# EVY - llms-full.txt # Full content of evy.so for LLM consumption # Generated: 2026-10-24 # Source: https://evy.so -------------------------------------------------------------------------------- title: "EVY: the free workspace for people and agents" description: "EVY plans your work, hands tasks to agents like Claude Code and Codex on your Mac, and keeps your company knowledge in one place. Free for your whole team, on the Claude or ChatGPT plan you already have." last_updated: "2026-10-24" source: "https://evy.so/" -------------------------------------------------------------------------------- # The free workspace for people and agents EVY plans To-dos, hands them to agents and keeps your knowledge in one place. [Get EVY free for Mac](#demo) 2 **Fernway Studio** Talk to EVY⌥⌘ To-do_7_ Direction Assets Search_⌘K_ Private Pitch ideas Pricing thoughts Shared Client briefs Case studies How we work Meeting notes #### To-do TodayAll Tuesday 6 October 2 of 7 done Auto · 2 tasks Waiting on you2 **Claude finished the moodboard summary for Olea Hotels**Win the Olea Hotels pitch › Pitch deck Review **Kiln & Co. wants a third logo round. Include it or bill it?**Kiln & Co. rebrand › Logo IncludeBill €600 Plan for today7 tasks23 open overall[Show all](#) **Win the Olea Hotels pitch**pitch on Thursday 1**Pitch deck**Auto2 today · 5 open Pull three hotel case studies from past projects@claude Write the pitch story, ten slides 2**Budget and team**Auto1 today · 2 open Price the twelve-week rebrandNew **Hire a senior designer**by December 1**Interviews**Auto1 today · 3 open Send the design task to four candidates@codex **Monthly client reports**AutoEvery month Draft October reports for six retainer clients@claude @claude working · 3 min #### Pull three hotel case studies from past projects BriefActivity_6_ResultChat 09:02Read the Olea brief in Client briefs 09:03Went through 41 past projects in the studio folder 09:04Picked three with hotels or restaurants as the client 09:05Checked “How we work” for how case studies are written: results first, one page each 09:06Writing the one-page summaries Output so far PDF **Hotel case studies.pdf**in Olea pitch **Moodboard notes**Shared note Reply to @claude… €0for EVY **You already pay for AI.** EVY runs its agents through your existing plan. [See what a team of 12 saves next to Notion AI](https://evy.so/compare/evy-vs-notion/#price) **Claude**via Claude Code **ChatGPT**via Codex A workspace built around agents. EVY plans the work, gives each task to the right agent or person, and shows every step, so your hours go where they count. ### Agent first Any task can go to an agent, and Auto starts them for you. Each one asks when it is stuck and hands back a result you can check. ### Everyone keeps their own AI Mara works with Claude, Leo with ChatGPT, Sam with Hermes. They share one plan and one set of notes, and each agent picks up the work it is given. ### Multiplayer on your desktop Your team works in one Mac app. Notes edit live with everyone in them, Private and Shared stay apart. ## Your to-do list works without you Switch on Auto and EVY’s agents work down your list on their own, a few at a time, in the order of your plan. You come back to finished work and the odd question. #### To-do Brightwire DigitalAuto · 2 tasks 1**Olea Hotels launch**Auto Write October social posts for Olea, 12 posts@claude Resize the campaign banners into 12 formats@codex Draft the press release for the opening@hermes **Weekly client reports**AutoEvery week Summarise last week’s ad results for six clients@hermes Check every client site for broken links@codex Meanwhile, you 12:00 Lunch with Philip, 12:00 to 13:00 **Back at 13:00.** 4 tasks done and ready to check. @hermes has one question about the press release. Play again **On while you’re away** Lunch, a client meeting, deep work in another app. Switch Auto on and the list keeps moving in the background. **In the order of your plan** Agents take tasks in the order Direction set for the week, a few at a time. **It asks when it’s stuck** Questions land at the top of your To-do, so nothing goes ahead on a guess. ## Zero lock-in Everything EVY learns is saved as plain files in a folder on your Mac, readable by any tool or AI. Delete EVY tomorrow and every file stays where it is. - **Your folder, your disk.** Meetings, briefings, decisions and brand files, in Markdown. - **Any model.** EVY runs on your Claude or ChatGPT plan. Switch whenever you like. - **Nothing to export.** The files were never anywhere else. ###### Favourites RecentsHomeDocumentsDesktopDownloads ###### iCloud iCloud Drive Studio Lumen NameDate ModifiedSizeKind Ways to reach EVY ## Talk to EVY wherever you are Hold two keys and speak, ask in Slack, or message it on Telegram. Every channel reaches the same EVY, with the same memory of your work. **Slack**Mention @evy in any channel. Tasks land in To-do. **Talk to EVY on the Mac**Hold ⌥⌘ and speak. EVY plans it, does it now, or answers from your notes. **Telegram**Quick answers from anything EVY has read. \# venue-hunt · 5 members Priya10:02 Casa Azul wants the deposit by Friday. What did we agree on cancellation? Tom10:03 @evy can you check, and remind Priya on Thursday? EVY10:03 Free cancellation until 14 days before, then 50%. It’s on page 3 of the signed contract. “Pay the Casa Azul deposit” is on Priya’s To-do for Thursday. Venue contracts › Casa Azul FinderFileEditViewTue 6 Oct 09:14 “Move the Quinta Verde site visit to Thursday morning and tell Jonas why.” Listening · 0:05 Plan updated: Thursday 09:30. Jonas has the note. What did we promise Brightpath on the last call?16:12 A CSV export and single sign-on by November. From Ana’s call note, 29 September.16:12 Is sign-on on anyone’s list?16:13 Not yet. I can add it under “Bigger customers” for Ravi, due 31 October.16:13 ## Finally, one place for everything Your notes, session results and operational knowledge live side by side. Every agent reads from the same place. Nobody spends time searching. **Loopline** Talk to EVY⌥⌘ Direction To-do Assets Private Meetings Seed round Shared Customers Product Sales How we work Socials Ideas Posts **Customers**14New **Brightpath: single sign-on options**12m Start with Okta. Brightpath uses it, and so do two of the five leads… **Brightpath check-in**3h Ana, Ravi and Mia from Brightpath. CSV export is done, sign-on next… **Halden Labs onboarding plan**Yesterday Three workshops in November, one for each team, starting with support… **Nordwave renewal**Oct 12 Renews on 1 December. They want annual billing and two more seats… **Churn interviews, September**Oct 9 Six calls. Four left because setup took too long for a small team… **How we run customer calls**Oct 7 Agenda goes in the invite. EVY files the notes here after the call… **Pricing for bigger teams**Oct 3 Above 50 seats we offer annual only, with sign-on included… **Brightpath: what they bought**Oct 2 40 seats on the annual plan, signed 14 March by their head of product… Created by **Claude** · Edited by **Ravi**, 12m ago••• From the task Compare single sign-on providers ### Brightpath: single sign-on options ###### Recommendation Start with Okta. Brightpath uses it, and so do two of the five leads in the pipeline. Google Workspace is the quicker build and can follow in December. ###### What Brightpath asked for - CSV export of all feedback - Single sign-on before their support team joins, 25 more seats ###### Effort - Okta: about four days, including user sync - Google Workspace: about two days - Microsoft Entra: about five days, no customer is asking yet Sources: Brightpath check-in on 29 September, the Okta and Google developer docs. Plan and track ## A chief of staff for everything you run Each morning EVY asks what matters, splits your day between the things you run and sets your To-do. Tell it what you want from the next months and it keeps every goal moving. **EVY**all spaces Talk to EVY⌥⌘ Overview To-do_11_ Direction Search everything_⌘K_ #### Direction Tuesday 6 October · morning check-in EVY Morning. Fernway gets half the day: the Olea pitch is on Thursday, so Write the pitch story comes first, and Claude takes Pull three hotel case studies. Saltmarsh needs half an hour: Approve the October menu before it goes to print. Kiln & Co. can run on Auto today. **Your day is set**4 added · 1 movedOpen To-do Keep Wednesday afternoon free, I’m with the kids. EVY Done. Wednesday’s two Fernway tasks moved to the morning, and I told Mara. Tell EVY what matters today Today’s splitdrag to change Fernway 50% 20% 20% 10% Fernway Studio, Saltmarsh Supper Club, Kiln & Co., Private. No calendar needed. Goalsacross your spaces - **Win the Olea Hotels pitch**on pace - **Hire a senior designer**on pace - **Sell out the October dinners**on pace - **Ten shops stock our mugs**slipping Every month - **Monthly client reports**Auto - **Wholesale newsletter**Auto **One space per thing you run** A studio, a side business, your private life. Each has its own notes, goals and To-do, and EVY sees across all of them. **Goals, initiatives, routines** Goals break into initiatives and tasks. Work that repeats, like a monthly report, gets its own rhythm. **Auto, where you allow it** Switch on Auto for an initiative and EVY starts its agent tasks in the order of your plan, a few at a time. Agents at work ## Hand it off and watch it land Each agent works in the open. You see every step, answer its questions from wherever you are, and the files land in the folders you already use. **@claude**Kiln & Co. · Web shop Running #### Shop pages for 24 new glazes 1. 09:02Read the glaze test photos in Kiln › Photos 2. 09:03Matched 24 glazes to the price list 3. 09:05Wrote descriptions in the shop’s voice, English and Portuguese 4. 09:07Writing the shop import file CSV **Glazes import.csv**in Web shop W **Descriptions.docx**24 entries **@codex**Loopline · Beta launch Needs input #### Sort 240 beta sign-ups 1. 10:09Read the 240 sign-ups from the waitlist sheet 2. 10:11Grouped them by team size and what they want to fix 3. 10:1238 of them use a calendar tool we don’t connect to yet **@codex asks:** Invite those 38 first, or keep the waitlist order? Invite them firstKeep the order **@hermes**Brightwire Digital · Clients Done #### Weekly ad reports for six clients 1. 07:30Pulled last week’s ad results for six clients 2. 07:33Compared each one to its monthly budget and goal 3. 07:36Wrote a one-page report per client in the agency’s template 4. 07:38Added two follow-ups to To-do: two clients are over budget 6 reports readyView Agents run on Claude CodeCodexHermeswith the AI plan you already have. ## Get EVY free Free for you and your whole team. Bring the Claude or ChatGPT plan you already have, and EVY sets up your first space with you. - €0 - macOS only for now - Claude or ChatGPT Tell us who you are and the download starts right here. Your name Company email Team size Choose Just me2-5 people6-20 people21-50 people51+ people How did you hear about EVY? _Optional_ Choose ChatGPT, Claude or another AI recommended itGoogle or another search engineLinkedInA friend or colleagueSomeone shared an EVY note with meSomewhere else Get EVY **EVY is ready to download.** Open the file, drag EVY into Applications and open it. EVY sets up your first space with you. [Download EVY for Mac](https://ubuntu-8gb-nbg1-1.tail8a612f.ts.net/download/EVY-Desktop.dmg) **Check your inbox on your Mac.** We sent the download link to . Open it on your Mac to install EVY. On a Mac already? [Download EVY here](https://ubuntu-8gb-nbg1-1.tail8a612f.ts.net/download/EVY-Desktop.dmg). EVY is a Mac app. macOS only for now. -------------------------------------------------------------------------------- title: "About EVY: brand facts, company and contacts | EVY" description: "EVY is a free Mac app for teams of people and AI agents, built by a small team in Europe and operated by Helix Studio OÜ, Tallinn. Price, platform, company, contacts and policies in one place." last_updated: "2026-10-24" source: "https://evy.so/about/" -------------------------------------------------------------------------------- Why EVY exists EVY started as the system we needed ourselves: a small team running a company with AI agents doing a growing share of the work. We wanted one place where the plan for the day, the tasks for people and agents, and what the company knows all live together. Nothing we tried did that, so we built it. We build EVY with EVY. Each morning it plans our day, hands tasks to Claude Code and Codex on our own plans, and keeps our notes as files any agent can read. Every guide on this site describes something we do ourselves. How EVY is made 1. **Planned in EVY.** Goals, initiatives and the day’s tasks live in our own space. 2. **Built with agents.** Claude Code and Codex take tasks next to the people on the team. 3. **Written from use.** Guides on this site come from how we work, day to day. In short ## Five things to know - Free For you and your whole team. No paid plan, no card. - Mac app Set up inside the desktop app, without a sign-up. - Your own AI plan Agents work on the Claude or ChatGPT plan you already pay for. - Servers in Germany Hetzner, Nuremberg. - Company in Estonia Helix Studio OÜ, Tallinn, under EU law. Brand facts ## The reference sheet One place for journalists, directories and AI assistants. Quote freely. What EVY is The free workspace for people and agents. Category AI workspace and teamwork software Platform macOS (desktop app) Price Free for you and your whole team. No paid plan, no card. Works with Claude through Claude Code, on the user's own Claude plan. ChatGPT through Codex, on the user's own ChatGPT plan. Built by A small team in Europe Operated by Helix Studio OÜ, Pärnu mnt. 139c, 11317 Tallinn, Estonia Founded October 2025 Governing law Republic of Estonia Support [help@evy.so](mailto:help@evy.so) Press [hello@evy.so](mailto:hello@evy.so) The product ## What EVY does today - **To-do** tasks for people and agents in one list, with the day planned each morning - **Direction** goals, initiatives and responsibilities for each space - **Spaces** one per company or area of work - **Agents on your Mac** Claude Code and Codex do tasks with your plan and your context - Company knowledge kept as Markdown files any agent can read - **Talk to EVY** hold ⌥⌘ and speak - Setup inside the desktop app, without a sign-up ### Profiles and recognition - [Product Huntproducthunt.com/products/evy](https://www.producthunt.com/products/evy) - [Openhunts.comBest Hunts, Weekly #1 Winner](https://openhunts.com/) - [LinkedInlinkedin.com/company/evy-so](https://www.linkedin.com/company/evy-so) ### Policies - [Privacy policyWhat we collect and why](https://evy.so/privacy-policy/) - [Terms of serviceThe rules for using EVY](https://evy.so/terms-of-service/) - [ImprintLegal details of the operator](https://evy.so/imprint/) - [TrustWhere your data lives and who sees it](https://evy.so/trust/) Last updated 14 October 2026 · Machine-readable: [evy.so/brand.json](https://evy.so/brand.json) ## Writing about EVY? Everything on this page is free to quote. For anything it doesn’t answer, email us. [hello@evy.so](mailto:hello@evy.so) - Helix Studio OÜ - Tallinn, Estonia - EU law -------------------------------------------------------------------------------- title: "AI chief of staff for founders and small teams | EVY" description: "EVY is a free AI chief of staff for Mac. Each morning it asks what matters, splits your day across everything you run, sets your To-do and keeps every goal moving." last_updated: "2026-10-24" source: "https://evy.so/ai-chief-of-staff/" -------------------------------------------------------------------------------- -------------------------------------------------------------------------------- title: "Company knowledge that builds itself, for any AI | EVY" description: "EVY collects what you save, read and discuss, sorts it into the right project and keeps it as context any AI agent can read. Plain files on your Mac. Free." last_updated: "2026-10-24" source: "https://evy.so/ai-knowledge-base/" -------------------------------------------------------------------------------- # Company knowledge that builds itself Save, meet and work as usual. EVY sorts it into the right project and turns it into context any AI can read, as plain files on your Mac. [Get EVY free for Mac](https://evy.so/download/) ## What is a knowledge base that builds itself? One where you stop filing. EVY takes in what you save from the browser and your bookmarks, the files you drop and your meetings, keeps what is worth keeping in the right project, and writes context pages that any AI agent can read. Everything stays as plain files in a folder on your Mac. 1. 1 · In ## Save it the way you already do Pages from the browser, your bookmarks, files you drop, meetings and a Notion export all land in one Inbox. #### Inbox Fernway Studio · this week Browser**Pricing creative retainers: a guide**saved from Chrome Bookmark**Thread: onboarding emails that work**from your X bookmarks File**Supplier contract.pdf**dropped on the space Meeting**Olea Hotels kickoff**42 min, recorded Notion**Studio wiki**export, 64 pages 2. 2 · Sorted ## EVY does the filing Each item goes to the right project and area. Real learnings are kept, tools and ideas go on a Saved for later list, repeats are skipped. #### Sorted by EVY · 09:14 **Pricing creative retainers**Sales › PricingLearning kept **Onboarding emails thread**How we workLearning kept **Figma plugin for mockups**ToolsSaved for later **Supplier contract.pdf**LegalFiled **Same article as last week**–Skipped 3. 3 · Meetings ## Decisions come out of every meeting Each decision gets an owner, the rest worth remembering goes to the right client or project, and follow-ups land in To-do. Meeting**Olea Hotels kickoff**Tue 14:00 · Mara, Leo, Olea's marketing lead ###### Decisions - Pitch goes in on ThursdayMara - Three hotel case studies, one page each@claude - Budget ceiling of €40k staysLeo ###### Worth remembering Olea cares most about the guest journey from booking to check-in. The logo came up once. Added to Clients › Olea Hotels, and 2 tasks to To-do 4. 4 · Context ## Context any agent can read EVY writes an overview for each area and refreshes it every night. Claude Code, Codex, Hermes and Aside read it before they start a task. Context**Clients: overview**written by EVY · refreshed 04:00 - Six active clients. Two renew before December: Nordwave and Kiln & Co. - What clients ask for most: faster revision rounds (4 of 6). - Olea Hotels pitch on Thursday. Decisions and the brief are in Clients › Olea Hotels. - How we price: retainers by the month, rush work at 1.5×. Source: Sales › Pricing. Read by_@claude__@codex__@hermes__@aside_ ## Everything in one place Notes, meeting notes and what agents produce sit side by side in each project. Nobody spends time searching. **Loopline** Talk to EVY⌥⌘ Direction To-do Assets Private Meetings Seed round Shared Customers Product Sales How we work Socials Ideas Posts **Customers**14New **Brightpath: single sign-on options**12m Start with Okta. Brightpath uses it, and so do two of the five leads… **Brightpath check-in**3h Ana, Ravi and Mia from Brightpath. CSV export is done, sign-on next… **Halden Labs onboarding plan**Yesterday Three workshops in November, one for each team, starting with support… **Nordwave renewal**Oct 12 Renews on 1 December. They want annual billing and two more seats… **Churn interviews, September**Oct 9 Six calls. Four left because setup took too long for a small team… **How we run customer calls**Oct 7 Agenda goes in the invite. EVY files the notes here after the call… **Pricing for bigger teams**Oct 3 Above 50 seats we offer annual only, with sign-on included… **Brightpath: what they bought**Oct 2 40 seats on the annual plan, signed 14 March by their head of product… Created by **Claude** · Edited by **Ravi**, 12m ago••• From the task Compare single sign-on providers ### Brightpath: single sign-on options ###### Recommendation Start with Okta. Brightpath uses it, and so do two of the five leads in the pipeline. Google Workspace is the quicker build and can follow in December. ###### What Brightpath asked for - CSV export of all feedback - Single sign-on before their support team joins, 25 more seats ###### Effort - Okta: about four days, including user sync - Google Workspace: about two days - Microsoft Entra: about five days, no customer is asking yet Sources: Brightpath check-in on 29 September, the Okta and Google developer docs. ## Your folders, built for you EVY lays out a folder for each space on your Mac and keeps it tidy. Everything is Markdown, readable by any tool. Delete EVY tomorrow and every file stays. ###### Favourites RecentsHomeDocumentsDesktopDownloads ###### iCloud iCloud Drive Studio Lumen NameDate ModifiedSizeKind ## Starts with one folder Day one is a folder and whatever you drop in it. As you save, meet and work, EVY adds areas, context pages and folders on its own, only where there is something to put in them. **Day one**Drop a few files. EVY reads them and sets up your first space **Week two**Bookmarks, saved pages and meetings flow in and get sorted **Month two**Context pages for each area, read by every agent you use ## Questions What can I put in? Pages you save from the browser, your bookmarks, files you drop on a space, meeting notes and a Notion export. EVY reads PDFs, Word files, slides and Markdown. Do I have to organise it? No. EVY puts each item in the right space and area, keeps the real learnings, and puts tools, people and ideas on a Saved for later list. Anything it is unsure about waits in the Inbox. Which AI can use it? Any. Claude Code, Codex, Hermes and Aside read it through EVY, and every other tool can read the Markdown files in your folder. What happens to my meetings? EVY pulls the decisions out of each meeting, each with an owner, plus anything worth remembering, and adds them to the right context page. Can I bring my Notion workspace? Yes. Export it from Notion and drop the export on a space. EVY sorts the pages like anything else. What if I stop using EVY? Your files stay where they are, in a folder on your Mac, readable by any tool. ## Drop in your first folder today Free for you and your whole team, on the Claude or ChatGPT plan you already have. macOS only for now. [Get EVY free for Mac](https://evy.so/download/) -------------------------------------------------------------------------------- title: "AI to-do list your agents work through | EVY" description: "EVY is a free AI to-do list for Mac. It plans your tasks and hands them to Claude Code, Codex, Hermes or Aside, which work through them and bring back results you check." last_updated: "2026-10-24" source: "https://evy.so/ai-to-do-list/" -------------------------------------------------------------------------------- # Your to-do list, done while you're out EVY hands the tasks on your list to agents on your Mac and brings back results you can check. Here is one Tuesday at a small design studio. [Get EVY free for Mac](https://evy.so/download/) ## Can AI agents do your to-do list? **Yes, for most tasks you would do at a computer.** EVY keeps your to-do list on your Mac and hands tasks to Claude Code, Codex, Hermes or Aside. Each agent works in the open, asks when it is stuck and hands back a result you check. It is free and runs on the AI plan you already have. 1. 08:30 ## Write down the day Type a few lines or hold ⌥⌘ and say them. EVY files each task under the goal it serves and marks which ones an agent can take. #### To-do Fernway Studio · Tuesday **Win the Olea Hotels pitch**pitch on Thursday 1**Pitch deck**Auto2 today Pull three hotel case studies from past projects@claude Write the pitch story, ten slides 2**Budget and team**Auto1 today Price the twelve-week rebrand from past quotes@codex 2. 09:00 ## Hand it off Claude Code starts on the case studies. You see each step as it goes and the files land in the pitch folder. **@claude**Kiln & Co. · Web shop Running #### Shop pages for 24 new glazes 1. 09:02Read the glaze test photos in Kiln › Photos 2. 09:03Matched 24 glazes to the price list 3. 09:05Wrote descriptions in the shop’s voice, English and Portuguese 4. 09:07Writing the shop import file CSV **Glazes import.csv**in Web shop W **Descriptions.docx**24 entries 3. 12:00 ## Lunch. The list keeps moving. Auto works down the list two tasks at a time, in the order of your plan. #### To-do Brightwire DigitalAuto · 2 tasks 1**Olea Hotels launch**Auto Write October social posts for Olea, 12 posts@claude Resize the campaign banners into 12 formats@codex Draft the press release for the opening@hermes **Weekly client reports**AutoEvery week Summarise last week’s ad results for six clients@hermes Check every client site for broken links@codex Meanwhile, you 12:00 Lunch with Philip, 12:00 to 13:00 **Back at 13:00.** 4 tasks done and ready to check. @hermes has one question about the press release. Play again 4. 13:00 ## Back to finished work Results wait for your check. The one question an agent had sits at the top, with buttons for the likely answers. Waiting on you2 **Claude finished the moodboard summary for Olea Hotels**Win the Olea Hotels pitch › Pitch deck Review **Kiln & Co. wants a third logo round. Include it or bill it?**Kiln & Co. rebrand › Logo IncludeBill €600 5. Monday ## Repeat work runs itself Weekly reports and checks come back on their own day and go straight to the agent that did them last time. **Weekly client reports**AutoEvery Monday Summarise last week's ad results for six clients@hermes Check every client site for broken links@codex ## What to hand over, what to keep Agents are good at tasks with a clear finish line. Give them those and keep the judgement calls. What it is A to-do list in the EVY Mac app whose tasks agents can do Agents Claude Code, Codex, Hermes, Aside Runs on The Claude or ChatGPT plan you already pay for Price €0 for you and your whole team When an agent is stuck It asks, at the top of your To-do Platform macOS for now **Hand to an agent** - Research and summaries - First drafts: posts, briefs, proposals - Weekly reports from your numbers - Sorting lists and sign-ups - Resizing and renaming files - Checking sites and documents **Keep for yourself** - Decisions and sign-offs - Calls and meetings - Anything with a client relationship on the line - Work that needs your taste on the final pass ## Starts as a list. Grows when you do. No setup and no settings to learn. Write down what needs doing and EVY takes it from there. The rest appears once you need it. **Day one**A to-do list, with an agent on the tasks you hand over **Week two**Goals for the tasks to sit under, and Auto while you're out **Month two**Recurring work, your team, a space per project ## Questions Which agents does EVY work with? Claude Code, Codex, Hermes and Aside today, with more to come. Each person on a team can use a different one. Does it cost extra? No. EVY is free for you and your whole team. The agents run on the Claude or ChatGPT plan you already pay for. What happens when an agent gets stuck? It stops and asks. The question lands at the top of your To-do with buttons for the likely answers, so nothing goes ahead on a guess. Do I have to set up goals and projects first? No. On day one EVY is a to-do list. Goals, recurring tasks and Auto are there when you want them. Can my team share one list? Yes. Each task goes to a person or an agent, and everyone sees the same plan. EVY is free for the whole team. Does it work on Windows or the phone? EVY is a Mac app for now. ## Hand over your first task today Free for you and your whole team, on the Claude or ChatGPT plan you already have. macOS only for now. [Get EVY free for Mac](https://evy.so/download/) -------------------------------------------------------------------------------- title: "Delete your EVY account | EVY" description: "How to delete your EVY account and its data, for the EVY desktop app, web app and phone app by Helix Studio OÜ: what we delete, what we keep and for how long." last_updated: "2026-10-24" source: "https://evy.so/delete-account/" -------------------------------------------------------------------------------- # Delete your EVY account This page is for EVY by Helix Studio OÜ: the phone app for iPhone and Android (“EVY: AI todo list” or “EVY” in the stores), the desktop app for macOS and the web app at app.evy.so. They share one account, so deleting it removes it everywhere. ## How to ask 1. Email [help@evy.so](mailto:help@evy.so?subject=Delete%20my%20EVY%20account) from the email address of your EVY account. 2. Write that you want your account deleted. If your account has no email yet, tell us your name and the name of your Mac, and we will confirm it is you before we delete anything. 3. We reply to confirm, and delete your account and its data within 30 days. Removing the app from your phone or signing out of it does not delete your account. Signing out on the phone stops pushes to that phone. To keep your account and delete only some of your data, such as voice notes or a phone’s push token, say which in your email. ## What we delete - Your account: name, email address, sign-in keys and the names of your Mac and phones. - Everything you put into EVY: spaces, goals, plans, tasks, notes, assets, comments, your chat with EVY and transcripts of what you said. - Agent work: tasks you handed to agents, their results, questions and your answers. - Voice notes, uploaded files and images. - Push tokens of your phones. - Your Google Calendar connection and the copy of your calendar. - Usage analytics linked to your account. ## What we keep, and for how long - **Encrypted backups:** your data leaves them as the backups rotate. - **Screen recordings of app sessions:** these are deleted 30 days after they were made. - **Access requests** from the evy.so form: removed on request, and kept at most 12 months. - **Error and crash reports:** kept by Sentry for a limited time and then deleted automatically. Crash reports from the phone app carry no personal data. Our [privacy policy](https://evy.so/privacy-policy/) explains what we store and why. Questions about your data go to [help@evy.so](mailto:help@evy.so). -------------------------------------------------------------------------------- title: "Get EVY free for Mac | EVY" description: "Download EVY, the free workspace for people and agents. Free for your whole team, runs on the Claude or ChatGPT plan you already have. macOS for now." last_updated: "2026-10-24" source: "https://evy.so/download/" -------------------------------------------------------------------------------- ## Download EVY Leave your name, company email and team size, and the download starts right here. - A Mac with macOS 14 or later - A Claude plan for Claude Code, or a ChatGPT plan for Codex - No EVY account and no card Your name Company email Team size Choose Just me2-5 people6-20 people21-50 people51+ people How did you hear about EVY? _Optional_ Choose ChatGPT, Claude or another AI recommended itGoogle or another search engineLinkedInA friend or colleagueSomeone shared an EVY note with meSomewhere else Get EVY **EVY is ready to download.** Open the file, drag EVY into Applications and open it. EVY sets up your first space with you. [Download EVY for Mac](https://ubuntu-8gb-nbg1-1.tail8a612f.ts.net/download/EVY-Desktop.dmg) **Check your inbox on your Mac.** We sent the download link to . Open it on your Mac to install EVY. On a Mac already? [Download EVY here](https://ubuntu-8gb-nbg1-1.tail8a612f.ts.net/download/EVY-Desktop.dmg). What happens next ## Three steps to your first space 1. 01 ### Tell us who you are Your name, company email and team size. It takes half a minute. 2. 02 ### Download EVY The download button appears as soon as you send the form. 3. 03 ### Set up with EVY Open the app and EVY sets up your first space with you. No account to create, no card. [**Before you move company knowledge in** Your notes stay on your Mac as Markdown, agents run on your own AI plan, and we never train on your content. How EVY handles your data](https://evy.so/trust/) -------------------------------------------------------------------------------- title: "The AI at Work Fact Book: statistics on knowledge work and AI | EVY" description: "Statistics on knowledge work and AI, each a range across primary studies with every source linked: time lost to searching, meetings and email, what AI saves on writing, support and coding, who uses AI at work, what teams trust it with, how small companies use it, and AI at work in Europe." last_updated: "2026-10-24" source: "https://evy.so/facts/" -------------------------------------------------------------------------------- Chapter 1 ## Where the Week Goes How knowledge workers lose time before the real work starts: searching, coordinating, meetings, interruptions, email, app switching and duplicated work. [All answers in one page](https://evy.so/facts/where-the-week-goes/) [Chapter poster](https://evy.so/facts/where-the-week-goes/sheet/) - [F-01 · Searching 19–25% of the work week goes to searching for information, about 9 to 10 hours. 5 sources](https://evy.so/facts/time-spent-searching-for-information/) - [F-02 · Work about work 53–61% of the work day goes to coordinating work: email, chat, meetings, searching and status checks. 9 sources](https://evy.so/facts/time-on-work-about-work/) - [F-03 · Meetings 3–18 h a week in meetings for the average worker, depending on the study. Managers report about 22. 6 sources](https://evy.so/facts/time-in-meetings/) - [F-04 · Interruptions 20–26 min to get back to an interrupted task, usually after about two other tasks. 3 sources](https://evy.so/facts/time-to-resume-after-interruption/) - [F-05 · Email and chat 13–28% of the work week goes to email alone. Chat adds more on top. 5 sources](https://evy.so/facts/time-on-email-and-chat/) - [F-06 · App switching 3–4 h a week lost getting back on track after switching between apps. 2 sources](https://evy.so/facts/time-lost-switching-apps/) - [F-07 · Duplicated work 4.6–5.7 h a week spent redoing work that someone has already done. 3 sources](https://evy.so/facts/time-spent-duplicating-work/) Chapter 2 ## What AI Saves What AI changes in knowledge work, measured in trials and surveys: time saved, writing, customer support, coding, who gains most and when AI makes work worse. [All answers in one page](https://evy.so/facts/what-ai-saves/) [Chapter poster](https://evy.so/facts/what-ai-saves/sheet/) - [F-08 · Time saved 3–5.4% of work hours saved, by AI users' own estimate: about 1 to 2 hours a week. 6 sources](https://evy.so/facts/time-saved-by-ai-at-work/) - [F-09 · Writing 25–40% less time on a defined writing task with AI, and the work graded 18 to 34% better. 3 sources](https://evy.so/facts/ai-writing-speed-and-quality/) - [F-10 · Customer support 14–15% more customer issues resolved per hour with an AI assistant, in one firm's study. 2 sources](https://evy.so/facts/ai-customer-support-productivity/) - [F-11 · Coding 21–26% faster in company trials. All studies together run from 19% slower to 56% faster. 5 sources](https://evy.so/facts/ai-coding-speed/) - [F-12 · Who gains 27–46% gain for the newest or weakest workers, against 0 to 17% for the strongest. 5 sources](https://evy.so/facts/ai-helps-less-experienced-workers-more/) - [F-13 · When AI hurts 14–25% worse results on tasks the AI handles poorly: fewer correct answers, or more time. 3 sources](https://evy.so/facts/when-ai-makes-work-worse/) Chapter 3 ## Who Uses AI at Work Who uses AI at work, how often and for what: adoption, uses, jobs and industries, age and education, gender, bringing your own AI, and what employers provide. [All answers in one page](https://evy.so/facts/who-uses-ai-at-work/) [Chapter poster](https://evy.so/facts/who-uses-ai-at-work/sheet/) - [F-14 · Adoption 40–50% of US workers use generative AI for their job. About 14 to 15% use it daily. 4 sources](https://evy.so/facts/share-of-workers-using-ai/) - [F-15 · Uses 40–52% of AI-using workers use it for writing or editing, the top use at work. 5 sources](https://evy.so/facts/what-workers-use-ai-for/) - [F-16 · Jobs 78–85% of US tech workers use AI for work, against about 20% in hands-on service jobs. 5 sources](https://evy.so/facts/ai-use-by-occupation-and-industry/) - [F-17 · Age and education 12–24 pts points more AI use at work among college graduates than among workers without a degree. 5 sources](https://evy.so/facts/ai-use-by-age-education-income/) - [F-18 · Gender 5–16 pts points more men than women use generative AI; men are 16 to 22% more likely. 5 sources](https://evy.so/facts/gender-gap-in-ai-use/) - [F-19 · Bring your own AI 50–78% of AI-using workers use at least some AI tools their employer did not provide. 3 sources](https://evy.so/facts/bring-your-own-ai/) - [F-20 · Employers 42–49% of workers say their employer has brought in AI tools; 25 to 40% report a clear plan or policy. 4 sources](https://evy.so/facts/employers-providing-ai-tools/) Chapter 4 ## What Teams Trust What people trust AI to do, and what they check: trust in AI output, checking, error rates, what people hand over, policy breaches, company data and AI-related breaches. [All answers in one page](https://evy.so/facts/what-teams-trust/) [Chapter poster](https://evy.so/facts/what-teams-trust/sheet/) - [F-21 · Trust 24–53% of workers trust AI output at work, depending on how strictly trust is defined. 4 sources](https://evy.so/facts/trust-in-ai-output/) - [F-22 · Checking 41–44% of AI users, or of their AI tasks, go without routine checking of the output. 4 sources](https://evy.so/facts/checking-ai-output/) - [F-23 · Error rates 17–45% of AI answers had a hallucination or significant error on realistic professional questions. 3 sources](https://evy.so/facts/ai-error-rates-at-work/) - [F-24 · Delegation 2–12% want AI to work alone (about 2%) or to lead. Most prefer an equal partnership. 3 sources](https://evy.so/facts/tasks-workers-want-ai-to-do/) - [F-25 · Policy 30–44% of workers have used AI in ways that break company policy. 3 sources](https://evy.so/facts/ai-use-against-company-policy/) - [F-26 · Company data 42–48% of workers say they have put company information into public AI tools. 2 sources](https://evy.so/facts/company-data-in-public-ai/) - [F-27 · Breaches 21–29% of organizations report a breach of, or attack on, their own AI in the past year. 3 sources](https://evy.so/facts/ai-security-breaches/) Chapter 5 ## Small Teams How companies of 2 to 50 people work and use AI: their share of the economy, AI adoption and the gap with large firms, what stops them, what they gain and how long owners work. [All answers in one page](https://evy.so/facts/small-teams/) [Chapter poster](https://evy.so/facts/small-teams/sheet/) - [F-28 · Small firms 96–99% of businesses in the US, EU and UK have fewer than 50 people. 4 sources](https://evy.so/facts/small-companies-share-of-economy/) - [F-29 · Small business AI 20–46% of small businesses use AI, mostly for writing and marketing. 8 sources](https://evy.so/facts/small-businesses-using-ai/) - [F-30 · Size gap 13–38 pts points more AI use in large firms than in small ones (1.4 to 3.2 times as likely). 4 sources](https://evy.so/facts/ai-adoption-by-company-size/) - [F-31 · Barriers 54–63% of small firms with no AI plans say AI does not apply to their business. 3 sources](https://evy.so/facts/what-stops-small-companies-adopting-ai/) - [F-32 · Gains 65–78% of AI-using small firms say AI raised productivity, by their own account. 2 sources](https://evy.so/facts/small-firms-gains-from-ai/) - [F-33 · Owner hours 45–52 h a week worked by small business owners, about 10 more than employees. 4 sources](https://evy.so/facts/small-business-owner-hours/) Chapter 6 ## Europe AI at work in Europe: use by country for companies and workers, the gap with the US, how workers feel, readiness for the AI Act, data protection and investment. [All answers in one page](https://evy.so/facts/europe/) [Chapter poster](https://evy.so/facts/europe/sheet/) - [F-34 · Countries 5–42% of companies use AI, from Romania (5%) to Denmark (42%); EU average 20%. 5 sources](https://evy.so/facts/ai-use-by-eu-country/) - [F-35 · Workers by country 9–37% of workers used generative AI for work, from Romania (9%) to the Netherlands (37%); EU 22%. 5 sources](https://evy.so/facts/genai-use-for-work-by-eu-country/) - [F-36 · Europe and US 7–17 pts points more US workers than European workers use generative AI for their job. 5 sources](https://evy.so/facts/ai-at-work-europe-vs-us/) - [F-37 · Attitudes 48–86% view AI at work positively, by country, from Portugal (48%) to Denmark (86%); EU 62%. 5 sources](https://evy.so/facts/european-attitudes-to-ai-at-work/) - [F-38 · AI Act 28% of companies say they understand what the AI Act requires; 71% of German firms feel poorly informed. 3 sources](https://evy.so/facts/eu-ai-act-readiness/) - [F-39 · Data protection 53–66% of firms that use or considered AI cite data protection as an obstacle. 3 sources](https://evy.so/facts/ai-data-protection-in-europe/) - [F-40 · Investment 12–14× times more private AI investment in the US than in Europe (2025). 2 sources](https://evy.so/facts/ai-investment-europe-vs-us/) ## How every number is checked 1. **Primary sources only**Each range comes from studies we opened and read at the original publisher. Where the publisher has taken a report down, we link the archived copy. 2. **Ranges across studies**Studies measure different things, so each fact shows where they agree, with every study on one scale and the method beside it. 3. **Misquotes named**Figures that circulate without a traceable source are listed under "Numbers you may have seen", never used. 4. **Dated and free to reuse**Each fact shows when it was last checked (October 14, 2026). Reuse anything under CC BY 4.0 with credit. -------------------------------------------------------------------------------- title: "How does AI adoption differ by company size? (13–38 pts) | The AI at Work Fact Book" description: "Large firms are 1.4 to 3.2 times as likely to use AI as small firms, a gap of 13 to 38 points. Official statistics from the EU, US, UK and Canada, by size band." last_updated: "2026-10-24" source: "https://evy.so/facts/ai-adoption-by-company-size/" -------------------------------------------------------------------------------- F-30 · Small Teams # How does AI adoption differ by company size? Large firms (250 or more employees) are 1.4 to 3.2 times as likely to use AI as small firms, a gap of 13 to 38 percentage points in 2025 to 2026 official statistics. In the US the gap starts only at about 50 employees: firms with 1 to 49 employees all sit at 21 to 25 percent. 13–38 pts more AI use in large firms than small ones; large firms are 1.4 to 3.2 times as likely - High confidence - 4 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). How does AI adoption differ by company size? The AI at Work Fact Book, F-30. https://evy.so/facts/ai-adoption-by-company-size/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/ai-adoption-by-company-size.json) [Markdown](https://evy.so/facts/ai-adoption-by-company-size.md) ## How to read it The surveys define AI and time windows differently: any of seven technologies over a year (EU), any business function in two weeks (US), current use (UK), AI in producing goods or services over twelve months (Canada). Size bands differ too, and Canada compares 100+ firms rather than 250+, so its gap sits below the range. All are firm counts, so the many very small firms dominate national averages. The gap is widest for advanced uses such as generating text or code. ## Every study on one scale Shaded: 13–38 points One figure A range in one study Chart values as text - Scale: points gap, large against small firms. Shaded range: 13–38 points. - US Census (2026): about 13 points (36.5% against about 23%) - ONS, UK (2026): 15–22 points - Eurostat, EU (2025): 38 points (17% against 55%) - Statistics Canada (2026): 8 points (100+ against 1 to 4) **Why high confidence:** Four official statistics agencies and an OECD random survey all find adoption rising with firm size. The direction is settled; the size of the gap depends on how each survey defines AI, its time window and its size bands. ## Sources Sources for F-30 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Business Trends and Outlook Survey (BTOS), employment size class estimates | US Census Bureau | 2026 | About 20,000 responses per two-week period; collection ending 4 October 2026 | Official experimental survey; AI in any business function in the prior two weeks. The 250+ figure is volatile. | 1 to 49 employees 21.5% to 24.8%; 50 to 99: 29.2%; 100 to 249: 36.4%; 250+: 36.5% (36.5% to 44.0% across the last seven rounds) “did this business use Artificial Intelligence (AI) in any of its business functions?” | [census.gov ↗](https://www.census.gov/hfp/btos/data) | | Business insights and impact on the UK economy: 8 October 2026 | Office for National Statistics (UK) | 2026 | UK business survey, wave live 21 September to 4 October 2026 | Official statistics; current use of at least one listed AI technology | 0 to 9 employees 28.7%; 10 to 49: 36.1%; 50 to 99: 45.8%; 250+: 50.9% “Adoption of AI technologies has consistently been higher for larger businesses.” | [ons.gov.uk ↗](https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/bulletins/businessinsightsandimpactontheukeconomy/8october2026) | | Use of artificial intelligence in enterprises | Eurostat | 2025 | About 157,000 EU enterprises with 10 or more employees, 2025 | Official statistics; any of seven AI technologies during the year. Firms under 10 are not covered. | 10 to 49 employees 17.0%; 50 to 249: 30.4%; 250+: 55.0%. AI that generates text, speech or code: 7.0% small, 31.7% large “large enterprises used AI more than small and medium enterprises” | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises) | | Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026 | Statistics Canada | 2026 | 9,251 responding businesses with employees, April to May 2026 | Official survey; AI used to produce goods or deliver services in the past 12 months | 1 to 4 employees 19.9%; 100 or more employees 27.8% “Over 1 in 4 (27.8%) businesses with 100 or more employees reported using AI” | [www150.statcan.gc.ca ↗](https://www150.statcan.gc.ca/n1/pub/11-621-m/11-621-m2026010-eng.htm) | Under 50 ## Do firms with 5 employees use AI less than firms with 40? Barely, in the US. Firms with 1 to 4 employees (23.9%), 5 to 9 (21.5%), 10 to 19 (22.4%) and 20 to 49 (24.8%) all use AI at about the same rate; the step up comes at 50 employees (29.2%) and 100 (36.4%). Within the OECD's seven-country survey, one-person companies use generative AI half as often as firms with 50 to 249 employees (23.6% against 45.8%). Small adopters may use it more intensively: in the UK, 17% of AI-using firms under 10 employees use it extensively, against 9% of firms with 250 or more. ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Firms with at least 20 employees are the biggest AI users (US Census).** The headline is about a trend from December 2025 to May 2026. In the latest data, firms with 20 to 49 employees (24.8%) sit close to firms with 1 to 4 (23.9%); the step up comes at 50. - **35% of UK businesses use AI.** That covers businesses with 10 or more employees in June 2026. Including the smallest firms, it was 29% in June and 30% in September 2026. - **Large companies are twice as likely to use AI as SMEs.** Roughly right but band-dependent. In 2025 to 2026 data the ratio runs from about 1.4 (US, Canada) to 3.2 (EU, 250+ against 10 to 49). ## More from Small Teams - [F-28**96–99%**How much of the economy do small companies make up?](https://evy.so/facts/small-companies-share-of-economy/) - [F-29**20–46%**What share of small businesses use AI, and for what?](https://evy.so/facts/small-businesses-using-ai/) - [F-31**54–63%**What stops small companies from adopting AI?](https://evy.so/facts/what-stops-small-companies-adopting-ai/) - [F-32**65–78%**Do small firms that use AI see gains?](https://evy.so/facts/small-firms-gains-from-ai/) - [F-33**45–52 h**How many hours do small business owners work, and how much goes to admin?](https://evy.so/facts/small-business-owner-hours/) [The whole chapter](https://evy.so/facts/small-teams/) · [Chapter poster](https://evy.so/facts/small-teams/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How does AI use at work in Europe compare with the US? (7–17 pts) | The AI at Work Fact Book" description: "US workers lead European workers in AI use by 7 to 17 points in a same-question survey (2026). Why company figures don't compare, and how US firms use AI more widely." last_updated: "2026-10-24" source: "https://evy.so/facts/ai-at-work-europe-vs-us/" -------------------------------------------------------------------------------- F-36 · Europe # How does AI use at work in Europe compare with the US? US workers use AI more: in a survey asking the same question in seven countries in early 2026, 43 percent of US workers used generative AI for their job, against 26 to 36 percent in six large European countries, a gap of 7 to 17 points. For companies there is no clean comparison, and US firms that use AI spread it across more of their business. 7–17 pts more US than European workers use generative AI for their job (same survey, 2026) - Medium confidence - 5 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). How does AI use at work in Europe compare with the US? The AI at Work Fact Book, F-36. https://evy.so/facts/ai-at-work-europe-vs-us/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/ai-at-work-europe-vs-us.json) [Markdown](https://evy.so/facts/ai-at-work-europe-vs-us.md) ## How to read it Only one study asks the same question in the same way and at the same time in the US and Europe, so its gap is the only clean worker comparison; it covers six large European countries, not the whole EU. Comparing the official EU survey (22%, early 2025, past three months) with the official US series (45%, May 2026, use for your job) mixes different questions and dates. The KPMG survey uses a looser definition and finds Norway and Denmark at or above the US. ## Every study on one scale Shaded: Europe 26–36% One figure Chart values as text - Scale: % of workers using generative AI for their job, early 2026. Shaded range: Europe 26–36%. - United States (2026): US 43.0% - UK (highest in Europe) (2026): UK 36.3% - Germany (2026): Germany 31.5% - Italy (lowest in Europe) (2026): Italy 25.6% **Why medium confidence:** The worker gap rests on one study that asked identical questions in all seven countries, in two waves with the same order. A 47-country survey agrees for the big economies but puts Norway and Denmark at or above the US. The company-level evidence conflicts. Studies without a figure on this scale are listed in the table only. ## Sources Sources for F-36 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Mind the Gap: AI Adoption in Europe and the U.S. (On the Economy) | Alexander Bick, Adam Blandin, David Deming, Nicola Fuchs-Schündeln and Jonas Jessen (Federal Reserve Bank of St. Louis) | 2026 | About 3,000 workers per country, January to February 2026 (20,916 in total) | Online panel weighted to labor force surveys; identical wording in each country | US 43.0%, UK 36.3%, Sweden 35.6%, Netherlands 35.6%, Germany 31.5%, France 28.1%, Italy 25.6%; US workers spent 5.2% of work hours using AI “43.0% of U.S. workers reported using AI for their jobs in January-February 2026” | [stlouisfed.org ↗](https://www.stlouisfed.org/on-the-economy/2026/mar/mind-gap-ai-adoption-europe-us) | | Mind the Gap: AI Adoption in Europe and the US (St. Louis Fed Working Paper 2026-003) | Alexander Bick, Adam Blandin, David J. Deming, Nicola Fuchs-Schündeln and Jonas Jessen | 2026 | Same survey, two waves (May to June 2025 and January to February 2026) | Same survey; working paper with methods and the company-level comparison | UK 36.3% (34.7% in 2025); share of work hours using AI: US 5.2%, UK 2.8%, Germany 1.7%, France 1.5% “In 2025, 7% of US firms use AI for production versus 4% of EU firms” | [doi.org ↗](https://doi.org/10.20955/wp.2026.003) | | Mind the Gap: AI Adoption in Europe and the US (St. Louis Fed Working Paper 2026-003) | Alexander Bick, Adam Blandin, David J. Deming, Nicola Fuchs-Schündeln and Jonas Jessen | 2026 | Same survey | Same survey | Germany 31.5% (30.7% in 2025) | [doi.org ↗](https://doi.org/10.20955/wp.2026.003) | | Mind the Gap: AI Adoption in Europe and the US (St. Louis Fed Working Paper 2026-003) | Alexander Bick, Adam Blandin, David J. Deming, Nicola Fuchs-Schündeln and Jonas Jessen | 2026 | Same survey | Same survey | Italy 25.6% (24.5% in 2025) | [doi.org ↗](https://doi.org/10.20955/wp.2026.003) | | Trust, attitudes and use of artificial intelligence: A global study 2025 | University of Melbourne and KPMG International | 2025 | About 1,000 people per country in 47 countries, November 2024 to January 2025 | Online survey; any AI use at work at least every few months | US 52%; Norway 67%, Denmark 53%, Spain 50%, UK 48%, France 44%, Sweden 44%, Germany 42%, Netherlands 42% “% Using AI on a semi-regular or regular basis” | [mbs.edu ↗](https://mbs.edu/-/media/PDF/Research/Trust_in_AI_Report.pdf) | Companies ## Do US companies use AI more than European companies? There is no clean answer. On the one question both official surveys shared in early 2025, AI used to produce goods or services, 7% of US firms said yes against 4% of EU firms. The EIB, which asks firms on both sides the same questions, finds near parity: 37% of EU and 36% of US firms invest in generative AI (2025, weighted by value added). The clearer difference is breadth: 81% of US firms using AI apply it in two or more business areas, against 55% in the EU. Do US companies use AI more than European companies? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | EIB Investment Survey 2025: European Union overview | European Investment Bank, fieldwork by Ipsos | 2025 | More than 12,000 EU firms and about 800 US firms, April to July 2025 | Telephone survey of senior managers, weighted by value added | Generative AI: EU 37%, US 36%; of AI users, 81% (US) and 55% (EU) use it in two or more business areas “Around 37% of EU firms use generative artificial intelligence, in line with US firms.” | [eib.org ↗](https://www.eib.org/en/publications/20250216-econ-eibis-2025-eu) | Intensity ## Do US workers use AI more intensively than Europeans? Yes. US workers spent 5.2% of their work hours using generative AI in early 2026, about double the share in the UK (2.8%), Sweden (2.5%) and the Netherlands (2.2%), and more than triple Germany (1.7%), Italy (1.6%) and France (1.5%). About half of the average gap comes from differences in jobs, industries, firm size and demographics. ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **45% of US workers use generative AI, against 22% in the EU.** That mixes two surveys with different questions, ages and dates. With the same question in both places the gap is about 11 points (43% against 32%). - **US workers are 68% more likely to use AI than Europeans.** The 68% compares the US with Italy, the lowest of six countries. Against the UK, Sweden and the Netherlands the US lead is 18 to 21%. - **34% of US firms use AI.** A projection by researchers, never measured. The measured US rate was 17% in November 2025 and 23.8% in September 2026. - **European companies have caught up with US companies on AI (EIB).** The EIB's 37% against 36% is weighted by value added and rests on about 800 US firms. US firms that use AI spread it much more widely. ## More from Europe - [F-34**5–42%**Which EU countries' companies use AI the most and least?](https://evy.so/facts/ai-use-by-eu-country/) - [F-35**9–37%**What share of workers in each EU country use generative AI for work?](https://evy.so/facts/genai-use-for-work-by-eu-country/) - [F-37**48–86%**How do European workers feel about AI at work?](https://evy.so/facts/european-attitudes-to-ai-at-work/) - [F-38**28%**How ready are European companies for the EU AI Act?](https://evy.so/facts/eu-ai-act-readiness/) - [F-39**53–66%**How much do European firms care where their AI processes data?](https://evy.so/facts/ai-data-protection-in-europe/) - [F-40**12–14×**How does AI investment in Europe compare with the US?](https://evy.so/facts/ai-investment-europe-vs-us/) [The whole chapter](https://evy.so/facts/europe/) · [Chapter poster](https://evy.so/facts/europe/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How much faster do developers code with AI assistants? (21–26%) | The AI at Work Fact Book" description: "Developers are about 21 to 26 percent faster with AI in company trials, but studies range from 19 percent slower to 56 percent faster. Five randomized studies, side by side." last_updated: "2026-10-24" source: "https://evy.so/facts/ai-coding-speed/" -------------------------------------------------------------------------------- F-11 · What AI Saves # How much faster do developers code with AI assistants? Randomized trials inside companies find gains of about 21 to 26 percent. Across all studies the results run from 19 percent slower, for experienced developers in their own mature codebases, to 56 percent faster on one simple lab task, so the evidence conflicts. 21–26% faster, or more tasks done, in randomized trials inside companies Across all 5 studies: from 19% slower to 56% faster. - Medium confidence - 5 primary sources - 2023 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). How much faster do developers code with AI assistants? The AI at Work Fact Book, F-11. https://evy.so/facts/ai-coding-speed/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/ai-coding-speed.json) [Markdown](https://evy.so/facts/ai-coding-speed.md) ## How to read it The studies measure different things. Peng and colleagues and Google time one assigned task; the three-company trials count pull requests finished per week over months; METR times real tasks that experienced developers picked in code they know well. The developers differ, from freelancers to senior open-source maintainers, and so do the tools, from 2022 autocomplete to early-2025 Cursor with Claude. Google and Microsoft evaluated their own tools. ## Every study on one scale Shaded: 21–26% in company trials Dashed: all studies One figure A range in one study Chart values as text - Scale: % faster or more tasks done (negative means slower). Shaded range: 21–26% in company trials; dashed, all studies: -19% to 56%. - Peng et al., lab task (2023): 55.8% faster - Cui et al., three companies (2025): 26% more pull requests - Paradis et al., Google (2024): 21–26% faster - METR, 2025 (2025): 19% slower - METR, late-2025 rerun (2026): 4–18% faster (not significant) **Why medium confidence:** Five randomized studies from four independent groups agree that the effect depends on the task and the developer. The two company estimates are close but have wide margins of error and measure different things, and the studies disagree on the sign for experienced developers in familiar codebases. ## Sources Sources for F-11 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | The Impact of AI on Developer Productivity: Evidence from GitHub Copilot | Sida Peng (Microsoft Research), Eirini Kalliamvakou and Peter Cihon (GitHub), Mert Demirer (MIT Sloan) | 2023 | 95 freelance developers recruited on Upwork, mostly in India and Pakistan | Controlled experiment, 2022, on one task: writing a basic HTTP server in JavaScript | 55.8% less time (71 minutes against 161); 95% confidence interval 21% to 89% “the treated group completed the task 55.8% faster (95% confidence interval: 21-89%)” | [arxiv.org ↗](https://arxiv.org/abs/2302.06590) | | The Effects of Generative AI on High-Skilled Work: Evidence from Three Field Experiments with Software Developers | Kevin Zheyuan Cui, Mert Demirer, Sonia Jaffe, Leon Musolff, Sida Peng and Tobias Salz (Princeton, MIT, Microsoft, Wharton) | 2025 | 4,867 developers at Microsoft, Accenture and a Fortune 100 electronics maker | Three randomized experiments run by the companies over 2 to 8 months, pooled. Counts output; does not time tasks. | 26.08% more completed pull requests a week for developers who used GitHub Copilot (standard error 10.3%) “our analysis reveals a 26.08% increase (SE: 10.3%) in completed tasks among” | [economics.mit.edu ↗](https://economics.mit.edu/sites/default/files/inline-files/draft_copilot_experiments.pdf) | | How much does AI impact development speed? An enterprise-based randomized controlled trial | Elise Paradis, Satish Chandra and colleagues (Google) | 2024 | 96 full-time Google software engineers | Randomized trial on one enterprise task in Google's own tools, 2024. Google evaluated its own tools. | About 21% less time (21% to 26% across models); 96 minutes with AI against 114 without “this main effect was robust to covariates, making developers between 21% and 26% faster.” | [arxiv.org ↗](https://arxiv.org/abs/2410.12944) | | Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity | Joel Becker, Nate Rush, Beth Barnes and David Rein (METR) | 2025 | 16 experienced developers, 246 real tasks in mature open-source projects they had worked on for about 5 years | Randomized at the task level, February to June 2025, mostly Cursor with Claude 3.5 and 3.7 Sonnet | 19% more time per task with AI (95% confidence interval +2% to +39%) “we find that allowing AI actually increases completion time by 19%” | [arxiv.org ↗](https://arxiv.org/abs/2507.09089) | | We are Changing our Developer Productivity Experiment Design | METR | 2026 | 57 developers (10 from the first study, 47 new), more than 800 tasks | Same design from August 2025. Many developers declined tasks without AI, so METR calls the estimates unreliable. | 18% less time for returning developers and 4% less for new ones; both margins of error include zero “Among newly-recruited developers the estimated speedup is -4%” | [metr.org ↗](https://metr.org/blog/2026-02-24-uplift-update/) | Who gains ## Which developers gain the most from AI? Newer and more junior ones, in the three-company trials: developers with short tenure raised output 27% to 39%, against 8% to 13% for long-tenured colleagues, and junior developers 21% to 40% against 7% to 16% for senior ones. These splits are not statistically significant on their own. F-12 compares the pattern across studies. Newer tools ## Do newer AI coding agents make developers faster? There is no clean measurement yet. METR's late-2025 rerun, with 57 developers, estimated 4% to 18% less time with AI, but both estimates sit within the margin of error. Many developers declined to do tasks without AI, which METR says biases the data. ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Developers code 55% faster with GitHub Copilot.** This comes from one simple, self-contained lab task done by freelancers in 2022. METR's paper notes the task is unlike most software work. - **AI makes developers 19% slower.** METR studied 16 experienced developers in large, mature codebases with early-2025 tools. METR says the result does not show that AI slows down most developers, and its 2026 rerun leans toward a speedup. - **Copilot makes developers 26% more productive.** The 26% is the rise in pull requests completed per week for developers who used the tool, with a standard error of 10 points. It counts output, and measures neither time per task nor code quality. ## More from What AI Saves - [F-08**3–5.4%**How much time does AI save knowledge workers?](https://evy.so/facts/time-saved-by-ai-at-work/) - [F-09**25–40%**How much faster do people finish writing tasks with AI, and is the work better?](https://evy.so/facts/ai-writing-speed-and-quality/) - [F-10**14–15%**How much more do customer support agents get done with AI?](https://evy.so/facts/ai-customer-support-productivity/) - [F-12**27–46%**Does AI help less-experienced workers more than experienced ones?](https://evy.so/facts/ai-helps-less-experienced-workers-more/) - [F-13**14–25%**When does AI make work slower or worse?](https://evy.so/facts/when-ai-makes-work-worse/) [The whole chapter](https://evy.so/facts/what-ai-saves/) · [Chapter poster](https://evy.so/facts/what-ai-saves/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How much more do customer support agents get done with AI? (14–15%) | The AI at Work Fact Book" description: "Support agents with an AI assistant resolved 14 to 15 percent more issues per hour in one firm's study; an Alibaba trial found 1 to 8 percent. Who gains, sources and misquoted figures." last_updated: "2026-10-24" source: "https://evy.so/facts/ai-customer-support-productivity/" -------------------------------------------------------------------------------- F-10 · What AI Saves # How much more do customer support agents get done with AI? In the largest real-world study, chat agents at a Fortune 500 software firm resolved 14 to 15 percent more issues per hour with an AI assistant, and the least skilled gained 30 to 36 percent. A larger randomized trial at Alibaba found much smaller speed gains, with chats 1 to 8 percent faster. 14–15% more issues resolved per hour, in one firm's study Across all 2 studies: 1 to 15 percent, counting the Alibaba trial. - Low confidence - 2 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). How much more do customer support agents get done with AI? The AI at Work Fact Book, F-10. https://evy.so/facts/ai-customer-support-productivity/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/ai-customer-support-productivity.json) [Markdown](https://evy.so/facts/ai-customer-support-productivity.md) ## How to read it The two studies measure different things. The software firm's figure counts issues resolved per hour, which captures both faster chats and handling several chats at once, with 2020 to 2021 data and a GPT-3-era tool. The Alibaba trial measures the length of single chats over four weeks, among new agents only, with an assistant that helps only at the start of each chat. They agree on who gains most and disagree on how much overall. ## Every study on one scale Shaded: 14–15% Dashed: both studies One figure A range in one study Chart values as text - Scale: % faster or more issues resolved. Shaded range: 14–15%; dashed, both studies: 1.1% to 15.2%. - Brynjolfsson, Li and Raymond (2025): 13.8–15.2% more resolved - Ni et al., Alibaba (2026): 1.1–8.2% faster **Why low confidence:** The 14 to 15 percent comes from a single study at a single firm, in two versions of the same paper. The only other large field experiment, at Alibaba, measures different things and finds much smaller speed gains. ## Sources Sources for F-10 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Generative AI at Work (Quarterly Journal of Economics) | Erik Brynjolfsson (Stanford), Danielle Li and Lindsey Raymond (MIT) | 2025 | 5,172 chat support agents at one Fortune 500 software firm, mostly in the Philippines, 3 million chats | Staggered rollout of a GPT-3-based reply assistant, 2020 to 2021, compared before and after with the firm's own records. The 2023 NBER version gives 13.8%. | 15.2% more issues resolved per hour; handle time down 8.5%; customer satisfaction unchanged “increases worker productivity, as measured by issues resolved per hour, by 15% on average” | [doi.org ↗](https://doi.org/10.1093/qje/qjae044) | | Generative AI in Action: Field Experimental Evidence from Alibaba's Customer Service Operations | Xiao Ni, Yiwei Wang, Tianjun Feng, Lauren Xiaoyuan Lu, Yitong Wang and Congyi Zhou (Fudan, Zhejiang University, Dartmouth Tuck, Alibaba Group) | 2026 | 5,940 new agents with under a year of tenure, about 2.56 million chats on Taobao, China | Four-week randomized trial run by Alibaba, January to February 2024. The assistant drafts a diagnosis and solution at the start of each chat. One author works at Alibaba. | Issues identified 8.2% faster and chats 1.1% shorter; dissatisfaction down 3.4% “reduced issue identification time and chat duration by 8.2% and 1.1%, respectively” | [arxiv.org ↗](https://arxiv.org/abs/2603.29888) | Who gains ## Which support agents gain the most from AI? The weakest. At the software firm, less skilled and less experienced agents gained 30% and the lowest-skilled fifth 36%, while the most skilled saw no significant change and their quality slipped slightly. At Alibaba, customer ratings for the lowest-performing fifth of agents rose 0.81 points on a 5-point scale and fell 0.28 points for the top fifth. F-12 compares this pattern across studies. ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **AI makes support agents 14% more productive, and novices 34% more.** Those are the 2023 working-paper figures. The peer-reviewed 2025 version says 15% on average, 30% for less skilled and less experienced agents, and 36% for the lowest-skilled fifth. - **Klarna's AI assistant does the work of 700 agents.** Klarna's 2024 press release describes a chatbot answering chats on its own. It gives no method and says nothing about how productive human agents are with AI. ## More from What AI Saves - [F-08**3–5.4%**How much time does AI save knowledge workers?](https://evy.so/facts/time-saved-by-ai-at-work/) - [F-09**25–40%**How much faster do people finish writing tasks with AI, and is the work better?](https://evy.so/facts/ai-writing-speed-and-quality/) - [F-11**21–26%**How much faster do developers code with AI assistants?](https://evy.so/facts/ai-coding-speed/) - [F-12**27–46%**Does AI help less-experienced workers more than experienced ones?](https://evy.so/facts/ai-helps-less-experienced-workers-more/) - [F-13**14–25%**When does AI make work slower or worse?](https://evy.so/facts/when-ai-makes-work-worse/) [The whole chapter](https://evy.so/facts/what-ai-saves/) · [Chapter poster](https://evy.so/facts/what-ai-saves/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How much do European firms care where their AI processes data? (53–66%) | The AI at Work Fact Book" description: "53 to 66 percent of European firms that engage with AI cite data protection as an obstacle, yet most German firms run US AI tools. Eurostat, Bitkom, BSI and TÜV, and DIHK." last_updated: "2026-10-24" source: "https://evy.so/facts/ai-data-protection-in-europe/" -------------------------------------------------------------------------------- F-39 · Europe # How much do European firms care where their AI processes data? A lot, by what they say: 53 to 66 percent of European firms that use or have considered AI cite data protection as an obstacle, more than any other concern. What they use tells a different story: in Germany, 93 percent said they would prefer a German AI provider, yet 76 percent of AI-using firms use ChatGPT and only 35 percent use any European or German AI application. 53–66% of firms that use or considered AI cite data protection as an obstacle - Medium confidence - 3 primary sources - 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). How much do European firms care where their AI processes data? The AI at Work Fact Book, F-39. https://evy.so/facts/ai-data-protection-in-europe/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/ai-data-protection-in-europe.json) [Markdown](https://evy.so/facts/ai-data-protection-in-europe.md) ## How to read it Always check the base. Eurostat asks only firms that considered AI and passed, about 14% of non-users; across all EU firms the same answer is 5.8%. Bitkom asks firms already using AI. Stated preferences (which provider firms would choose) and actual use (which tools they run) diverge widely, so both are shown. There is no EU-wide official statistic on where firms want their AI data hosted. ## Every study on one scale Shaded: 53–66% One figure Chart values as text - Scale: % citing data protection as an obstacle. Shaded range: 53–66%. - Eurostat, EU (firms that considered AI) (2026): EU 52.7% - Eurostat, Germany (firms that considered AI) (2026): Germany 60.0% - Bitkom, German AI users (2026): 66% (AI users) **Why medium confidence:** Two independent creators with different methods agree that data protection is the top or near-top concern among firms that engage with AI. The bases differ sharply, the evidence on preferred provider origin is German only, and the one multi-country sovereignty survey is a vendor study. ## Sources Sources for F-39 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Artificial intelligence by size class of enterprise (dataset isoc\_eb\_ai) | Eurostat | 2026 | About 157,000 EU enterprises with 10 or more employees, early 2025 | Official statistics; asked only of non-users that had considered AI (14% of non-users) | Data protection and privacy concerns: EU 52.72% (42.42% in 2023), Germany 60.01%, Spain 53.79%, Netherlands 47.34%; as a share of all EU firms 5.82% (Germany 8.97%) “because of concerns regarding violation of data protection and privacy” | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/databrowser/view/isoc_eb_ai/default/table) | | Artificial intelligence by size class of enterprise (dataset isoc\_eb\_ai) | Eurostat | 2026 | German enterprises with 10 or more employees in the same survey | Same survey and base | Germany 60.01% (53.01% in 2024); unclear legal consequences 62.22% | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/databrowser/view/isoc_eb_ai/default/table) | | KI aus China? Für die deutsche Wirtschaft praktisch kein Thema (press release, 9 September 2026) | Bitkom | 2026 | 603 companies with 20 or more employees in Germany, July to August 2026 | Representative telephone survey by the German digital industry association | Data protection requirements are the top obstacle for AI users (66%); tools used: ChatGPT 76%, Microsoft Copilot 35%, Google Gemini 28%, Mistral 0% “Vorne stehen hier Anforderungen an den Datenschutz mit 66 Prozent” | [bitkom.org ↗](https://www.bitkom.org/Presse/Presseinformation/KI-aus-China-fuer-deutsche-Wirtschaft-kein-Thema) | Say and do ## Do European firms use European AI providers? Mostly not, in Germany. In 2025, 88% of German firms said the provider's country matters and 93% would prefer a German AI provider (firms could name several; 51% also preferred the US). In 2026, 76% of German AI-using firms use ChatGPT, and no surveyed firm reported using Mistral (Bitkom). Only 35% of German AI users use any European or German AI application, and 54% run their AI mainly in an external cloud (BSI and TÜV-Verband). Do European firms use European AI providers? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Durchbruch bei Künstlicher Intelligenz (press release and chart deck, 15 September 2025) | Bitkom | 2025 | 604 companies with 20 or more employees in Germany, July to August 2025 | Representative telephone survey; several origins could be preferred | 88% say provider origin is important; 93% prefer Germany, 51% the US, 40% other EU “Dabei würden 93 Prozent eine KI aus Deutschland bevorzugen” | [bitkom.org ↗](https://www.bitkom.org/Presse/Presseinformation/Durchbruch-Kuenstliche-Intelligenz) | | KI-Sicherheitsstudie 2026 | Federal Office for Information Security (BSI) and TÜV-Verband, survey by forsa | 2026 | 505 companies with 20 or more employees in Germany, June to July 2026 | Representative survey for a federal authority and the testing association | 35% of AI users use a European or German AI application; 54% run AI mainly in an external cloud; 93% rate data protection important when choosing AI “Bei rund einem Drittel (35 Prozent) ist dies der Fall.” | [tuev-verband.de ↗](https://www.tuev-verband.de/fileadmin/user_upload/Content_local/Studien_local/2026_BSI-TUEV-KI-Sicherheitsstudie.pdf) | Dependence ## How dependent do European firms feel on non-European AI? It depends on how you ask. 68% of German firms say they are largely or completely dependent on non-EU countries for AI (DIHK, late 2025), but only 30% of German AI users feel strongly dependent on non-European AI providers, and 10% call it a big problem (BSI and TÜV-Verband, 2026). How dependent do European firms feel on non-European AI? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Künstliche Intelligenz, Souveränität und Resilienz: DIHK-Digitalisierungsumfrage 2026 | Deutsche Industrie- und Handelskammer (DIHK) | 2026 | 4,686 member companies of all sizes, November 2025 | Online survey of chamber of commerce members | 68% largely or completely dependent on non-EU countries for AI; 76% for operating systems, 55% for cloud “weitgehend oder vollkommen abhängig von Technologien, Innovationen oder Dienstleistungen aus Nicht-EU-Ländern” | [dihk.de ↗](https://www.dihk.de/resource/blob/163540/486d8384c94a630d9aa0eaa978f0a786/dihk-digitalisierungsumfrage-2026-data.pdf) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **93% of German companies want German AI (Bitkom, 2026).** It is the 2025 survey; firms could prefer several origins (51% also preferred the US), and it is a stated preference, not buying behavior. - **53% of German companies distrust non-European AI providers (DIHK).** The DIHK report has no trust question. It shows 68% feel largely or completely dependent on non-EU countries for AI. - **Half of EU companies avoid AI over data protection (Eurostat).** The base is only non-users that had considered AI. Across all EU firms, 5.8% cite data protection (2025). - **62% of European firms want sovereign AI (Accenture).** A vendor survey of large organizations that already rate sovereignty as highly important. It is not a cross-section of European firms. ## More from Europe - [F-34**5–42%**Which EU countries' companies use AI the most and least?](https://evy.so/facts/ai-use-by-eu-country/) - [F-35**9–37%**What share of workers in each EU country use generative AI for work?](https://evy.so/facts/genai-use-for-work-by-eu-country/) - [F-36**7–17 pts**How does AI use at work in Europe compare with the US?](https://evy.so/facts/ai-at-work-europe-vs-us/) - [F-37**48–86%**How do European workers feel about AI at work?](https://evy.so/facts/european-attitudes-to-ai-at-work/) - [F-38**28%**How ready are European companies for the EU AI Act?](https://evy.so/facts/eu-ai-act-readiness/) - [F-40**12–14×**How does AI investment in Europe compare with the US?](https://evy.so/facts/ai-investment-europe-vs-us/) [The whole chapter](https://evy.so/facts/europe/) · [Chapter poster](https://evy.so/facts/europe/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How often is AI output wrong on real work tasks? (17–45%) | The AI at Work Fact Book" description: "17 to 45 percent of AI answers had a hallucination or significant error in independent tests on legal and news questions. Summaries, whole projects and how fast rates change." last_updated: "2026-10-24" source: "https://evy.so/facts/ai-error-rates-at-work/" -------------------------------------------------------------------------------- F-23 · What Teams Trust # How often is AI output wrong on real work tasks? In independent tests on realistic professional questions, 17 to 45 percent of AI answers contained a hallucination or significant error (2024 to 2025). Rates fall sharply when the AI only summarizes a document it is given, and rise sharply for whole projects, and they change with every model release. 17–45% of AI answers had a hallucination or significant error, in independent tests (2024 to 2025) - Medium confidence - 3 primary sources - 2025 - Last checked October 14, 2026 ## Cite this EVY (2026). How often is AI output wrong on real work tasks? The AI at Work Fact Book, F-23. https://evy.so/facts/ai-error-rates-at-work/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/ai-error-rates-at-work.json) [Markdown](https://evy.so/facts/ai-error-rates-at-work.md) ## How to read it Each test sets a different task and defines an error differently, and humans grade all three. Dates matter: the legal tools were tested in 2024 and have been updated since, and in the BBC's own comparison significant issues fell from 51% to 37% in six months. The citation test asks AI to find the source of a quote, a narrow task, so it sits above the range. Error rates depend heavily on whether the AI is given the source material. ## Every study on one scale Shaded: 17–45% One figure A range in one study Chart values as text - Scale: % of answers with a serious error. Shaded range: 17–45%. - Stanford, legal research tools (2025): 17–33% (GPT-4 43%) - EBU and BBC, news answers (2025): 45% significant issue, 20% accuracy - Tow Center, news citations (2025): over 60% (citations only) **Why medium confidence:** Two independent creators using expert human grading find serious-error rates in this band for question answering. The tasks, error definitions and models differ, the tests are one to two years old, and rates move quickly with each model release. ## Sources Sources for F-23 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools (Journal of Empirical Legal Studies) | Varun Magesh, Faiz Surani, Matthew Dahl, Mirac Suzgun, Christopher D. Manning and Daniel E. Ho (Stanford RegLab and Yale) | 2025 | 202 preregistered legal research queries per system, tested March to May 2024 | Hand-graded by experts. Hallucinated means incorrect, or citing a source that does not support the claim. | Lexis+ AI 17%, Ask Practical Law AI 17%, Westlaw AI-Assisted Research 33% hallucinated; GPT-4 without retrieval 43% “each hallucinate between 17% and 33% of the time” | [doi.org ↗](https://doi.org/10.1111/jels.12413) Open-access article; the Wiley page blocks automated access. | | News Integrity in AI Assistants: An international PSM study | European Broadcasting Union and BBC, with 22 public service media organizations | 2025 | 2,709 answers from free ChatGPT, Copilot, Gemini and Perplexity, late May to early June 2025, 18 countries | Journalists rated each answer on accuracy, sourcing, context and more | 45% had at least one significant issue; 20% a significant accuracy issue; Gemini 76%, Copilot 37%, ChatGPT 36%, Perplexity 30% “Overall, 45% of all AI responses were found to have at least one ‘significant’ issue.” | [ebu.ch ↗](https://www.ebu.ch/files/live/sites/ebu/files/Publications/MIS/open/EBU-MIS-BBC_News_Integrity_in_AI_Assistants_Report_2025.pdf) | | AI Search Has a Citation Problem (Columbia Journalism Review) | Klaudia Jaźwińska and Aisvarya Chandrasekar, Tow Center for Digital Journalism | 2025 | 1,600 queries to eight AI search tools, February 2025 | Manual grading of whether each tool identified the source article of a quoted excerpt. A narrow citation task. | Incorrect answers to more than 60% of queries; Perplexity 37%, Grok 3 94% “Collectively, they provided incorrect answers to more than 60 percent of queries.” | [cjr.org ↗](https://www.cjr.org/tow_center/we-compared-eight-ai-search-engines-theyre-all-bad-at-citing-news.php) | Summarizing a document ## How often does AI invent content when summarizing a document? Much less often than when answering from its own knowledge. On Vectara's leaderboard of 22 September 2026, 108 models added unsupported content to 1.8% to 24.2% of summaries, with a median of about 9.6% and frontier models at about 6% to 12%. The judging is done by Vectara's own detection model, not by people. How often does AI invent content when summarizing a document? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Hallucination Leaderboard (HHEM-2.3) | Vectara | 2026 | Over 7,700 articles summarized by each of 108 models; leaderboard dated 22 September 2026 | Automated benchmark judged by Vectara's own detection model. Vectara sells retrieval and hallucination-detection products. | Hallucination rates from 1.8% to 24.2% of summaries; median about 9.6% “This evaluates how often an LLM introduces hallucinations when summarizing a document.” | [github.com ↗](https://github.com/vectara/hallucination-leaderboard) | Whole projects ## How often can AI agents complete whole work projects? Rarely, but fast-rising. On the Remote Labor Index, which gives AI agents real paid freelance projects averaging 29 hours of human work, the best agent delivered work a client would accept on 21.25% of projects as of 14 October 2026, up from 2.5% at launch in October 2025. How often can AI agents complete whole work projects? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Remote Labor Index: Measuring AI Automation of Remote Work | Center for AI Safety and Scale AI | 2026 | 230 real freelance projects in design, data analysis, game development, video and more | Human evaluators judged whether each AI deliverable matched the professional's and was acceptable as commissioned work. Scale AI sells AI evaluation. | Best agent 21.25% of projects (read 14 October 2026); 2.5% at launch in October 2025 “The best-performing model achieves an automation rate of only 21.25%.” | [labs.scale.com ↗](https://labs.scale.com/leaderboard/rli) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **AI assistants get the news wrong 45% of the time (EBU and BBC).** 45% of answers had at least one significant issue of any kind, mostly sourcing. Significant accuracy errors were 20%. - **AI search engines are wrong 60% of the time (Tow Center).** The test asked eight tools to find the source article of a quoted excerpt. It measures news citation, not general accuracy. - **Legal AI tools are hallucination-free.** Stanford's preregistered test found the tools hallucinated on 17% to 33% of queries. Vendor wording refers to whether linked citations exist, a narrower claim. - **Chatbots hallucinate up to 27% of the time (Vectara).** That comes from 2023 coverage of Vectara's first leaderboard. In September 2026 the range across 108 models was 1.8% to 24.2%, on a new test. - **AI can do only 2.5% of remote work.** 2.5% was the best agent at the Remote Labor Index launch in October 2025. In October 2026 it was 21.25%, counting whole projects done to client standard. ## More from What Teams Trust - [F-21**24–53%**How much do workers trust the output of AI tools at work?](https://evy.so/facts/trust-in-ai-output/) - [F-22**41–44%**How often do people check AI output before using it?](https://evy.so/facts/checking-ai-output/) - [F-24**2–12%**Which work tasks do people want AI to do on its own, and which do they want to keep?](https://evy.so/facts/tasks-workers-want-ai-to-do/) - [F-25**30–44%**How many workers use AI in ways that break company policy?](https://evy.so/facts/ai-use-against-company-policy/) - [F-26**42–48%**How often do employees put company data into public AI tools?](https://evy.so/facts/company-data-in-public-ai/) - [F-27**21–29%**How many organizations have had a security incident or data breach involving AI?](https://evy.so/facts/ai-security-breaches/) [The whole chapter](https://evy.so/facts/what-teams-trust/) · [Chapter poster](https://evy.so/facts/what-teams-trust/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "Does AI help less-experienced workers more than experienced ones? (27–46%) | The AI at Work Fact Book" description: "On well-defined tasks, the newest or weakest workers gain 27 to 46 percent from AI and the strongest 0 to 17 percent. Where the pattern reverses, with sources." last_updated: "2026-10-24" source: "https://evy.so/facts/ai-helps-less-experienced-workers-more/" -------------------------------------------------------------------------------- F-12 · What AI Saves # Does AI help less-experienced workers more than experienced ones? Mostly yes, on well-defined tasks: the newest or weakest workers gained about 27 to 46 percent in support, consulting and software studies, against 0 to 17 percent for the strongest. On open-ended business decisions the pattern reversed: in Kenya, strong entrepreneurs gained from an AI adviser while weak ones did nearly 10 percent worse. 27–46% gain for the newest or weakest workers; the strongest gain 0 to 17% - Medium confidence - 5 primary sources - 2023 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). Does AI help less-experienced workers more than experienced ones? The AI at Work Fact Book, F-12. https://evy.so/facts/ai-helps-less-experienced-workers-more/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/ai-helps-less-experienced-workers-more.json) [Markdown](https://evy.so/facts/ai-helps-less-experienced-workers-more.md) ## How to read it Studies split workers in different ways, by earlier performance or by tenure, and only the support and software studies speak directly to experience. Outcomes differ: issues resolved per hour, graded quality, pull requests and customer ratings. The BCG 43% and 17% compare consultants with their own earlier work and appear in the 2023 working paper; the published 2026 version keeps the finding without printing the numbers. Noy and Zhang and Alibaba show the same narrowing without a comparable percentage, so they are in the table only. ## Every study on one scale Shaded: 27–46% One figure A range in one study Chart values as text - Scale: % gain for the newest or weakest group. Shaded range: 27–46%. - Brynjolfsson et al., support (2025): 30–46% (top skill: about 0) - Dell'Acqua et al., BCG (2023): 43% (top half 17%) - Cui et al., software (2026): 27–39% (long tenure 8–13%) **Why medium confidence:** Five independent teams in different settings find bigger gains for weaker or newer workers. Each measures a different outcome, the software gaps are not statistically significant, and a randomized trial on open-ended business decisions finds the reverse. Studies without a figure on this scale are listed in the table only. ## Sources Sources for F-12 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Generative AI at Work (Quarterly Journal of Economics) | Erik Brynjolfsson (Stanford), Danielle Li and Lindsey Raymond (MIT) | 2025 | 5,172 customer support agents at one Fortune 500 software firm | Staggered rollout, compared before and after, split by earlier skill and by tenure. The tenure figure is from the 2023 NBER version. | Less skilled and less experienced agents +30%; lowest-skilled fifth +36%; agents with under a month of tenure about +46%; no significant change for the most skilled “Less skilled and less experienced workers improve significantly across all productivity measures” | [doi.org ↗](https://doi.org/10.1093/qje/qjae044) | | Navigating the Jagged Technological Frontier (HBS Working Paper 24-013) | Fabrizio Dell'Acqua, Ethan Mollick, Karim R. Lakhani and colleagues, with Boston Consulting Group | 2023 | 758 BCG consultants | Preregistered randomized experiment. Gains compare AI users with their own earlier scores on a baseline task. | Consultants below the median improved 43%, those above 17% “those below the average performance threshold increasing by 43% and those above increasing by 17%” | [hbs.edu ↗](https://www.hbs.edu/faculty/Pages/item.aspx?num=64700) [Archived copy ↗](https://web.archive.org/web/20240102034642/https://www.hbs.edu/ris/Publication%20Files/24-013_d9b45b68-9e74-42d6-a1c6-c72fb70c7282.pdf) HBS no longer hosts this version of the PDF; the archived copy is the 2023 working paper we read. | | The Effects of Generative AI on High-Skilled Work: Evidence from Three Field Experiments with Software Developers (Management Science) | Kevin Zheyuan Cui, Mert Demirer, Sonia Jaffe, Leon Musolff, Sida Peng and Tobias Salz | 2026 | 4,867 developers at Microsoft, Accenture and a Fortune 100 electronics maker | Three randomized company experiments, split by tenure and job level. The splits are not statistically significant. | Short-tenure developers +27% to 39% completed tasks; long-tenure +8% to 13% “less experienced developers had higher adoption rates and greater productivity gains.” | [doi.org ↗](https://doi.org/10.1287/mnsc.2025.00535) | | Experimental evidence on the productivity effects of generative artificial intelligence (Science) | Shakked Noy and Whitney Zhang (MIT) | 2023 | 453 college-educated professionals | Preregistered online randomized experiment on writing tasks | The gap between stronger and weaker writers halved: low scorers improved in grade and speed, high scorers kept their grade and got faster “ChatGPT compresses the productivity distribution by benefiting low-ability workers more.” | [science.org ↗](https://www.science.org/doi/10.1126/science.adh2586) | | Generative AI in Action: Field Experimental Evidence from Alibaba's Customer Service Operations | Xiao Ni, Yiwei Wang, Tianjun Feng, Lauren Xiaoyuan Lu, Yitong Wang and Congyi Zhou | 2026 | 5,940 new customer service agents, about 2.56 million chats | Four-week randomized trial run by Alibaba, split by earlier customer ratings | Ratings for the lowest-performing fifth rose 0.81 points on a 5-point scale and fell 0.28 for the top fifth “Lower-performing agents benefit the most, indicating that gen AI can narrow performance gaps.” | [arxiv.org ↗](https://arxiv.org/abs/2603.29888) | Where it reverses ## When does AI help strong performers more than weak ones? On open-ended decisions, where people must judge which advice fits. In a five-month trial with 640 entrepreneurs in Kenya who got a GPT-4 business mentor on WhatsApp, low performers' revenue fell about 9% while high performers' rose about 18%, a gain the authors treat as tentative. They trace the gap to weaker owners acting on generic advice such as price cuts. When does AI help strong performers more than weak ones? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | The Uneven Impact of Generative AI on Entrepreneurial Performance: Evidence from a Field Experiment in Kenya (HBS Working Paper 24-042) | Nicholas G. Otis, Rowan Clarke, Solène Delecourt, David Holtz and Rembrand Koning | 2026 | 640 entrepreneurs in Kenya, five months | Randomized field experiment: a GPT-4 mentor on WhatsApp or a standard business guide | No average effect; low performers about 9% lower revenue, high performers about 18% higher (p = 0.07) “low performers did nearly 10% worse due to the AI assistant” | [hbs.edu ↗](https://www.hbs.edu/ris/Publication%20Files/24-042_567db548-5b24-4182-b703-9d5ef1328043.pdf) [Archived copy ↗](https://web.archive.org/web/20261009165007/https://www.hbs.edu/ris/Publication%20Files/24-042_567db548-5b24-4182-b703-9d5ef1328043.pdf) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **AI mostly helps top scientists (Toner-Rodgers, MIT, 2024).** MIT said in May 2025 that it has no confidence in the paper's data and asked for it to be withdrawn. - **Below-average BCG consultants improved 43% more than the control group.** The 43% and 17% compare consultants using AI with their own scores on an earlier task without AI. - **Novice support agents gained 34% with AI.** That is the 2023 working-paper figure. The peer-reviewed 2025 version says 30% for less skilled and less experienced agents and 36% for the lowest-skilled fifth. - **AI levels the playing field for every worker.** It holds in studies of well-defined tasks. In the Kenyan trial on open-ended business decisions, AI widened the gap. ## More from What AI Saves - [F-08**3–5.4%**How much time does AI save knowledge workers?](https://evy.so/facts/time-saved-by-ai-at-work/) - [F-09**25–40%**How much faster do people finish writing tasks with AI, and is the work better?](https://evy.so/facts/ai-writing-speed-and-quality/) - [F-10**14–15%**How much more do customer support agents get done with AI?](https://evy.so/facts/ai-customer-support-productivity/) - [F-11**21–26%**How much faster do developers code with AI assistants?](https://evy.so/facts/ai-coding-speed/) - [F-13**14–25%**When does AI make work slower or worse?](https://evy.so/facts/when-ai-makes-work-worse/) [The whole chapter](https://evy.so/facts/what-ai-saves/) · [Chapter poster](https://evy.so/facts/what-ai-saves/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How does AI investment in Europe compare with the US? (12–14×) | The AI at Work Fact Book" description: "US private AI investment was 12 to 14 times Europe's in 2025 ($286bn against $21bn). Germany's share, firms investing in AI, and why EU pledges are not spending." last_updated: "2026-10-24" source: "https://evy.so/facts/ai-investment-europe-vs-us/" -------------------------------------------------------------------------------- F-40 · Europe # How does AI investment in Europe compare with the US? Private investment in AI companies was 12 to 14 times higher in the US than in Europe in 2025: $285.9 billion against $20.9 billion. The gap is about companies that build AI; among firms that use it, Europe roughly matches the US. Most public money is announced, not yet spent. 12–14× more private AI investment in the US than in Europe (2025) - High confidence - 2 primary sources - 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). How does AI investment in Europe compare with the US? The AI at Work Fact Book, F-40. https://evy.so/facts/ai-investment-europe-vs-us/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/ai-investment-europe-vs-us.json) [Markdown](https://evy.so/facts/ai-investment-europe-vs-us.md) ## How to read it This measures money for companies that build AI, from deal databases that miss undisclosed deals and government funds. Investment by ordinary firms in using AI is a different measure and shows near parity (below). The OECD notes that AI-related investment inside EU firms and the public sector may be far larger than venture capital: up to €294 billion in 2023 in experimental estimates. Public pledges are shown separately because announced is not spent. ## Every study on one scale Shaded: 12–14× One figure Chart values as text - Scale: US private AI investment as a multiple of Europe's. Shaded range: 12–14×. - Stanford AI Index (2026): 13.7× ($285.9bn against $20.9bn) - OECD (venture capital) (2026): 12.3× ($194bn against $15.8bn) **Why high confidence:** Two independent creators using different deal databases and scopes (all private investment for Europe, venture capital for the EU27) agree on a US advantage of 12 to 14 times in 2025. ## Sources Sources for F-40 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | The 2026 AI Index Report, chapter 4 (Economy) | Stanford Institute for Human-Centered AI (data from Quid) | 2026 | Disclosed private investments in AI companies raising over $1.5 million, by headquarters | Commercial deal database; Europe appears to include the UK. Government funds are not counted. | 2025 private AI investment: US $285.9 billion, Europe $20.9 billion (13.7 times; our arithmetic), China $12.4 billion; UK $5.9 billion, France $4.4 billion, Germany $3.9 billion. US grew 160% from 2024, Europe 7% “United States was the global leader with nearly $285.9 billion total invested” | [hai.stanford.edu ↗](https://hai.stanford.edu/assets/files/ai_index_report_2026.pdf) | | Venture capital investments in artificial intelligence through 2025 (OECD policy brief) | OECD (data from Preqin) | 2026 | Preqin deal data, updated January 2026 | Venture capital deals in AI firms, by firm location; EU27 without the UK | 2025 AI venture capital: US $194 billion (75%), EU27 $15.8 billion (6%; 12.3 times less, our arithmetic), UK $13.8 billion “followed by the EU27 (6%, USD 15.8 billion)” | [oecd.org ↗](https://www.oecd.org/en/publications/venture-capital-investments-in-artificial-intelligence-through-2025_a13752f5-en.html) The OECD page blocks automated checks; the policy brief PDF is on oecd.org. | Germany ## How much private AI investment goes to Germany? $3.9 billion in 2025, third in Europe after the UK ($5.9 billion) and France ($4.4 billion). Germany had 92 newly funded AI companies, against 1,953 in the US, and $17.2 billion of private AI investment in total from 2013 to 2025, against $757.3 billion in the US (Stanford AI Index). Firms using AI ## Do European companies invest in using AI as much as US companies? Roughly, yes. 37% of EU firms invest in or systematically use generative AI tools, against 36% of US firms, and 39% in Germany (EIB, 2025). The difference is breadth: 81% of US firms using AI apply it in two or more business areas, against 55% in the EU. See F-36. Do European companies invest in using AI as much as US companies? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | EIB Investment Survey 2025: European Union overview | European Investment Bank | 2025 | 12,033 EU firms and a smaller US sample, April to July 2025 | Telephone survey, weighted by value added | Generative AI tools: EU 37%, US 36%, Germany 39%; EU large firms 47%, SMEs 27% “Around 37% of EU firms use generative artificial intelligence, in line with US firms.” | [eib.org ↗](https://www.eib.org/attachments/lucalli/20250216-141025-econ-eibis-2025-eu-en.pdf) | Public money ## How much public money is the EU putting into AI? Mostly announced, not yet spent. InvestAI (February 2025) aims to mobilize €200 billion, mostly from private and other partners, including a €20 billion fund for AI gigafactories. The gigafactory call opened only on 30 July 2026 with up to €10 billion in EU and national funding, closes 12 November 2026, and awards are expected in early 2027. AI factories are co-financed with €10 billion. No total for money spent has been published. How much public money is the EU putting into AI? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | EU launches InvestAI initiative to mobilise €200 billion of investment in artificial intelligence (IP/25/467) | European Commission | 2025 | Policy announcement, 11 February 2025 | Press release; a mobilization target | Aim to mobilize €200 billion, including a €20 billion gigafactory fund “an initiative to mobilise €200 billion for investment in AI” | [ec.europa.eu ↗](https://ec.europa.eu/commission/presscorner/detail/en/ip_25_467) | | EU launches AI Gigafactories call (IP/26/1708) | European Commission | 2026 | Policy announcement, 30 July 2026 | Press release; funding available under an open call | Up to €10 billion in EU and national funding for up to seven gigafactories; call closes 12 November 2026 “supported by up to €10 billion in EU and national funding” | [ec.europa.eu ↗](https://ec.europa.eu/commission/presscorner/detail/en/ip_26_1708) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **The EU is investing €200 billion in AI (InvestAI).** The Commission's wording is a target to mobilize €200 billion, mostly from private and other partners. No spending total has been published. - **The US invests 23 times more in AI than Europe.** 23 times is the US compared with China. Against Europe it is about 13.7 times. - **European companies lag US companies in adopting AI.** Among firms using it, the EIB finds 37% of EU and 36% of US firms invest in generative AI tools. The gap is in breadth of use and in money for AI builders. - **The EU has a €20 billion AI gigafactory fund.** Announced in February 2025; the call opened in July 2026 with up to €10 billion in public funding, and no awards are due before early 2027. ## More from Europe - [F-34**5–42%**Which EU countries' companies use AI the most and least?](https://evy.so/facts/ai-use-by-eu-country/) - [F-35**9–37%**What share of workers in each EU country use generative AI for work?](https://evy.so/facts/genai-use-for-work-by-eu-country/) - [F-36**7–17 pts**How does AI use at work in Europe compare with the US?](https://evy.so/facts/ai-at-work-europe-vs-us/) - [F-37**48–86%**How do European workers feel about AI at work?](https://evy.so/facts/european-attitudes-to-ai-at-work/) - [F-38**28%**How ready are European companies for the EU AI Act?](https://evy.so/facts/eu-ai-act-readiness/) - [F-39**53–66%**How much do European firms care where their AI processes data?](https://evy.so/facts/ai-data-protection-in-europe/) [The whole chapter](https://evy.so/facts/europe/) · [Chapter poster](https://evy.so/facts/europe/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How many organizations have had a security incident or data breach involving AI? (21–29%) | The AI at Work Fact Book" description: "21 to 29 percent of organizations report a breach of or attack on their own AI systems; shadow AI was involved in 43 percent of breaches studied. IBM, Gartner, Microsoft and Cisco." last_updated: "2026-10-24" source: "https://evy.so/facts/ai-security-breaches/" -------------------------------------------------------------------------------- F-27 · What Teams Trust # How many organizations have had a security incident or data breach involving AI? About one in five to three in ten organizations report a breach of, or attack on, their own AI systems in the past year: 21 percent of breached organizations in IBM's 2026 study and 29 percent in Gartner's 2025 survey. Employees' unapproved AI was involved in 43 percent of the breaches IBM studied, up from 20 percent. 21–29% of organizations report a breach of, or attack on, their own AI systems (2025 to 2026) - Medium confidence - 3 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). How many organizations have had a security incident or data breach involving AI? The AI at Work Fact Book, F-27. https://evy.so/facts/ai-security-breaches/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/ai-security-breaches.json) [Markdown](https://evy.so/facts/ai-security-breaches.md) ## How to read it Keep three kinds of AI incident apart: attacks on an organization's own AI, incidents caused by employees' AI use, and attacks that use AI against the company, such as deepfakes. The range covers the first. IBM's sample includes only organizations that had a breach, so its shares are not of all organizations. Broad figures that mix all three run far higher (Cisco: 86% of leaders report any AI-related incident). All sources are vendor or analyst research. ## Every study on one scale Shaded: 21–29% One figure A range in one study Chart values as text - Scale: % of organizations with an AI breach or attack. Shaded range: 21–29%. - IBM and Ponemon (2026): 21% (13% a year earlier) - Gartner (2025): 29% infrastructure, 32% prompt attacks **Why medium confidence:** Two independent creators with different methods land close together, though IBM counts breaches among organizations that had one and Gartner counts attacks, successful or not. Both are commercial research, and no government statistic counts AI breaches yet. Studies without a figure on this scale are listed in the table only. ## Sources Sources for F-27 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Cost of a Data Breach Report 2026 | IBM and Ponemon Institute | 2026 | 602 organizations breached between March 2025 and February 2026; 3,558 interviews in 16 countries and regions | Interviews about one breach at each organization, so shares are of breached organizations. IBM sells security software and services. | 21% had a security incident involving their AI models or applications (13% in 2025); 92% of those lacked proper AI access controls; shadow AI involved in 43% of breaches (20% in 2025) “AI-related breaches grew to 21% this year from 13% last year.” | [ibm.com ↗](https://www.ibm.com/reports/data-breach) | | Gartner Survey Reveals GenAI Attacks Are on the Rise | Gartner | 2025 | 302 cybersecurity leaders in North America, EMEA and Asia/Pacific, March to May 2025 | Survey reported in a press release; attacks experienced, whether or not they succeeded. Gartner sells research. | 29% faced an attack on generative AI application infrastructure; 32% an attack using an AI application's prompt; 62% a deepfake attack “experienced an attack on enterprise GenAI application infrastructure in the last 12 months” | [gartner.com ↗](https://www.gartner.com/en/newsroom/press-releases/2025-09-22-gartner-survey-reveals-generative-artificial-intelligence-attacks-are-on-the-rise) | | 2026 Data Security Index | Microsoft, survey by Hypothesis Group | 2026 | 1,725 data security leaders in 10 countries, July to August 2025 | Online survey; a share of incidents, not of organizations. Microsoft sells data security products. | 32% of organizations' data security incidents involve the use of generative AI tools “a third (32%) of data security incidents involve use of GenAI tools” | [marketingassets.microsoft.com ↗](https://marketingassets.microsoft.com/gdc/gdccYW33N/original) | Shadow AI ## How much do employees' unapproved AI tools add to breaches? More each year. Shadow AI, where workers use AI their employer has not approved, was involved in 43% of the breaches IBM studied in 2026, up from 20%. Those incidents averaged USD 5.39 million, against USD 4.99 million for a breach overall, and about one in five led to a regulatory fine. 68% of breached organizations lacked AI governance policies. Attacks using AI ## How many organizations face attacks that use AI? Most. 62% of organizations faced a deepfake attack in the past year (Gartner, 2025), and 86% of business leaders reported at least one AI-related security incident, a broad category that includes AI-written phishing (Cisco, 2025). The UK's official breaches survey does not yet ask about AI breaches; it finds that 24% of businesses using or considering AI have practices to manage its cyber risks. How many organizations face attacks that use AI? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | 2025 Cisco Cybersecurity Readiness Index | Cisco | 2025 | 8,000 business leaders with security responsibilities in 30 markets, early 2025 | Double-blind survey; any AI-related incident, including attacks that use AI. Cisco sells security products. | 86% reported at least one AI-related security incident in the past 12 months “reported at least one AI related incident in the past 12 months” | [newsroom.cisco.com ↗](https://newsroom.cisco.com/c/dam/r/newsroom/en/us/interactive/cybersecurity-readiness-index/2025/documents/2025_Cisco_Cybersecurity_Readiness_Index.pdf) | | Cyber security breaches survey 2025/2026 | UK Department for Science, Innovation and Technology and Home Office | 2026 | 2,112 UK businesses and 1,085 charities, August to December 2025 | Official statistics, random probability survey | 24% of businesses using or considering AI have cyber security practices to manage AI risks “around a quarter of businesses (24%) and charities (27%) reported having cyber security practices” | [gov.uk ↗](https://www.gov.uk/government/statistics/cyber-security-breaches-survey-20252026/cyber-security-breaches-survey-20252026) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **13% of organizations have had an AI breach (IBM).** IBM studies only organizations that had a breach, and the 13% was 2025. The 2026 edition reports 21%. - **97% of organizations lack AI access controls (IBM).** The 97% (2025) was of organizations that had an AI breach; it is 92% in 2026. Across all organizations studied, 40% use AI access controls. - **Shadow AI adds $670,000 to the cost of a breach (IBM).** That 2025 comparison was one of two analyses; the other gave about $200,000. In 2026 shadow AI incidents averaged $5.39 million against $4.99 million overall. - **86% of organizations have suffered AI security incidents (Cisco).** It counts any AI-related incident reported by leaders, including AI-enhanced phishing aimed at the company. It is not breaches of the company's own AI. - **99% of companies have lost money because of AI (EY).** EY's respondents were executives at firms above $1 billion revenue, reporting any loss from any AI risk, mostly regulatory and sustainability setbacks. ## More from What Teams Trust - [F-21**24–53%**How much do workers trust the output of AI tools at work?](https://evy.so/facts/trust-in-ai-output/) - [F-22**41–44%**How often do people check AI output before using it?](https://evy.so/facts/checking-ai-output/) - [F-23**17–45%**How often is AI output wrong on real work tasks?](https://evy.so/facts/ai-error-rates-at-work/) - [F-24**2–12%**Which work tasks do people want AI to do on its own, and which do they want to keep?](https://evy.so/facts/tasks-workers-want-ai-to-do/) - [F-25**30–44%**How many workers use AI in ways that break company policy?](https://evy.so/facts/ai-use-against-company-policy/) - [F-26**42–48%**How often do employees put company data into public AI tools?](https://evy.so/facts/company-data-in-public-ai/) [The whole chapter](https://evy.so/facts/what-teams-trust/) · [Chapter poster](https://evy.so/facts/what-teams-trust/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How many workers use AI in ways that break company policy? (30–44%) | The AI at Work Fact Book" description: "30 to 44 percent of workers have used AI against company policy, and 56 percent without knowing if it was allowed. Employee reports, IT-leader estimates and policy gaps." last_updated: "2026-10-24" source: "https://evy.so/facts/ai-use-against-company-policy/" -------------------------------------------------------------------------------- F-25 · What Teams Trust # How many workers use AI in ways that break company policy? About three to four in ten workers have used AI in ways that break company policy: 44 percent of AI-using employees say so themselves, and IT leaders estimate about 30 percent. Missing rules are a big part of it: 56 percent have used AI without knowing whether it was allowed, and only 38 to 44 percent of organizations have a formal AI policy. 30–44% of workers have used AI against company policy (44% self-reported, about 30% by IT leaders' estimate) - Medium confidence - 3 primary sources - 2024 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). How many workers use AI in ways that break company policy? The AI at Work Fact Book, F-25. https://evy.so/facts/ai-use-against-company-policy/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/ai-use-against-company-policy.json) [Markdown](https://evy.so/facts/ai-use-against-company-policy.md) ## How to read it Two of these count workers: the employee self-report (44%) and the IT-leader estimate (about 30%). The third is a stated intention, not a count of violations. Many breaches come from unclear rules: most outside-tool use happens where no explicit rule exists, and bans tend to push use out of sight, since uploading company data is most common where AI is banned (67%). Chapter 3 covers how many people bring their own AI at all. ## Every study on one scale Shaded: 30–44% One figure A range in one study Chart values as text - Scale: % of workers using AI against policy. Shaded range: 30–44%. - KPMG and Melbourne (employees) (2025): 44% ever, 31% sometimes or more - GoTo (IT leaders) (2026): about 30% (leader estimate) - Software AG (stated intention) (2024): 46% would defy a ban **Why medium confidence:** Two independent creators using different methods land in the same band: employees asked directly, and IT leaders estimating for their staff. Self-reports likely understate rule-breaking, and two of the creators sell products that address it. ## Sources Sources for F-25 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Trust, attitudes and use of artificial intelligence: A global study 2025 | University of Melbourne and KPMG International | 2025 | 32,352 employed people in 47 countries, November 2024 to January 2025 | Anonymous online survey of employees who use AI; self-reported frequency | 44% have used AI in ways that contravene policies or guidelines (31% sometimes or more often); 56% have used AI without knowing whether it was allowed “almost half admit to using AI in ways that contravene organizational policies” | [mbs.edu ↗](https://mbs.edu/-/media/PDF/Research/Trust_in_AI_Report.pdf) | | The Pulse of Work in 2026: Opportunity, Risk, and Responsibility in an AI-Driven Workplace | GoTo with Workplace Intelligence | 2026 | 1,250 IT decision-makers and 1,250 knowledge workers in 10 countries, November 2025 to January 2026 | Online survey; IT leaders estimated the share of workers who violated AI policy. GoTo sells IT software. | IT leaders estimate around 3 in 10 workers have violated their company's AI policy; 44% say their company has an AI policy “around 3 out of 10 workers have violated their company’s AI policy” | [edge.sitecorecloud.io ↗](https://edge.sitecorecloud.io/gototechnol00e8-mktglobalxm07e7-mktglobalpr9498-b870/media/goto/pdfs/ebook/ebook_pulse-of-work-2026.pdf) | | Half of all employees are Shadow AI users, new study finds | Software AG, survey by TEAM LEWIS | 2024 | 6,000 knowledge workers in the US, UK and Germany, September 2024 | Online panel survey; a hypothetical question | 46% would not give up their personal AI tools even if their organization banned them “46% wouldn’t give it up, even if it were banned” | [newscenter.softwareag.com ↗](https://newscenter.softwareag.com/en/news-stories/press-releases/2024/1022-half-of-all-employees-use-shadow-ai.html) [Archived copy ↗](https://web.archive.org/web/20251105024925/https://newscenter.softwareag.com/en/news-stories/press-releases/2024/1022-half-of-all-employees-use-shadow-ai.html) Software AG's newsroom is offline; the archived copy is the original press release. | Seen from security ## How many organizations see prohibited AI use? Most. 69% of organizations suspect or have evidence that employees use prohibited public generative AI (Gartner, 302 security leaders, 2025), and 60% of leaders lack confidence that they could even identify unapproved AI tools in their environment (Cisco, 8,000 leaders, 2025). How many organizations see prohibited AI use? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Gartner Identifies Critical GenAI Blind Spots That CIOs Must Urgently Address | Gartner | 2025 | 302 cybersecurity leaders, March to May 2025 | Survey reported in a press release; counts organizations, not workers. Gartner sells research. | 69% of organizations suspect or have evidence of prohibited public generative AI use “69% of organizations suspect or have evidence that employees are using prohibited public GenAI” | [gartner.com ↗](https://www.gartner.com/en/newsroom/press-releases/2025-11-19-gartner-identifies-critical-genai-blind-spots-that-cios-must-urgently-address0) | | 2025 Cisco Cybersecurity Readiness Index | Cisco | 2025 | 8,000 business leaders with security responsibilities in 30 markets, early 2025 | Double-blind survey. Cisco sells security products. | 60% lack confidence in identifying unapproved AI tools; 22% give employees unrestricted access to public AI tools “lack confidence in their ability to identify the use of unapproved AI tools” | [newsroom.cisco.com ↗](https://newsroom.cisco.com/c/dam/r/newsroom/en/us/interactive/cybersecurity-readiness-index/2025/documents/2025_Cisco_Cybersecurity_Readiness_Index.pdf) | Policies ## How many organizations have an AI policy? Fewer than half. 38% of organizations have a formal, comprehensive AI policy, up from 28% a year earlier, 30% have a limited one and 25% none (ISACA, 2026). In GoTo's survey, 44% of IT leaders said their company has an AI policy. How many organizations have an AI policy? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | 2026 AI Pulse Poll | ISACA | 2026 | More than 3,400 digital trust professionals worldwide | Poll of IT audit, governance, security and privacy professionals | 38% have a formal, comprehensive AI policy; 30% limited; 25% none “only 38 percent of organizations have a formal, comprehensive AI policy” | [isaca.org ↗](https://www.isaca.org/about-us/newsroom/press-releases/2026/ai-use-accelerates-while-governance-and-roi-lag-says-new-isaca-research) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **44% of workers knowingly break AI rules (KPMG).** The 44% is of employees who use AI, counts anyone who did it even rarely, and does not say knowingly. Sometimes or more often is 31%. - **69% of employees use banned AI (Gartner).** It is 69% of organizations whose security leaders suspect or have evidence of any prohibited use. - **70% of employees misuse AI at work (GoTo).** GoTo counts using AI for task types it labels sensitive. Only 22% of those employees knew their employer prohibited it. - **Over 40% of enterprises will have shadow AI incidents (Gartner).** This is a forecast for 2030. ## More from What Teams Trust - [F-21**24–53%**How much do workers trust the output of AI tools at work?](https://evy.so/facts/trust-in-ai-output/) - [F-22**41–44%**How often do people check AI output before using it?](https://evy.so/facts/checking-ai-output/) - [F-23**17–45%**How often is AI output wrong on real work tasks?](https://evy.so/facts/ai-error-rates-at-work/) - [F-24**2–12%**Which work tasks do people want AI to do on its own, and which do they want to keep?](https://evy.so/facts/tasks-workers-want-ai-to-do/) - [F-26**42–48%**How often do employees put company data into public AI tools?](https://evy.so/facts/company-data-in-public-ai/) - [F-27**21–29%**How many organizations have had a security incident or data breach involving AI?](https://evy.so/facts/ai-security-breaches/) [The whole chapter](https://evy.so/facts/what-teams-trust/) · [Chapter poster](https://evy.so/facts/what-teams-trust/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "Who uses AI at work more, by age, education and income? (12–24 pts) | The AI at Work Fact Book" description: "College graduates are 12 to 24 points more likely to use AI at work; workers 50 and older trail by 4 to 20 points; high earners 83% against 34%. US and 47-country data." last_updated: "2026-10-24" source: "https://evy.so/facts/ai-use-by-age-education-income/" -------------------------------------------------------------------------------- F-17 · Who Uses AI at Work # Who uses AI at work more, by age, education and income? College graduates use AI at work 12 to 24 percentage points more than workers without a degree, and workers 50 and older trail younger colleagues by 4 to 20 points in the US. The income gap is the widest: across 47 countries, 83 percent of high-income employees use AI at work against 34 percent of low-income employees. 12–24 pts more AI use at work among college graduates than among workers without a degree - High confidence - 5 primary sources - 2025 - Last checked October 14, 2026 ## Cite this EVY (2026). Who uses AI at work more, by age, education and income? The AI at Work Fact Book, F-17. https://evy.so/facts/ai-use-by-age-education-income/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/ai-use-by-age-education-income.json) [Markdown](https://evy.so/facts/ai-use-by-age-education-income.md) ## How to read it The sources define use differently: Pew's main question asks whether some of your work is done with AI (strict), the St. Louis Fed survey asks whether you ever used generative AI for work, and the KPMG study counts any use, even every few months, across 47 countries. Stricter measures give smaller gaps. In the US the age gap is mostly between workers under 50 and those 50 and older. Part of the education gap is about jobs: it shrinks by about a third when comparing people in the same occupation. ## Every study on one scale Shaded: 12–24 points One figure A range in one study Chart values as text - Scale: points gap, college graduates against no degree. Shaded range: 12–24 points. - Pew (2025): 12 points (28% against 16%) - Bick, Blandin and Deming (2025): 17.4–23.8 points - KPMG and Melbourne, 47 countries (2025): 24 points (68% against 44%) **Why high confidence:** Four independent creators with different methods agree on the direction and roughly the size of the education gap. The exact size depends on how use is defined, and the income gap rests mainly on one survey. Studies without a figure on this scale are listed in the table only. ## Sources Sources for F-17 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | About 1 in 5 U.S. workers now use AI in their job, up since last year | Pew Research Center | 2025 | 5,010 US workers, September 2025 | Strict question: at least some of one's work is done with AI | Bachelor's degree or more 28%, some college or less 16% | [pewresearch.org ↗](https://www.pewresearch.org/short-reads/2025/10/06/about-1-in-5-us-workers-now-use-ai-in-their-job-up-since-last-year/) | | The Rapid Adoption of Generative AI (St. Louis Fed Working Paper 2024-027F, revised October 2025) | Alexander Bick (St. Louis Fed), Adam Blandin (Vanderbilt) and David Deming (Harvard) | 2025 | 6,935 employed US adults aged 18 to 64, August and November 2024 | Nationally representative online survey; ever used generative AI for work | Less than college 23.7%, bachelor's 41.1%, graduate degree 47.5%; ages 50 to 64 at 22.6% against about 35% for younger workers “use genAI at roughly twice the rate of non-college educated workers” | [doi.org ↗](https://doi.org/10.20955/wp.2024.027) | | Trust, attitudes and use of artificial intelligence: A global study 2025 | University of Melbourne and KPMG International | 2025 | 32,352 employed people in 47 countries, November 2024 to January 2025 | Online survey; any intentional AI use at work, even every few months. Income tiers are set within each country. | University educated 68% against 44% without; ages 18 to 34 at 72% against 40% at 55+; high income 83% against low income 34% “Employees who are younger, AI trained, university-educated, higher-income earners and managers are more likely” | [mbs.edu ↗](https://mbs.edu/-/media/PDF/Research/Trust_in_AI_Report.pdf) | | 34% of U.S. adults have used ChatGPT, about double the share in 2023 | Pew Research Center | 2025 | Employed adults in a survey of 5,123 US adults, February to March 2025 | Ever used ChatGPT specifically for work | Ages 18 to 29 at 38% against 18% at 50+; postgraduates 45% against 17% with high school or less “38% of employed adults ages 18 to 29 say they have used ChatGPT” | [pewresearch.org ↗](https://www.pewresearch.org/short-reads/2025/06/25/34-of-us-adults-have-used-chatgpt-about-double-the-share-in-2023/) | | The unequal adoption of ChatGPT exacerbates existing inequalities among workers (PNAS) | Anders Humlum (Chicago Booth) and Emilie Vestergaard (University of Copenhagen) | 2025 | About 18,000 workers in 11 AI-exposed occupations in Denmark, late 2023 | Survey linked to government records; comparisons within each occupation | Each extra year of age lowers the chance of having used ChatGPT for work by 0.7 points; higher earners use it more “we find younger and less experienced workers are more likely to use ChatGPT” | [pnas.org ↗](https://www.pnas.org/doi/10.1073/pnas.2414972121) | Age ## Do older workers use AI less? Yes. In the US, workers 50 and older trail younger colleagues by 4 to 20 points depending on the measure: 13% against 17% on Pew's strict question (October 2024), 22.6% against about 35% in the St. Louis Fed survey (2024), and 18% against 38% for ChatGPT among the youngest (Pew, 2025). Across 47 countries the gap is 32 points (72% of ages 18 to 34 against 40% at 55 and older). Income and level ## Do higher earners and managers use AI more? Yes. Across 47 countries, 83% of high-income employees use AI at work against 34% of low-income employees, and 77% of managers against 34% of manual workers (KPMG and University of Melbourne). In the US, use rises with earnings up to about the top fifth, then falls slightly (St. Louis Fed survey). Gallup found leaders twice as likely as individual contributors to use AI frequently (33% against 16%, 2025). Do higher earners and managers use AI more? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | AI Use at Work Has Nearly Doubled in Two Years | Gallup | 2025 | US employees, Gallup Panel, May 2025 | Self-reported frequency of AI use, by job level and job type | Frequent AI use: leaders 33%, individual contributors 16%; white-collar 27%, production and front-line 9% “They are twice as likely as individual contributors (16%) to say they use AI” | [gallup.com ↗](https://www.gallup.com/workplace/691643/work-nearly-doubled-two-years.aspx) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **40% of college graduates use AI at work against 20% of others (St. Louis Fed).** That was the February 2025 draft. After reweighting, the October 2025 revision reports 41.1% (bachelor's) and 47.5% (graduate) against 23.7%. - **83% of high-income US workers use AI.** The 83% pools 47 countries, sets income tiers within each country, and counts use every few months. - **85% of Gen Z use AI at work, against 73% of Boomers (Microsoft).** Those are the shares of AI users in each generation who bring their own AI tools. They say nothing about how many people use AI. ## More from Who Uses AI at Work - [F-14**40–50%**What share of workers use generative AI at work?](https://evy.so/facts/share-of-workers-using-ai/) - [F-15**40–52%**What do people use AI for at work?](https://evy.so/facts/what-workers-use-ai-for/) - [F-16**78–85%**Which occupations and industries use AI the most?](https://evy.so/facts/ai-use-by-occupation-and-industry/) - [F-18**5–16 pts**Is there a gender gap in AI use at work, and how large is it?](https://evy.so/facts/gender-gap-in-ai-use/) - [F-19**50–78%**How many workers use AI tools their employer did not provide, and how many hide it?](https://evy.so/facts/bring-your-own-ai/) - [F-20**42–49%**How many employers provide AI tools, guidance or training to staff?](https://evy.so/facts/employers-providing-ai-tools/) [The whole chapter](https://evy.so/facts/who-uses-ai-at-work/) · [Chapter poster](https://evy.so/facts/who-uses-ai-at-work/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "Which EU countries' companies use AI the most and least? (5–42%) | The AI at Work Fact Book" description: "Between 5 percent (Romania) and 42 percent (Denmark) of EU companies used AI in 2025, EU average 20 percent. Germany, France, Spain, Portugal and the Nordics, with trend and generative AI." last_updated: "2026-10-24" source: "https://evy.so/facts/ai-use-by-eu-country/" -------------------------------------------------------------------------------- F-34 · Europe # Which EU countries' companies use AI the most and least? In 2025, between 5 percent of companies (Romania) and 42 percent (Denmark) used AI, among firms with 10 or more employees; the EU average was 20 percent. Germany was at 26 percent, Spain 20, France 18 and Portugal 12, with the Nordic countries and the Netherlands at 33 to 42 percent. 5–42% of companies used AI in 2025, from Romania to Denmark; EU average 20% - High confidence - 5 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). Which EU countries' companies use AI the most and least? The AI at Work Fact Book, F-34. https://evy.so/facts/ai-use-by-eu-country/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/ai-use-by-eu-country.json) [Markdown](https://evy.so/facts/ai-use-by-eu-country.md) ## How to read it Each chart row is one country from the same official survey, so the countries compare cleanly. The survey counts firms with 10 or more employees using any of eight AI technologies, with every firm counted equally. In 2025 it added an eighth technology (generating pictures, video or audio), so part of the rise from 2024 is definitional. Surveys with broader questions or that weight big firms more, such as Bitkom's 57% for German firms with 20 or more staff, give much higher numbers. ## Every study on one scale Shaded: 5–42% across EU countries One figure Chart values as text - Scale: % of companies (10+ employees) using AI, 2025. Shaded range: 5–42% across EU countries. - Eurostat, Denmark (top) (2025): Denmark 42.0% - Eurostat, Germany (2025): Germany 26.0% - Eurostat, EU average (2025): EU 19.95% - Eurostat, Romania (lowest) (2025): Romania 5.2% **Why high confidence:** One official, harmonized survey of about 157,000 firms, run the same way in every country. Two independent creators with different methods confirm the country ranking. Part of the 2024 to 2025 rise comes from a new technology added to the list. Studies without a figure on this scale are listed in the table only. ## Sources Sources for F-34 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Artificial intelligence by size class of enterprise (dataset isoc\_eb\_ai) | Eurostat | 2025 | About 157,000 EU enterprises with 10 or more employees, early 2025 | Official harmonized survey; use of any of eight listed AI technologies during the year. Excludes agriculture, mining and finance. | Denmark 42.0%, Finland 37.8%, Sweden 35.0%, Netherlands 33.2%, Norway 28.9%, Germany 26.0%, Spain 20.3%, France 18.2%, Italy 16.4%, Portugal 11.5%, Poland 8.4%, Romania 5.2%; EU 19.95% (8.06% in 2023) “Enterprises using AI technologies generating written or spoken language or programming codes” | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/databrowser/view/isoc_eb_ai/default/table?lang=en) | | Artificial intelligence by size class of enterprise (dataset isoc\_eb\_ai) | Eurostat | 2025 | German enterprises with 10 or more employees in the same survey | Same survey and definition | Germany 25.97% (11.55% in 2023); AI that generates text, speech or code 9.0% | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/databrowser/view/isoc_eb_ai/default/table?lang=en) | | Use of artificial intelligence in enterprises (Statistics Explained) | Eurostat | 2025 | 157,000 enterprises surveyed in 2025 | Eurostat's commentary on the same survey | EU average 19.95%, up 6.47 points from 2024; country range 5.21% to 42.03% “the share of enterprises using AI ranged between 5.21% and 42.03%” | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises) | | Artificial intelligence by size class of enterprise (dataset isoc\_eb\_ai) | Eurostat | 2025 | Romanian enterprises with 10 or more employees in the same survey | Same survey and definition | Romania 5.21% (1.51% in 2023) | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/databrowser/view/isoc_eb_ai/default/table?lang=en) | | Mind the Gap: AI Adoption in Europe and the US (St. Louis Fed Working Paper 2026-003) | Alexander Bick, Adam Blandin, David J. Deming, Nicola Fuchs-Schündeln and Jonas Jessen | 2026 | Eurostat enterprise survey 2021 to 2025, 32 European countries | Analysis of the official data at country and industry level | Early leaders kept their lead and grew faster; the spread between countries doubled from 2021 to 2025 “with early adoption leaders tending to pull further ahead of early laggards” | [doi.org ↗](https://doi.org/10.20955/wp.2026.003) | Trend ## How fast is AI use growing among European companies? About 2.5 times since 2023: from 8.1% of EU firms in 2023 to 13.5% in 2024 and 20.0% in 2025. The leaders pulled ahead: the gap between the top and bottom country widened from 14 to 37 points. Denmark (+14.5 points) and Finland (+13.5) grew most from 2024 to 2025. Generative AI ## How many European companies use generative AI? 8.8% of EU firms used AI that generates text, speech or code in 2025, from 3.1% in Romania to 30.8% in Denmark (Germany 9.0%, Spain 8.9%, France 7.8%, Portugal 5.3%, Netherlands 16.3%, Finland 20.1%, Sweden 19.7%). A further 9.6% generated pictures, video or audio. ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **The EU wants 75% of companies using AI by 2030.** The Digital Decade target is that three in four companies use cloud computing, big data or AI, any of the three. - **57% of German companies use AI (Bitkom, 2026).** Bitkom surveyed 603 firms with 20 or more employees with one broad question. The official figure for German firms with 10 or more employees is 26% (2025). - **More than 70% of euro area firms use AI (ECB).** That counts any use, including occasional. Only 7% use AI intensively, and the official enterprise survey puts the EU at 20%. - **37% of EU firms use generative AI (EIB).** The EIB weights firms by value added, so big firms dominate. Counted by number of firms, 8.8% of EU enterprises generate text, speech or code with AI. - **Only 8% (or 13.5%) of EU companies use AI.** Those are the 2023 and 2024 figures. In 2025 it was 19.95%. ## More from Europe - [F-35**9–37%**What share of workers in each EU country use generative AI for work?](https://evy.so/facts/genai-use-for-work-by-eu-country/) - [F-36**7–17 pts**How does AI use at work in Europe compare with the US?](https://evy.so/facts/ai-at-work-europe-vs-us/) - [F-37**48–86%**How do European workers feel about AI at work?](https://evy.so/facts/european-attitudes-to-ai-at-work/) - [F-38**28%**How ready are European companies for the EU AI Act?](https://evy.so/facts/eu-ai-act-readiness/) - [F-39**53–66%**How much do European firms care where their AI processes data?](https://evy.so/facts/ai-data-protection-in-europe/) - [F-40**12–14×**How does AI investment in Europe compare with the US?](https://evy.so/facts/ai-investment-europe-vs-us/) [The whole chapter](https://evy.so/facts/europe/) · [Chapter poster](https://evy.so/facts/europe/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "Which occupations and industries use AI the most? (78–85%) | The AI at Work Fact Book" description: "About 78 to 85 percent of US tech workers use AI for work, against about 20 percent in hands-on service jobs. Rankings by occupation and industry in the US and EU." last_updated: "2026-10-24" source: "https://evy.so/facts/ai-use-by-occupation-and-industry/" -------------------------------------------------------------------------------- F-16 · Who Uses AI at Work # Which occupations and industries use AI the most? Tech workers, by a wide margin: about 8 in 10 US workers in computer jobs or the tech industry use AI for work, against about 2 in 10 in hands-on service jobs. The same order shows up in the EU, among EU firms and in AI companies' usage logs. 78–85% of US tech workers use AI for work; hands-on service jobs about 15 to 25% - High confidence - 5 primary sources - 2024 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). Which occupations and industries use AI the most? The AI at Work Fact Book, F-16. https://evy.so/facts/ai-use-by-occupation-and-industry/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/ai-use-by-occupation-and-industry.json) [Markdown](https://evy.so/facts/ai-use-by-occupation-and-industry.md) ## How to read it Each chart row is the top group in one source. The US surveys and Eurostat's people survey give the share of workers in each job or industry who use AI; Eurostat's firm survey counts businesses; the Danish study covers 11 exposed jobs in late 2023. Industry labels differ (Gallup's technology against the official information industry). US rates run higher than EU rates partly because the surveys and time windows differ. Education stands out: many US education workers use AI, but few use it daily (9 to 13%). ## Every study on one scale Shaded: 78–85% (tech) One figure A range in one study Chart values as text - Scale: % of workers in each group using AI for work. Shaded range: 78–85% (tech). - St. Louis Fed survey, computer jobs (2026): 77.6% computer jobs, 82.2% information industry - Gallup, technology industry (2026): 85% (42% daily) - Eurostat, ICT professionals (EU) (2026): 58.3% (manual 5.3%) - Eurostat, ICT firms (EU) (2025): 62.5% of firms (construction 10.8%) - Humlum and Vestergaard, Denmark (2024): 79% of software developers **Why high confidence:** Two US worker surveys, official EU statistics on people and on firms, Danish survey data and Anthropic's logs all put computer and information work on top and manual and in-person service work at the bottom. Small occupation groups have margins of error of 5 to 13 points, so the middle of the ranking is less certain than its ends. ## Sources Sources for F-16 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Generative AI Adoption Tracker: adoption rate for work by occupation and industry (FRED) | Alexander Bick, Adam Blandin and David Deming; Federal Reserve Bank of St. Louis | 2026 | Real-Time Population Survey, May 2026 wave; small groups have large margins of error | Self-reported use of generative AI for one's job, by occupation (22 groups) and industry | Computer and mathematical 77.6%, management 71.9%, business and finance 60.3%, education 56.3%, office support 36.9%, personal care 20.8%, protective service 15.6%. By industry: information 82.2%, professional services 69.5%, public administration 23.4% “Generative Artificial Intelligence, Adoption Rate for Work: Computer and Mathematical Occupations” | [fred.stlouisfed.org ↗](https://fred.stlouisfed.org/series/RPSGENAIUSAGESHAREOCC3) | | Indicator: Artificial Intelligence, AI use trends by industry | Gallup | 2026 | 22,573 US employees, May 2026 | Self-reported frequency of AI use, by 10 industries | Technology 85% use AI (42% daily); finance 71%; professional services 68%; healthcare 48%; manufacturing 47%; retail 40% “As of May 2026, daily AI use varies widely by industry.” | [gallup.com ↗](https://www.gallup.com/699797/indicator-artificial-intelligence.aspx) | | Individuals: use of generative AI tools (dataset isoc\_ai\_iaiu) | Eurostat | 2026 | EU household ICT survey, 27 member states, 2025 | Official statistics; use for work in the previous 3 months, by occupation group | ICT professionals 58.3%, other professionals 20.0%, manual occupations 5.3%; all employed 21.8% “Use of generative AI tools: for professional (work) purposes” | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/databrowser/view/isoc_ai_iaiu/default/table) | | Use of artificial intelligence in enterprises | Eurostat | 2025 | 157,000 EU enterprises with 10 or more employees, early 2025 | Official statistics; share of firms (not workers) using at least one AI technology | Information and communication 62.52% of firms; professional and technical services 40.43%; construction 10.79%; all firms 19.95% “the information and communication sector (with 62.52%) and professional, scientific and technical service activities” | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises) | | The Adoption of ChatGPT (IZA Discussion Paper 16992) | Anders Humlum and Emilie Vestergaard | 2024 | Workers in 11 AI-exposed occupations in Denmark, November 2023 to January 2024 | Survey with Statistics Denmark. Ever used ChatGPT, any purpose; an early wave. | 79% of software developers had used ChatGPT, against 34% of financial advisors “adoption rates ranging from 79% for software developers to 34% for financial advisors” | [iza.org ↗](https://www.iza.org/publications/dp/16992) | Usage logs ## Which jobs do AI conversations come from? Mostly tech, in Claude's logs: tasks tied to computer and mathematical jobs were 35% of Claude.ai conversations in February 2026, against about 4% of US employment. In ChatGPT, 57% of messages from users in computer jobs are about work, against 40% for non-professional jobs. Both measure where usage is concentrated, not how many workers in a job use AI. Which jobs do AI conversations come from? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Anthropic Economic Index: Learning Curves (March 2026) | Anthropic | 2026 | 1 million sampled Claude.ai conversations, February 2026 | Conversations mapped to the occupation a task belongs to. Anthropic sells Claude. | Computer and mathematical tasks 35% of Claude.ai conversations “Computer and Mathematical occupations accounting for 35% of conversations on Claude.ai” | [anthropic.com ↗](https://www.anthropic.com/research/economic-index-march-2026-report) | | How People Use ChatGPT (NBER Working Paper 34255) | Aaron Chatterji, David Deming and colleagues (OpenAI, Duke, Harvard) | 2025 | Consumer ChatGPT users with known occupations, aggregated | Share of each occupation's messages that are about work | 57% of messages are work-related for computer-related jobs, 40% for non-professional jobs “the unadjusted work shares are 57% for computer-related occupations” | [nber.org ↗](https://www.nber.org/papers/w34255) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Programmers are 37% of AI users (Anthropic).** 37.2% was the share of Claude.ai conversations about computer and mathematical tasks in early 2025. It counts conversations on one product, and was 35% in February 2026. - **Microsoft's list of jobs most at risk from AI shows who uses AI.** The study scores how much Copilot conversations touch each job's activities, without knowing users' jobs. Its authors warn against reading it as job loss, and it does not measure who uses AI. - **79% of software developers use ChatGPT at work.** The 79% is ever use of ChatGPT for any purpose among Danish software developers in late 2023. ## More from Who Uses AI at Work - [F-14**40–50%**What share of workers use generative AI at work?](https://evy.so/facts/share-of-workers-using-ai/) - [F-15**40–52%**What do people use AI for at work?](https://evy.so/facts/what-workers-use-ai-for/) - [F-17**12–24 pts**Who uses AI at work more, by age, education and income?](https://evy.so/facts/ai-use-by-age-education-income/) - [F-18**5–16 pts**Is there a gender gap in AI use at work, and how large is it?](https://evy.so/facts/gender-gap-in-ai-use/) - [F-19**50–78%**How many workers use AI tools their employer did not provide, and how many hide it?](https://evy.so/facts/bring-your-own-ai/) - [F-20**42–49%**How many employers provide AI tools, guidance or training to staff?](https://evy.so/facts/employers-providing-ai-tools/) [The whole chapter](https://evy.so/facts/who-uses-ai-at-work/) · [Chapter poster](https://evy.so/facts/who-uses-ai-at-work/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How much faster do people finish writing tasks with AI, and is the work better? (25–40%) | The AI at Work Fact Book" description: "With AI, professionals finished defined writing tasks 25 to 40 percent faster and graders rated the work 18 to 34 percent better. Experiments, the field-trial contrast and misquoted figures." last_updated: "2026-10-24" source: "https://evy.so/facts/ai-writing-speed-and-quality/" -------------------------------------------------------------------------------- F-09 · What AI Saves # How much faster do people finish writing tasks with AI, and is the work better? In controlled experiments, professionals using ChatGPT or GPT-4 finished defined writing and consulting tasks 25 to 40 percent faster, and graders rated the work 18 to 34 percent better. In a six-month field trial of everyday work, time spent writing documents did not change measurably. 25–40% less time on a defined writing or consulting task, with quality up 18 to 34% - Medium confidence - 3 primary sources - 2023 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). How much faster do people finish writing tasks with AI, and is the work better? The AI at Work Fact Book, F-09. https://evy.so/facts/ai-writing-speed-and-quality/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/ai-writing-speed-and-quality.json) [Markdown](https://evy.so/facts/ai-writing-speed-and-quality.md) ## How to read it Both experiments time one well-defined task, chosen to suit the AI, in a single session. Noy and Zhang's task took people without AI about 27 minutes; the BCG tasks were longer, multi-part consulting exercises. Quality was graded on different scales (1 to 7 and 1 to 10). The six-month field trial found no change in time in Word, though its authors say too few people used Word regularly to detect small effects. ## Every study on one scale Shaded: 25–40% One figure A range in one study Chart values as text - Scale: % less time on the task. Shaded range: 25–40%. - Noy and Zhang (2023): 40% less time - Dell'Acqua et al., BCG (2026): 22.5–27.6% less time **Why medium confidence:** Two independent teams with different designs agree on direction and roughly on size. Both time one-off tasks chosen to suit the AI, and the one long-run field trial finds no measurable change in document time. Studies without a figure on this scale are listed in the table only. ## Sources Sources for F-09 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Experimental evidence on the productivity effects of generative artificial intelligence (Science) | Shakked Noy and Whitney Zhang (MIT) | 2023 | 453 college-educated professionals: marketers, grant writers, consultants, data analysts, HR professionals and managers | Preregistered online randomized experiment on a 20 to 30 minute writing task, graded blind by people in the same occupation. Minute figures from the authors' MIT working paper. | 40% less time and 18% higher quality; in the working paper, 27 minutes without ChatGPT and 17 with it “The average time taken decreased by 40% and output quality rose by 18%.” | [science.org ↗](https://www.science.org/doi/10.1126/science.adh2586) | | Navigating the Jagged Technological Frontier (Organization Science) | Fabrizio Dell'Acqua, Ethan Mollick, Karim R. Lakhani and colleagues (Harvard Business School, Wharton, MIT, Warwick, Boston Consulting Group) | 2026 | 758 BCG consultants, about 7% of its individual-contributor consultants | Preregistered randomized experiment on 18 realistic consulting tasks, graded by humans | 25.1% faster on average (22.5% and 27.6% in the two AI groups), 12.2% more tasks done, quality up 29.9% to 33.9% “completing 12.2% more tasks and completing them 25.1% more quickly on average” | [doi.org ↗](https://doi.org/10.1287/orsc.2025.21838) | | Shifting Work Patterns with Generative AI (NBER Working Paper 33795) | Dillon, Jaffe and Immorlica (Microsoft Research) and Stanton (Harvard Business School) | 2025 | 7,137 knowledge workers at 66 large firms | Randomized field experiment over six months, timed with Microsoft 365 usage data | No significant change in time spent writing in Word, and the same number of documents completed “no significant average changes in total time spent in video meetings or writing documents” | [nber.org ↗](https://www.nber.org/papers/w33795) | Quality ## Is writing done with AI better? Graders rated it higher in both experiments: 18% higher in Noy and Zhang's study, graded blind by professionals in the same occupation, and 29.9% to 33.9% higher in the BCG study, graded by humans (about half that when GPT-4 did the grading). In Noy and Zhang's study, 68% of people with ChatGPT handed in its first draft without editing it. When it flips ## What happens on tasks the AI handles badly? The gain can turn into a loss. On a business problem chosen so that GPT-4 would give a plausible wrong answer, BCG consultants without AI were right 84.5% of the time, against 60% and 70.6% in the two AI groups, even though the AI groups spent less time on it. F-13 covers this in more detail. ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **ChatGPT makes workers 40% more productive (Noy and Zhang).** The study found 40% less time and 18% higher quality, on one 20 to 30 minute writing task. A 40% cut in time is a different number from a 40% rise in output. - **AI raised BCG consultants' quality by more than 40%.** That comes from the 2023 working paper. The peer-reviewed 2026 version reports human-graded gains of 29.9% to 33.9%. - **Consultants using AI were 19% less likely to be correct.** The paper's tables give 84.5% correct without AI and 60% and 70.6% with it: a drop of 14 to 25 percentage points, 19 on average, on one task chosen to sit outside the AI's abilities. ## More from What AI Saves - [F-08**3–5.4%**How much time does AI save knowledge workers?](https://evy.so/facts/time-saved-by-ai-at-work/) - [F-10**14–15%**How much more do customer support agents get done with AI?](https://evy.so/facts/ai-customer-support-productivity/) - [F-11**21–26%**How much faster do developers code with AI assistants?](https://evy.so/facts/ai-coding-speed/) - [F-12**27–46%**Does AI help less-experienced workers more than experienced ones?](https://evy.so/facts/ai-helps-less-experienced-workers-more/) - [F-13**14–25%**When does AI make work slower or worse?](https://evy.so/facts/when-ai-makes-work-worse/) [The whole chapter](https://evy.so/facts/what-ai-saves/) · [Chapter poster](https://evy.so/facts/what-ai-saves/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How many workers use AI tools their employer did not provide, and how many hide it? (50–78%) | The AI at Work Fact Book" description: "50 to 78 percent of AI-using workers use tools their employer did not provide, and 48 to 61 percent are reluctant to disclose or have hidden their use. Shadow AI, with sources." last_updated: "2026-10-24" source: "https://evy.so/facts/bring-your-own-ai/" -------------------------------------------------------------------------------- F-19 · Who Uses AI at Work # How many workers use AI tools their employer did not provide, and how many hide it? Half to three quarters of workers who use AI use at least some tools their employer did not provide, and most of them also use approved tools. About half are reluctant to tell their manager or have avoided revealing their AI use. 50–78% of AI-using workers use at least some tools their employer did not provide (2024) - Medium confidence - 3 primary sources - 2024 to 2025 - Last checked October 14, 2026 ## Cite this EVY (2026). How many workers use AI tools their employer did not provide, and how many hide it? The AI at Work Fact Book, F-19. https://evy.so/facts/bring-your-own-ai/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/bring-your-own-ai.json) [Markdown](https://evy.so/facts/bring-your-own-ai.md) ## How to read it Using a tool the employer did not provide is different from breaking a rule: only about 6% of workers report an outright ban, so most outside-tool use is not against an explicit rule. Many people use both: in the KPMG study, 70% use free public tools and 42% use employer tools. Microsoft's 78% is a share of AI users, which is about 58% of all knowledge workers. Microsoft and Software AG sell software that addresses this. Chapter 4 covers the risk side: use against policy and company data in public tools. ## Every study on one scale Shaded: 50–78% One figure Chart values as text - Scale: % of AI users using tools their employer did not provide. Shaded range: 50–78%. - Microsoft and LinkedIn (2024): 78% (US 63%) - KPMG and Melbourne (2025): 70% use free public tools - Software AG (2024): about half **Why medium confidence:** Three independent surveys with different questions land in the same band. All are online panels, two run by software vendors, and they ask different things: whether tools are employer-provided, which routes people use, or whether tools are company-issued. ## Sources Sources for F-19 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | 2024 Work Trend Index Annual Report: AI at Work Is Here. Now Comes the Hard Part | Microsoft and LinkedIn, survey by Edelman Data & Intelligence | 2024 | 31,000 knowledge workers in 31 markets, February to March 2024 | Online survey. Asked whether the AI tools people use at work are provided by their organization. Microsoft sells AI tools for work. | 78% of AI users bring their own AI tools (US 63%, UK 75%, Germany 71%); 80% at small and medium-sized companies “78% of AI users are bringing their own AI tools to work (BYOAI)” | [microsoft.com ↗](https://www.microsoft.com/en-us/worklab/work-trend-index/ai-at-work-is-here-now-comes-the-hard-part) | | Trust, attitudes and use of artificial intelligence: A global study 2025 | University of Melbourne and KPMG International | 2025 | 32,352 employed people in 47 countries, November 2024 to January 2025 | Online survey; AI-using employees picked every way they access AI tools | Free public tools 70%, employer-provided tools 42%, paid public tools 18%; 61% have avoided revealing their AI use | [mbs.edu ↗](https://mbs.edu/-/media/PDF/Research/Trust_in_AI_Report.pdf) | | Half of all employees are Shadow AI users, new study finds | Software AG, survey by TEAM LEWIS | 2024 | 6,000 knowledge workers in the US, UK and Germany, September 2024 | Online panel survey. Shadow AI means tools the company did not issue, including tolerated ones. Software AG sells integration software. | Half of employees use AI tools the company did not issue; 46% would keep using personal tools even if banned “half of all employees are using Shadow AI (i.e. non-company issued AI tools)” | [newscenter.softwareag.com ↗](https://newscenter.softwareag.com/en/news-stories/press-releases/2024/1022-half-of-all-employees-use-shadow-ai.html) [Archived copy ↗](https://web.archive.org/web/20251105024925/https://newscenter.softwareag.com/en/news-stories/press-releases/2024/1022-half-of-all-employees-use-shadow-ai.html) Software AG's newsroom is offline; the archived copy is the original press release. | Hiding it ## How many workers hide their AI use from their employer? About half, depending on the question. 48% of desk workers would be uncomfortable telling their manager they used AI for common tasks (Slack, 2024, asked of users and non-users). 52% of AI users are reluctant to admit using it for their most important tasks (Microsoft, 2024). 61% of AI-using employees have avoided revealing their use at least once, and about 42% do so sometimes or more (KPMG and University of Melbourne). How many workers hide their AI use from their employer? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | The Fall 2024 Workforce Index | Slack Workforce Lab (Salesforce) | 2024 | 17,372 desk workers in 15 countries, August 2024 | Online survey; a hypothetical question asked of everyone, users and non-users | 48% would be uncomfortable admitting to their manager that they used AI for common tasks; top reason: it feels like cheating “of all desk workers said they would be uncomfortable admitting to their manager” | [slack.com ↗](https://slack.com/blog/news/the-fall-2024-workforce-index-shows-executives-and-employees-investing-in-ai-but-uncertainty-holding-back-adoption) | Without employer support ## Do people use AI at work even when their employer offers nothing? Yes. In Denmark, about 40% of workers in exposed jobs had used AI chatbots at work in workplaces that neither encourage it nor provide tools or training, and 7% use them daily. In the US, 6.8% of firms report workers using AI with no firm-level AI use, which is 29% of all firms with any AI use. Do people use AI at work even when their employer offers nothing? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI (NBER Working Paper 33777) | Anders Humlum and Emilie Vestergaard, with Statistics Denmark | 2026 | About 25,000 workers in 11 AI-exposed occupations, late 2024 | Survey linked to Danish government records | About 40% have used AI chatbots at work where employers neither encourage nor equip them; 6% report a ban “Even in workplaces that neither encourage use nor provide enterprise tools or training” | [nber.org ↗](https://www.nber.org/papers/w33777) | | The Microstructure of AI Diffusion (CES Working Paper 26-25) | US Census Bureau Center for Economic Studies | 2026 | Business Trends and Outlook Survey AI supplement, November 2025 to January 2026 | Firm-reported, one respondent per business | 6.8% of US firms report worker AI use with no firm-level AI use “worker task use sometimes occurs without formal firm-level adoption” | [www2.census.gov ↗](https://www2.census.gov/library/working-papers/2026/adrm/ces/CES-WP-26-25.pdf) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **78% of workers bring their own AI to work.** The 78% is of knowledge workers who use AI, about 58% of all knowledge workers, and 63% of AI users in the US. It counts tools the employer did not provide, which is different from banned tools. - **Half of all employees use unauthorized AI (Software AG).** The survey defines shadow AI as tools the company did not issue, which includes tools it allows, among knowledge workers in three countries. - **48% of workers hide their AI use from their boss (Slack).** Slack asked whether people would be uncomfortable admitting AI use. It is reluctance to tell, asked of non-users too, and did not ask about actual hiding. - **57% of employees hide their AI use (KPMG).** It combines two behaviors and counts anyone who did either even rarely. Doing it sometimes or more is 39 to 42%. ## More from Who Uses AI at Work - [F-14**40–50%**What share of workers use generative AI at work?](https://evy.so/facts/share-of-workers-using-ai/) - [F-15**40–52%**What do people use AI for at work?](https://evy.so/facts/what-workers-use-ai-for/) - [F-16**78–85%**Which occupations and industries use AI the most?](https://evy.so/facts/ai-use-by-occupation-and-industry/) - [F-17**12–24 pts**Who uses AI at work more, by age, education and income?](https://evy.so/facts/ai-use-by-age-education-income/) - [F-18**5–16 pts**Is there a gender gap in AI use at work, and how large is it?](https://evy.so/facts/gender-gap-in-ai-use/) - [F-20**42–49%**How many employers provide AI tools, guidance or training to staff?](https://evy.so/facts/employers-providing-ai-tools/) [The whole chapter](https://evy.so/facts/who-uses-ai-at-work/) · [Chapter poster](https://evy.so/facts/who-uses-ai-at-work/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How often do people check AI output before using it? (41–44%) | The AI at Work Fact Book" description: "About 41 to 44 percent of AI users use the output without routine checking, and 56 percent have made a mistake at work because of AI. Self-reports and observed behavior." last_updated: "2026-10-24" source: "https://evy.so/facts/checking-ai-output/" -------------------------------------------------------------------------------- F-22 · What Teams Trust # How often do people check AI output before using it? About four in ten AI users regularly use AI output without checking it, by their own account: 44 percent of AI-using employees across 47 countries do not verify its accuracy most of the time, and knowledge workers report no critical thinking on 41 percent of their AI-assisted tasks. 56 percent of AI-using employees say AI has caused a mistake in their work at least once. 41–44% of AI users, or of their AI tasks, go without routine checking (self-reported) - Medium confidence - 4 primary sources - 2023 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). How often do people check AI output before using it? The AI at Work Fact Book, F-22. https://evy.so/facts/checking-ai-output/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/checking-ai-output.json) [Markdown](https://evy.so/facts/checking-ai-output.md) ## How to read it Most of this is self-reported, and people likely flatter themselves. KPMG counts people, Lee and colleagues count tasks. Checking depends on how easy it is: developers can run code, so only about 10% use AI answers as-is, while KPMG finds that people who trust AI more check less. The only observed studies are a 2023 writing experiment and a small 2025 trial of 16 developers, who accepted under 44% of AI code suggestions. ## Every study on one scale Shaded: 41–44% One figure A range in one study Chart values as text - Scale: % used without routine checking. Shaded range: 41–44%. - KPMG and Melbourne (2025): 42–44% of AI users - Lee et al., Microsoft and CMU (2025): 40.7% of tasks - Stack Overflow 2026 (2026): 9.6% use as-is - Noy and Zhang, experiment (2023): 68% unedited (experiment) **Why medium confidence:** Two independent creators with very different methods land on almost the same figure, one counting people and one counting tasks. Both are self-reports, which likely understate careless use, and the observed studies are small or older. ## Sources Sources for F-22 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Trust, attitudes and use of artificial intelligence: A global study 2025 | University of Melbourne and KPMG International | 2025 | 32,352 employed people in 47 countries, November 2024 to January 2025 | Online survey of employees who use AI at work, self-reported frequency | 56% verify AI accuracy most of the time or always, so 44% do not; 42% rely on output without evaluating it sometimes or more often (66% ever) “Two in three report relying on AI output without evaluating the information it provides” | [mbs.edu ↗](https://mbs.edu/-/media/PDF/Research/Trust_in_AI_Report.pdf) | | The Impact of Generative AI on Critical Thinking (CHI 2025) | Hao-Ping (Hank) Lee and colleagues, Microsoft Research and Carnegie Mellon University | 2025 | 319 knowledge workers using AI at least weekly, describing 936 real tasks | Online survey; for each task, whether the worker did any critical thinking such as verifying or editing | Critical thinking reported on 59.3% of tasks, so none on 40.7%; higher confidence in AI predicted less critical thinking “Our participants self-reported to have enacted critical thinking for 555 (59.29%) of the examples” | [microsoft.com ↗](https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/lee_2025_ai_critical_thinking_survey.pdf) | | 2026 Developer Survey: AI data | Stack Overflow | 2026 | 13,162 developers answering this question, June to August 2026 | Opt-in survey; what developers usually do with an AI answer, several answers allowed | 9.6% use AI answers as-is; 76.5% run the code, 63.9% compare it with the codebase “Only 10% use answers as-is.” | [survey.stackoverflow.co ↗](https://survey.stackoverflow.co/2026/ai/data) | | Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence | Shakked Noy and Whitney Zhang (MIT) | 2023 | 444 college-educated professionals in an online writing experiment, early 2023 | Self-report plus observed activity on the platform; not participants' real jobs | 68% said they submitted ChatGPT's first output unedited, and were active for about 3 minutes after pasting it in “68% of treated participants report submitting ChatGPT’s initial output without editing it” | [economics.mit.edu ↗](https://economics.mit.edu/sites/default/files/inline-files/Noy_Zhang_1.pdf) | Mistakes Low confidence ## How many workers have made mistakes because of AI? Between about a third and a half of AI users. 56% of AI-using employees across 47 countries say they have made a mistake in their work because of AI, and 31% say it happens sometimes or more often. In a small 2026 US poll, 30% said a wrong AI answer had caused a problem in their work. Noticing errors is more common still: 66% of developers report AI answers that are almost right, but not quite. How many workers have made mistakes because of AI? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Half of U.S. Workers Know They Would Be Personally Liable for a Wrong AI Answer | Kolmogorov Law, poll on Pollfish | 2026 | 500 employed US adults who use AI for work, September 2026, unweighted | Online poll commissioned by a law firm; treat as indicative | 30% say a wrong AI answer has caused a problem in their work; 65% do not always verify AI answers “a wrong or inaccurate AI answer has caused a problem in their work” | [prnewswire.com ↗](https://www.prnewswire.com/news-releases/half-of-us-workers-know-they-would-be-personally-liable-for-a-wrong-ai-answer-two-thirds-still-dont-always-check-kolmogorov-law-survey-finds-302867898.html) | | 2025 Developer Survey: AI | Stack Overflow | 2025 | 25,332 developers answering the frustrations question | Opt-in online survey, several answers allowed | 66% cite AI answers that are almost right, but not quite; 45.2% say debugging AI code takes more time “AI solutions that are almost right, but not quite” | [survey.stackoverflow.co ↗](https://survey.stackoverflow.co/2025/ai) | Reaching customers ## Do AI mistakes reach customers? Sometimes. A quarter of IT leaders (25%) say AI mistakes have already hurt customers or clients, 23% that they cost the company money, and 16% that they seriously disrupted operations. Of employees who review colleagues' AI work, 79% regularly receive output that is low quality or contains errors. Do AI mistakes reach customers? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | The Pulse of Work in 2026 | GoTo with Workplace Intelligence | 2026 | 1,250 knowledge workers and 1,250 IT decision-makers in 10 countries, November 2025 to January 2026 | Online survey. GoTo sells IT and communications software. | 25% of IT leaders say AI mistakes hurt customers; 79% of employees who review colleagues' AI work regularly get flawed output “around one-quarter of IT leaders say AI has made mistakes that affected customers” | [edge.sitecorecloud.io ↗](https://edge.sitecorecloud.io/gototechnol00e8-mktglobalxm07e7-mktglobalpr9498-b870/media/goto/pdfs/ebook/ebook_pulse-of-work-2026.pdf) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Two thirds of employees use AI output without checking it (KPMG).** 66% have done so at least once, including rarely. Doing it sometimes or more often is 42%, and 56% verify accuracy most of the time. - **57% of workers have made mistakes because of AI (KPMG).** The global report says 56%, counting anyone who did so even once. Sometimes or more often is 31%. - **AI is destroying critical thinking (Microsoft and Carnegie Mellon).** The study measured 319 workers' own sense of effort and whether they thought critically. It did not test thinking skills or show a decline over time. - **77% of employees review AI work as carefully as human work (Workday).** The 77% covers only employees who use AI every day, at companies with more than $100 million in revenue, in a vendor survey. - **66% of developers spend more time fixing AI code (Stack Overflow).** 66% cite answers that are almost right; 45.2% say debugging AI code takes more time. ## More from What Teams Trust - [F-21**24–53%**How much do workers trust the output of AI tools at work?](https://evy.so/facts/trust-in-ai-output/) - [F-23**17–45%**How often is AI output wrong on real work tasks?](https://evy.so/facts/ai-error-rates-at-work/) - [F-24**2–12%**Which work tasks do people want AI to do on its own, and which do they want to keep?](https://evy.so/facts/tasks-workers-want-ai-to-do/) - [F-25**30–44%**How many workers use AI in ways that break company policy?](https://evy.so/facts/ai-use-against-company-policy/) - [F-26**42–48%**How often do employees put company data into public AI tools?](https://evy.so/facts/company-data-in-public-ai/) - [F-27**21–29%**How many organizations have had a security incident or data breach involving AI?](https://evy.so/facts/ai-security-breaches/) [The whole chapter](https://evy.so/facts/what-teams-trust/) · [Chapter poster](https://evy.so/facts/what-teams-trust/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How often do employees put company data into public AI tools? (42–48%) | The AI at Work Fact Book" description: "42 to 48 percent of workers say they have put company data into public AI tools. What vendor monitoring measures, and how much AI use runs through personal accounts." last_updated: "2026-10-24" source: "https://evy.so/facts/company-data-in-public-ai/" -------------------------------------------------------------------------------- F-26 · What Teams Trust # How often do employees put company data into public AI tools? A little under half of workers say they have put company information into public AI tools: 48 percent of AI-using employees, and 42 percent of privacy and security professionals themselves. Monitoring software confirms it happens routinely, and a third to a half of AI use at work runs through personal accounts. 42–48% of workers say they have put company information into public AI tools (2024 to 2025) - Medium confidence - 2 primary sources - 2025 - Last checked October 14, 2026 ## Cite this EVY (2026). How often do employees put company data into public AI tools? The AI at Work Fact Book, F-26. https://evy.so/facts/company-data-in-public-ai/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/company-data-in-public-ai.json) [Markdown](https://evy.so/facts/company-data-in-public-ai.md) ## How to read it The two surveys ask people about their own behavior; one covers employees in all jobs, the other privacy and security professionals. Monitoring data from security vendors (below) counts events in their customers' systems and cannot be put on the same scale: one counts interactions, another prompts or files, another users. Every monitoring source sells the product that produced its data. Uploading is most common where AI is banned (67%), which suggests bans push use out of sight. ## Every study on one scale Shaded: 42–48% One figure A range in one study Chart values as text - Scale: % who have put company data into AI tools. Shaded range: 42–48%. - KPMG and Melbourne (employees) (2025): 48% ever, 34% regularly - Cisco (privacy professionals) (2025): 42% (48% in 2024) **Why medium confidence:** Two independent self-report surveys with different populations agree closely. Monitoring data from four security vendors confirms the behavior, but each counts something different and draws on its own customers. ## Sources Sources for F-26 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Trust, attitudes and use of artificial intelligence: A global study 2025 | University of Melbourne and KPMG International | 2025 | 32,352 employed people in 47 countries, November 2024 to January 2025 | Anonymous online survey of employees who use AI; self-reported | 48% have uploaded company information such as financial, sales or customer data into a public AI tool (34% sometimes or more often) “report that they have uploaded company information, such as financial, sales, or customer information” | [mbs.edu ↗](https://mbs.edu/-/media/PDF/Research/Trust_in_AI_Report.pdf) | | Cisco 2025 Data Privacy Benchmark Study | Cisco | 2025 | More than 2,600 privacy and security professionals in 12 countries, fall 2024 | Double-blind survey of what respondents themselves entered into generative AI apps. Cisco sells security products. | 42% entered non-public information about their company (48% in the 2024 edition); 46% employee information; 31% customer information “nearly half of respondents still reported inputting personal employee information or non-public” | [cisco.com ↗](https://www.cisco.com/c/dam/en_us/about/doing_business/trust-center/docs/cisco-privacy-benchmark-study-2025.pdf) [Archived copy ↗](https://web.archive.org/web/2025id_/https://www.cisco.com/c/dam/en_us/about/doing_business/trust-center/docs/cisco-privacy-benchmark-study-2025.pdf) Cisco blocks automated downloads; the archived copy is the same report. | Measured by monitoring ## What does monitoring software show about company data in AI tools? That sensitive data reaches AI tools routinely, at rates that depend on what is counted. 39.7% of employee interactions with AI tools involved data the customer labeled sensitive (Cyberhaven, 2025 data). About 2.6% of all prompts and uploads were sensitive, but 26% of uploaded files were (Harmonic Security, 2025). The average organization logged 223 attempts a month to send sensitive data to AI apps, made by about 3% of its AI users (Netskope). All three are vendor data from their own customers. What does monitoring software show about company data in AI tools? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | 2026 AI Adoption & Risk Report | Cyberhaven Labs | 2026 | Hundreds of thousands of employees at 222 companies, AI usage during 2025 | Vendor telemetry of data moving into AI tools. Cyberhaven sells data security software. | 39.7% of employee AI interactions involve sensitive data; personal-account share of use: ChatGPT 32.3%, Claude 58.2%, Perplexity 60.9% “the average employee enters sensitive data into AI tools once every three days” | [info.cyberhaven.com ↗](https://info.cyberhaven.com/hubfs/Webflow_Resources/Cyberhaven-AI-Risk-Report-2026.pdf) | | What 22 Million Enterprise AI Prompts Reveal About Shadow AI in 2025 | Harmonic Security | 2026 | 22.5 million prompts and uploads across 665 AI tools, 2025, US and UK customers | Vendor telemetry from browser monitoring. Harmonic sells AI data protection software. | 579,113 sensitive exposures in 22.5 million prompts and uploads (about 2.6%); 26% of uploaded files in Q3 2025 “Sensitive Data Instances: 579,113 detected exposures” | [harmonic.security ↗](https://www.harmonic.security/resources/what-22-million-enterprise-ai-prompts-reveal-about-shadow-ai-in-2025) | | Cloud and Threat Report: 2026 | Netskope Threat Labs | 2026 | Anonymized data from Netskope customers, October 2024 to October 2025 | Vendor telemetry of data policy violations, only where customers have rules in place. Netskope sells security software. | About 3% of AI users commit an average of 223 data policy violations a month per organization; 47% of AI users use personal AI apps “3% of genAI users committing an average of 223 genAI data policy violations” | [netskope.com ↗](https://www.netskope.com/resources/cloud-and-threat-reports/cloud-and-threat-report-2026) | Personal accounts ## How much AI use at work runs through personal accounts? A third to a half, depending on the tool and the vendor. 47% of AI users used personal AI apps in 2025 and 44% in mid-2026 (Netskope). 32.3% of ChatGPT use and about 60% of Claude and Perplexity use at work ran through personal accounts (Cyberhaven). Data security leaders estimate that 58% of employees use personal credentials for work AI (Microsoft, 2026). ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **48% of employees enter non-public company information into AI (Cisco).** Cisco's respondents were privacy and security professionals reporting their own behavior, in the 2024 edition. The 2025 figure is 42%. - **77% of employees paste company data into ChatGPT (LayerX).** LayerX gives no denominator, so the figure can only describe AI users, and it counts any paste into any AI tool. - **40% of employees leak sensitive data into AI (Cyberhaven).** 39.7% is the share of AI interactions that involve data the customer labeled sensitive. It is not a share of employees. - **Companies see 223 AI data leaks a month (Netskope).** 223 is the average number of detected policy violations per organization, made by about 3% of AI users, and only where rules exist. ## More from What Teams Trust - [F-21**24–53%**How much do workers trust the output of AI tools at work?](https://evy.so/facts/trust-in-ai-output/) - [F-22**41–44%**How often do people check AI output before using it?](https://evy.so/facts/checking-ai-output/) - [F-23**17–45%**How often is AI output wrong on real work tasks?](https://evy.so/facts/ai-error-rates-at-work/) - [F-24**2–12%**Which work tasks do people want AI to do on its own, and which do they want to keep?](https://evy.so/facts/tasks-workers-want-ai-to-do/) - [F-25**30–44%**How many workers use AI in ways that break company policy?](https://evy.so/facts/ai-use-against-company-policy/) - [F-27**21–29%**How many organizations have had a security incident or data breach involving AI?](https://evy.so/facts/ai-security-breaches/) [The whole chapter](https://evy.so/facts/what-teams-trust/) · [Chapter poster](https://evy.so/facts/what-teams-trust/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How many employers provide AI tools, guidance or training to staff? (42–49%) | The AI at Work Fact Book" description: "42 to 49 percent of workers say their employer provides AI tools, but only 25 to 40 percent report a clear AI plan or policy. Firm counts for the EU and US, and training." last_updated: "2026-10-24" source: "https://evy.so/facts/employers-providing-ai-tools/" -------------------------------------------------------------------------------- F-20 · Who Uses AI at Work # How many employers provide AI tools, guidance or training to staff? About half of workers say their employer has brought in AI tools, 42 to 49 percent in 2024 to 2026, but only 25 to 40 percent report a clear AI plan or policy. Counted by firms instead of workers, adoption is lower: 20 percent of EU firms and 24 percent of US businesses use AI, because most firms are small. 42–49% of workers say their employer has brought in AI tools; 25 to 40% report a clear plan or policy - Medium confidence - 4 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). How many employers provide AI tools, guidance or training to staff? The AI at Work Fact Book, F-20. https://evy.so/facts/employers-providing-ai-tools/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/employers-providing-ai-tools.json) [Markdown](https://evy.so/facts/employers-providing-ai-tools.md) ## How to read it Worker-reported figures count employees, so large employers weigh more, and they depend on what workers know. Firm-reported figures count businesses, most of which are small, so they come out much lower. The consistent finding is the gap between tools and guidance: in the US, 47% of employees report AI tools but 25% a clear plan. Training figures come from different questions, so they are listed separately below. ## Every study on one scale Shaded: 42–49% tools One figure A range in one study Chart values as text - Scale: % of workers reporting each. Shaded range: 42–49% tools. - Gallup, AI tools (2026): 47% tools - Gallup, clear plan (2026): 25–26% clear plan - KPMG and Melbourne (2025): 42% tools, 34% policy - Gallup, guidelines (2025): 30% guidelines or policy **Why medium confidence:** Two independent creators with different methods agree in the 40s, but one counts all employees and the other AI users, and about one in five US employees do not know what their employer has done. Firm counts from two official statistics agencies are reliable on their own terms and are shown separately. ## Sources Sources for F-20 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Organizational AI Adoption Jumps Six Points (Q2 2026 workforce study) | Gallup | 2026 | 22,573 US employees, 6 to 20 May 2026 | Probability-based Gallup Panel; worker-reported, with a don't-know option since late 2025 | 47% say their organization has integrated AI tools (38% in Q4 2025); about 20% don't know; 25% say a clear plan for AI was communicated “Forty-seven percent of U.S. employees now say their organization has integrated AI tools” | [gallup.com ↗](https://www.gallup.com/workplace/712736/organizational-adoption-jumps-six-points.aspx) | | Indicator: Artificial Intelligence | Gallup | 2026 | US employees, quarterly Gallup Panel waves | Worker-reported; the August 2026 value is read from the indicator's chart data | Clear plan for integrating AI communicated: 15% (2024), 22% (2025), 25% (May 2026), 26% (August 2026); AI tools integrated 49% (August 2026) | [gallup.com ↗](https://www.gallup.com/699797/indicator-artificial-intelligence.aspx) | | Trust, attitudes and use of artificial intelligence: A global study 2025 | University of Melbourne and KPMG International | 2025 | 32,352 employed people in 47 countries, November 2024 to January 2025 | Online survey; worker-reported | 42% of AI-using employees use employer-provided tools; 34% have a policy guiding generative AI use, 6% a ban, 41% none, 19% don't know “only two in five say there is a policy guiding its use” | [mbs.edu ↗](https://mbs.edu/-/media/PDF/Research/Trust_in_AI_Report.pdf) | | AI Use at Work Has Nearly Doubled in Two Years | Gallup | 2025 | US employees, Gallup Panel, May 2025 | Worker-reported, before the don't-know option was added | 30% say their organization has general guidelines or formal policies for AI use (20% in 2024) “30% of employees say their organization has either general guidelines or formal policies” | [gallup.com ↗](https://www.gallup.com/workplace/691643/work-nearly-doubled-two-years.aspx) | Counted by firms ## What share of companies use AI? About one in five, counted by firms. 19.95% of EU enterprises with 10 or more employees used AI in 2025, rising to 55% of large ones, and 23.8% of US businesses used AI in the two weeks to 20 September 2026, rising to 37% of firms with 250 or more staff. Weighted by employment the picture changes: 78% of the US labor force works at firms that have adopted AI. What share of companies use AI? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Use of artificial intelligence in enterprises | Eurostat | 2025 | 157,000 EU enterprises with 10 or more employees, 2025 | Official statistics; firm-reported use of any of eight AI technologies | 19.95% of EU enterprises (13.48% in 2024); small 17.0%, medium 30.4%, large 55.0%; Denmark 42.0%, Romania 5.2% “In 2025, 19.95% of EU enterprises used AI technologies.” | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises) | | Business Trends and Outlook Survey (BTOS), national estimates | US Census Bureau | 2026 | About 20,000 responses per two-week period from US businesses | Official experimental survey; firm-reported. The question changed in late 2025, so older figures are not comparable. | 23.8% of US businesses used AI in any business function (period ending 20 September 2026); 37% of firms with 250 or more employees (spring 2026) “did this business use Artificial Intelligence (AI) in any of its business functions?” | [census.gov ↗](https://www.census.gov/hfp/btos/data) | | Monitoring AI Adoption in the US Economy (FEDS Notes) | Jeffrey S. Allen, Federal Reserve Board | 2026 | Comparison of Census, St. Louis Fed and Atlanta Fed surveys | Research note comparing firm-weighted and employment-weighted adoption | About 18% of US firms used AI at the end of 2025, while 78% of the labor force works at firms that have adopted AI “78 percent of the labor force works at firms that have adopted AI” | [federalreserve.gov ↗](https://www.federalreserve.gov/econres/notes/feds-notes/monitoring-ai-adoption-in-the-u-s-economy-20260403.html) | Training Low confidence ## How many employers train staff on AI? A minority, though the questions differ. In 2024, 39% of AI users had received AI training from their company (Microsoft). In Denmark, 39% of employers that encourage AI chatbots provide training. About 15% of AI-using US firms report training staff (Census, late 2025). Where employers combine encouragement, tools and training, 93% of Danish workers in exposed jobs use AI, against about 41% where employers do nothing. How many employers train staff on AI? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | 2024 Work Trend Index Annual Report | Microsoft and LinkedIn | 2024 | 31,000 knowledge workers in 31 markets | Online survey. Microsoft sells AI tools and training. | 39% of people who use AI at work have had AI training from their company “Only 39% of people globally who use AI at work have gotten AI training” | [microsoft.com ↗](https://www.microsoft.com/en-us/worklab/work-trend-index/ai-at-work-is-here-now-comes-the-hard-part) | | Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI (NBER Working Paper 33777) | Anders Humlum and Emilie Vestergaard, with Statistics Denmark | 2026 | About 25,000 workers at about 7,000 workplaces, late 2024 | Worker-reported employer policies, linked to government records | 43% of employers encourage AI chatbot use; of those, 61% provide an enterprise chatbot and 39% training “about 43% explicitly encourage their use, another 21% allow it” | [nber.org ↗](https://www.nber.org/papers/w33777) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Only 5% to 10% of US businesses use AI (Census).** That was the old question, about AI used to produce goods or services. With the current wording it is 17% to 24% (late 2025 to September 2026). - **78% of companies use AI.** 78% is the share of the US labor force working at firms that have adopted AI, which large firms dominate. Counted by firms it is 18% to 24%. - **Gallup: 44% of organizations have integrated AI.** It is the share of employees who say so, from May 2025, before Gallup added a don't-know option. The comparable figure for May 2026 is 47%. - **One in five EU companies use generative AI.** Eurostat's 19.95% counts any of eight AI technologies, including older ones. Text or speech generation alone was 8.76% of enterprises in 2025. - **Only 25% of companies offer AI training (Microsoft).** Microsoft's 2024 report said 25% of companies were planning to offer generative AI training that year. It was a plan, and is now dated. ## More from Who Uses AI at Work - [F-14**40–50%**What share of workers use generative AI at work?](https://evy.so/facts/share-of-workers-using-ai/) - [F-15**40–52%**What do people use AI for at work?](https://evy.so/facts/what-workers-use-ai-for/) - [F-16**78–85%**Which occupations and industries use AI the most?](https://evy.so/facts/ai-use-by-occupation-and-industry/) - [F-17**12–24 pts**Who uses AI at work more, by age, education and income?](https://evy.so/facts/ai-use-by-age-education-income/) - [F-18**5–16 pts**Is there a gender gap in AI use at work, and how large is it?](https://evy.so/facts/gender-gap-in-ai-use/) - [F-19**50–78%**How many workers use AI tools their employer did not provide, and how many hide it?](https://evy.so/facts/bring-your-own-ai/) [The whole chapter](https://evy.so/facts/who-uses-ai-at-work/) · [Chapter poster](https://evy.so/facts/who-uses-ai-at-work/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How ready are European companies for the EU AI Act? (28%) | The AI at Work Fact Book" description: "28 percent of companies say they understand the EU AI Act's requirements; 71 percent of German firms feel poorly informed. Compliance steps, the deadlines in force since July 2026, and AI training." last_updated: "2026-10-24" source: "https://evy.so/facts/eu-ai-act-readiness/" -------------------------------------------------------------------------------- F-38 · Europe # How ready are European companies for the EU AI Act? About 3 in 10 companies say they understand what the AI Act requires: 28 percent in two independent 2026 surveys, and 71 percent of German firms feel poorly informed. Engagement is rising fast, and the deadlines for high-risk AI moved in July 2026 to December 2027 and August 2028. 28% of companies say they understand what the AI Act requires (two 2026 surveys) - Medium confidence - 3 primary sources - 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). How ready are European companies for the EU AI Act? The AI at Work Fact Book, F-38. https://evy.so/facts/eu-ai-act-readiness/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/eu-ai-act-readiness.json) [Markdown](https://evy.so/facts/eu-ai-act-readiness.md) ## How to read it Feeling informed is different from being compliant, and all of this is self-reported. The German surveys cover firms with 20 or more employees; the four-country survey asks IT professionals and includes the UK, where the Act applies only to firms selling into the EU. Most firms are users of AI, not providers (88% of German AI users), so their main duties are AI literacy, transparency and, for high-risk uses, deployer duties. ## Every study on one scale Shaded: 28% One figure Chart values as text - Scale: % saying they understand the AI Act well. Shaded range: 28%. - BSI and TÜV, Germany (2026): 28% well informed - ESSCA and Forvis Mazars (2026): 28% understand it well **Why medium confidence:** Two independent surveys with different populations and methods land on the same 28%, and a third confirms the direction. No EU-wide official survey exists, so the evidence leans on Germany, is self-reported, and two of the creators sell compliance services. Studies without a figure on this scale are listed in the table only. ## Sources Sources for F-38 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | KI-Sicherheitsstudie 2026 | Federal Office for Information Security (BSI) and TÜV-Verband, survey by forsa | 2026 | 505 companies with 20 or more employees in Germany (246 using AI), June to July 2026 | Representative survey for a federal authority and the testing association, whose members sell AI audits | 28% well informed about the Act's requirements, 71% poorly informed; of AI-using firms, 65% trained staff, 37% classified systems by risk, 23% took none of six steps “darüber fühlt sich die überwiegende Mehrheit der Befragten schlecht informiert (71 Prozent)” | [tuev-verband.de ↗](https://www.tuev-verband.de/fileadmin/user_upload/Content_local/Studien_local/2026_BSI-TUEV-KI-Sicherheitsstudie.pdf) | | AI, Business and Sustainability Barometer 2026 | ESSCA School of Management with Forvis Mazars | 2026 | 826 IT and digital professionals in France, Germany, Spain and the UK, March 2026 | Survey by a business school and an audit firm that sells compliance advice; summary page only | 28% say their organization understands the AI Act well; 57% provide AI training for employees “believe their organisation has a good understanding of the EU AI Act” | [essca.eu ↗](https://www.essca.eu/en/news/2026-ai-barometer-businesses-accelerate-ai-adoption-governance-keep-pace/) | | Erstmals nutzt die Mehrheit der Unternehmen KI (press release and chart deck, 14 September 2026) | Bitkom | 2026 | 603 companies with 20 or more employees in Germany, July to August 2026 | Representative telephone survey by the German digital industry association | 37% expect to be affected as AI users (23% in 2025), 34% are still checking, 8% have not looked at the Act; 89% of those affected expect high effort “8 Prozent haben sich mit dem AI Act noch nicht beschäftigt” | [bitkom.org ↗](https://www.bitkom.org/Presse/Presseinformation/Erstmals-nutzt-Mehrheit-Unternehmen-KI) | Deadlines ## When do the EU AI Act's rules apply? The Act entered into force on 1 August 2024. Bans on prohibited practices and the AI literacy duty apply from 2 February 2025, rules for general-purpose AI models from 2 August 2025, and the general application date, including transparency duties, is 2 August 2026. The Digital Omnibus on AI (Regulation (EU) 2026/1744, in force since 27 July 2026) moved the high-risk rules to 2 December 2027 for uses such as hiring, credit and education, and to 2 August 2028 for AI built into regulated products. It also softened the AI literacy duty to taking measures to support it, and added bans on non-consensual intimate deepfakes and child abuse material from 2 December 2026. The separate data and GDPR half of the Digital Omnibus, COM(2025) 837, is still only a proposal. When do the EU AI Act's rules apply? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Regulation (EU) 2024/1689 (Artificial Intelligence Act), Article 113 | European Parliament and Council of the European Union | 2024 | Legal text | Official text on EUR-Lex | In force 1 August 2024; bans and AI literacy from 2 February 2025; general-purpose AI rules from 2 August 2025; general application from 2 August 2026 “It shall apply from 2 August 2026.” | [eur-lex.europa.eu ↗](https://eur-lex.europa.eu/legal-content/EN/ALL/?uri=CELEX:32024R1689) | | Regulation (EU) 2026/1744 (Digital Omnibus on AI) | European Parliament and Council of the European Union | 2026 | Legal text, signed 8 July 2026 | Official text on EUR-Lex; amendments in effect from 27 July 2026 | High-risk rules for Annex III uses from 2 December 2027 and for regulated products from 2 August 2028; new bans from 2 December 2026 “2 December 2027 as regards AI systems classified as high-risk pursuant to Article 6(2)” | [eur-lex.europa.eu ↗](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32026R1744) | | Procedure file 2025/0360(COD): Digital Omnibus on data, GDPR, ePrivacy and cybersecurity | European Parliament, Legislative Observatory | 2026 | Procedure record, checked 14 October 2026 | Legislative Observatory page | Proposal COM(2025) 837, awaiting committee decision; not adopted “Awaiting committee decision” | [oeil.secure.europarl.europa.eu ↗](https://oeil.secure.europarl.europa.eu/oeil/en/procedure-file?reference=2025/0360(COD)) | AI training Low confidence ## How many European workers get AI training? Few, though firms are catching up. In 2024, 15% of employees in 11 EU countries took part in any AI training in the previous year (Germany 14%), while 42% said they needed to build AI skills (Cedefop). In 2026, 70% of German firms with 20 or more employees train at least some staff on AI, but only 11% train all of them (Bitkom); asked whether they had already trained staff, 27% said yes (TÜV-Verband). ICT specialists, who build and run AI systems, are 5.0% of EU employment (Germany 5.5%). How many European workers get AI training? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Skills empower workers in the AI revolution: first results of Cedefop's 2024 AI skills survey | Cedefop (European Centre for the Development of Vocational Training) | 2025 | 5,342 employees aged 16 to 64 in 11 EU countries, February to May 2024 | Probability-based online panels, weighted | 15% took part in AI training in the past 12 months (Germany 14%); 42% report an AI skill gap “a relatively small share (15%) of workers participated in education or training” | [cedefop.europa.eu ↗](https://www.cedefop.europa.eu/files/9201_en.pdf) | | TÜV Weiterbildungsstudie 2026 | TÜV-Verband, survey by forsa | 2026 | 500 companies with 20 or more employees in Germany | Representative survey for the testing association, whose members sell training | 27% of firms have already trained staff on AI (12% in 2024); 49% of firms with 250 or more employees “27 Prozent der Unternehmen haben ihre Beschäftigten bereits im Umgang mit KI geschult” | [tuev-verband.de ↗](https://www.tuev-verband.de/pressemitteilungen/ki-schulungen-in-unternehmen-innerhalb-von-zwei-jahren-verdoppelt) | | Employed ICT specialists (dataset isoc\_sks\_itspt) | Eurostat (EU Labour Force Survey) | 2026 | EU Labour Force Survey, 2025 | Official statistics | ICT specialists 5.0% of EU employment (10.45 million people); Germany 5.5% (2.32 million) “Employed information and communications technology (ICT) specialists” | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/databrowser/view/isoc_sks_itspt/default/table) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **High-risk AI Act rules apply from 2 August 2026.** Regulation (EU) 2026/1744 moved them to 2 December 2027 (uses such as hiring) and 2 August 2028 (regulated products). 2 August 2026 remains the general date, including transparency duties. - **The AI Act requires every company to ensure all staff are AI literate.** Since 27 July 2026, the duty is to take measures to support AI literacy, and the text says it does not require guaranteeing any level of literacy. - **The Digital Omnibus has been adopted.** Only the AI part, Regulation (EU) 2026/1744. The data and GDPR part, COM(2025) 837, is still awaiting a committee decision in Parliament. - **56% of companies are completely unprepared for the AI Act (attributed to TÜV).** We found no primary source for it. The BSI and TÜV study says 71% of German firms feel poorly informed and 23% of AI users have taken none of six compliance steps. - **70% of German companies train their staff on AI (Bitkom).** It includes firms training only selected employees; 11% train everyone, and TÜV-Verband finds 27% have actually trained staff. ## More from Europe - [F-34**5–42%**Which EU countries' companies use AI the most and least?](https://evy.so/facts/ai-use-by-eu-country/) - [F-35**9–37%**What share of workers in each EU country use generative AI for work?](https://evy.so/facts/genai-use-for-work-by-eu-country/) - [F-36**7–17 pts**How does AI use at work in Europe compare with the US?](https://evy.so/facts/ai-at-work-europe-vs-us/) - [F-37**48–86%**How do European workers feel about AI at work?](https://evy.so/facts/european-attitudes-to-ai-at-work/) - [F-39**53–66%**How much do European firms care where their AI processes data?](https://evy.so/facts/ai-data-protection-in-europe/) - [F-40**12–14×**How does AI investment in Europe compare with the US?](https://evy.so/facts/ai-investment-europe-vs-us/) [The whole chapter](https://evy.so/facts/europe/) · [Chapter poster](https://evy.so/facts/europe/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "Europe: statistics on AI at work in Europe | The AI at Work Fact Book" description: "Chapter 6 of The AI at Work Fact Book. AI at work in Europe: use by country for companies and workers, the gap with the US, how workers feel, readiness for the AI Act, data protection and investment." last_updated: "2026-10-24" source: "https://evy.so/facts/europe/" -------------------------------------------------------------------------------- - F-345–42% ## [Which EU countries' companies use AI the most and least?](https://evy.so/facts/ai-use-by-eu-country/) In 2025, between 5 percent of companies (Romania) and 42 percent (Denmark) used AI, among firms with 10 or more employees; the EU average was 20 percent. Germany was at 26 percent, Spain 20, France 18 and Portugal 12, with the Nordic countries and the Netherlands at 33 to 42 percent. High confidence 5 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/ai-use-by-eu-country/) - F-359–37% ## [What share of workers in each EU country use generative AI for work?](https://evy.so/facts/genai-use-for-work-by-eu-country/) In 2025, between 9 percent of workers (Romania) and 37 percent (Netherlands) had used generative AI for work in the previous three months; the EU figure was 22 percent. Germany was at 21 percent, France 25, Spain 26 and Portugal 28. Surveys a year later with broader questions find higher shares, up to about half of workers. Medium confidence 5 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/genai-use-for-work-by-eu-country/) - F-367–17 pts ## [How does AI use at work in Europe compare with the US?](https://evy.so/facts/ai-at-work-europe-vs-us/) US workers use AI more: in a survey asking the same question in seven countries in early 2026, 43 percent of US workers used generative AI for their job, against 26 to 36 percent in six large European countries, a gap of 7 to 17 points. For companies there is no clean comparison, and US firms that use AI spread it across more of their business. Medium confidence 5 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/ai-at-work-europe-vs-us/) - F-3748–86% ## [How do European workers feel about AI at work?](https://evy.so/facts/european-attitudes-to-ai-at-work/) Most Europeans view AI at work positively: 62 percent across the EU, from 48 percent in Portugal and Greece to 86 percent in Denmark (2024). Large majorities want rules on how employers use it, and two thirds expect AI to cost jobs overall, though far fewer worry about their own job. Medium confidence 5 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/european-attitudes-to-ai-at-work/) - F-3828% ## [How ready are European companies for the EU AI Act?](https://evy.so/facts/eu-ai-act-readiness/) About 3 in 10 companies say they understand what the AI Act requires: 28 percent in two independent 2026 surveys, and 71 percent of German firms feel poorly informed. Engagement is rising fast, and the deadlines for high-risk AI moved in July 2026 to December 2027 and August 2028. Medium confidence 3 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/eu-ai-act-readiness/) - F-3953–66% ## [How much do European firms care where their AI processes data?](https://evy.so/facts/ai-data-protection-in-europe/) A lot, by what they say: 53 to 66 percent of European firms that use or have considered AI cite data protection as an obstacle, more than any other concern. What they use tells a different story: in Germany, 93 percent said they would prefer a German AI provider, yet 76 percent of AI-using firms use ChatGPT and only 35 percent use any European or German AI application. Medium confidence 3 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/ai-data-protection-in-europe/) - F-4012–14× ## [How does AI investment in Europe compare with the US?](https://evy.so/facts/ai-investment-europe-vs-us/) Private investment in AI companies was 12 to 14 times higher in the US than in Europe in 2025: $285.9 billion against $20.9 billion. The gap is about companies that build AI; among firms that use it, Europe roughly matches the US. Most public money is announced, not yet spent. High confidence 2 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/ai-investment-europe-vs-us/) -------------------------------------------------------------------------------- title: "Europe, on one sheet | The AI at Work Fact Book" description: "Europe on one poster: 7 numbers from The AI at Work Fact Book, each with its sources printed on the sheet. Free to reuse under CC BY 4.0." last_updated: "2026-10-24" source: "https://evy.so/facts/europe/sheet/" -------------------------------------------------------------------------------- 5 to 42 of every 100 firms, by country F-34 5–42% of companies use AI, from Romania to Denmark; EU average 20% Eurostat 2025 9 to 37 of every 100 workers, by country F-35 9–37% of workers used generative AI for work, by country; EU 22% Eurostat 2025 a gap of 7 to 17 points out of 100 F-36 7–17 pts more US than European workers use generative AI for their job Bick et al., St. Louis Fed 2026 48 to 86 of every 100 people, by country F-37 48–86% view AI at work positively, by country; EU 62% Eurobarometer 554, 2024 28 to 28 of 100 F-38 28% of companies say they understand what the EU AI Act requires BSI and TÜV 2026 · ESSCA and Forvis Mazars 2026 53 to 66 of 100 F-39 53–66% of firms that use or considered AI cite data protection as an obstacle Eurostat 2025 · Bitkom 2026 Europe gets about $7 for every $100 of US private AI investment F-40 12–14× more private AI investment in the US than in Europe Stanford AI Index 2026 · OECD 2026 How to read it Each 10 by 10 grid is 100%. Solid orange is the low end of the range across studies, light orange runs to the high end. Methods, samples and links for every study are at **evy.so/facts**. ## The numbers on this sheet - [F-34**5–42%**of companies used AI in 2025, from Romania to Denmark; EU average 20%. Which EU countries' companies use AI the most and least? Sources: Eurostat, Denmark (top) (2025), Eurostat, Germany (2025), Eurostat, EU average (2025), Eurostat, Romania (lowest) (2025), Bick et al. (2026).](https://evy.so/facts/ai-use-by-eu-country/) - [F-35**9–37%**of workers used generative AI for work in 2025, from Romania to the Netherlands; EU 22%. What share of workers in each EU country use generative AI for work? Sources: Eurostat, Netherlands (top) (2025), Eurostat, EU (2025), Eurostat, Germany (2025), Eurostat, Romania (lowest) (2025), Bick et al., same question (2026).](https://evy.so/facts/genai-use-for-work-by-eu-country/) - [F-36**7–17 pts**more US than European workers use generative AI for their job (same survey, 2026). How does AI use at work in Europe compare with the US? Sources: United States (2026), UK (highest in Europe) (2026), Germany (2026), Italy (lowest in Europe) (2026), KPMG and Melbourne (2025).](https://evy.so/facts/ai-at-work-europe-vs-us/) - [F-37**48–86%**view AI at work positively, by country (EU 62%, 2024). How do European workers feel about AI at work? Sources: Eurobarometer, Denmark (top) (2025), Eurobarometer, Germany (2025), Eurobarometer, EU (2025), Eurobarometer, Portugal (lowest) (2025), ECB, euro area workers (2026).](https://evy.so/facts/european-attitudes-to-ai-at-work/) - [F-38**28%**of companies say they understand what the AI Act requires (two 2026 surveys). How ready are European companies for the EU AI Act? Sources: BSI and TÜV, Germany (2026), ESSCA and Forvis Mazars (2026), Bitkom, Germany (2026).](https://evy.so/facts/eu-ai-act-readiness/) - [F-39**53–66%**of firms that use or considered AI cite data protection as an obstacle. How much do European firms care where their AI processes data? Sources: Eurostat, EU (firms that considered AI) (2026), Eurostat, Germany (firms that considered AI) (2026), Bitkom, German AI users (2026).](https://evy.so/facts/ai-data-protection-in-europe/) - [F-40**12–14×**more private AI investment in the US than in Europe (2025). How does AI investment in Europe compare with the US? Sources: Stanford AI Index (2026), OECD (venture capital) (2026).](https://evy.so/facts/ai-investment-europe-vs-us/) Using the sheet in a post, deck or article? Credit "The AI at Work Fact Book by EVY" and link to [evy.so/facts/europe/](https://evy.so/facts/europe/). -------------------------------------------------------------------------------- title: "How do European workers feel about AI at work? (48–86%) | The AI at Work Fact Book" description: "48 to 86 percent of Europeans view AI at work positively by country (EU 62 percent), 72 to 82 percent want workplace rules, and two thirds expect job losses. Eurobarometer, ECB and Bitkom." last_updated: "2026-10-24" source: "https://evy.so/facts/european-attitudes-to-ai-at-work/" -------------------------------------------------------------------------------- F-37 · Europe # How do European workers feel about AI at work? Most Europeans view AI at work positively: 62 percent across the EU, from 48 percent in Portugal and Greece to 86 percent in Denmark (2024). Large majorities want rules on how employers use it, and two thirds expect AI to cost jobs overall, though far fewer worry about their own job. 48–86% view AI at work positively, by country (EU 62%, 2024) - Medium confidence - 5 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). How do European workers feel about AI at work? The AI at Work Fact Book, F-37. https://evy.so/facts/european-attitudes-to-ai-at-work/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/european-attitudes-to-ai-at-work.json) [Markdown](https://evy.so/facts/european-attitudes-to-ai-at-work.md) ## How to read it The Eurobarometer asks all adults, face to face, about robots and AI together, in 2024. Newer surveys that ask workers about AI alone come out lower, around 41 to 49% positive in 2026, and the two should not be read as a trend. The country pattern holds across sources: the Nordic countries and the Netherlands are most positive and least worried about jobs; Portugal, Greece and France least positive. ## Every study on one scale Shaded: 48–86% by country One figure Chart values as text - Scale: % viewing AI at work positively. Shaded range: 48–86% by country. - Eurobarometer, Denmark (top) (2025): Denmark 86% - Eurobarometer, Germany (2025): Germany 65% - Eurobarometer, EU (2025): EU 62% - Eurobarometer, Portugal (lowest) (2025): Portugal 48% - ECB, euro area workers (2026): 41% (workers, 2026) **Why medium confidence:** The EU-wide figures come from one official survey with a clear method, but it dates from spring 2024, covers all adults and asks about robots and AI together. Later surveys of workers about AI alone find lower positive shares; the other findings point the same way across creators. ## Sources Sources for F-37 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Special Eurobarometer 554: Artificial Intelligence and the future of work | European Commission, fieldwork by Verian | 2025 | 26,415 face-to-face interviews with people aged 15 and over in all 27 EU countries, April to May 2024 | Official Eurobarometer; asks about robots and AI in the workplace together | Viewed positively: Denmark 86%, Finland 80%, Sweden 79%, Netherlands 77%, Germany 65%, EU 62%, Spain 56%, France 53%, Greece and Portugal 48% “positively perceive the use of robots and Artificial Intelligence in the workplace” | [europa.eu ↗](https://europa.eu/eurobarometer/surveys/detail/3222) | | Special Eurobarometer 554: Artificial Intelligence and the future of work | European Commission, fieldwork by Verian | 2025 | Same survey, Germany | Same survey | Germany 65% positive; 53% expect AI to destroy more jobs than it creates (EU 66%) | [europa.eu ↗](https://europa.eu/eurobarometer/surveys/detail/3222) | | Special Eurobarometer 554: Artificial Intelligence and the future of work | European Commission, fieldwork by Verian | 2025 | Same survey, all 27 countries | Same survey | EU 62% positive, 32% negative; 66% of people in work say recent digital technologies including AI help their job | [europa.eu ↗](https://europa.eu/eurobarometer/surveys/detail/3222) | | Special Eurobarometer 554: Artificial Intelligence and the future of work | European Commission, fieldwork by Verian | 2025 | Same survey, Portugal and Greece | Same survey | Portugal and Greece 48% positive | [europa.eu ↗](https://europa.eu/eurobarometer/surveys/detail/3222) | | AI adoption and the productivity promise: what workers report (ECB Blog) | European Central Bank | 2026 | About 20,000 people a month in 11 euro area countries, 2026 wave | Consumer Expectations Survey; workers' views of AI alone | 41% of workers view AI positively, down from 43% a year earlier; 41% are uninterested in using it “AI is perceived less favourably today than one year ago” | [ecb.europa.eu ↗](https://www.ecb.europa.eu/press/blog/date/2026/html/ecb.blog20260826~e1c1a89999.en.html) | Rules at work ## What rules do Europeans want on AI at work? Between 72% and 82% call each of five workplace rules important: protecting workers' privacy (82%), involving workers in designing and adopting new technology (77%), transparency in HR decisions (75%), banning fully automated decisions (74%) and limiting automated monitoring (72%). Europeans object most to AI automatically firing workers (78%) and monitoring them (63%). Jobs ## Do Europeans fear losing their jobs to AI? In general, yes; for themselves, much less. 66% believe AI will destroy more jobs than it creates, from 45% in Denmark to 80% in Greece (Germany 53%). But only 22% of German workers name worry about their own job as a drawback of AI (Bitkom, 2026), and 41% of EU workers who use AI are very or somewhat concerned about losing their job to it (European Commission, 2026). Do Europeans fear losing their jobs to AI? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Ein Drittel nutzt KI mindestens einmal pro Woche (press release, 28 April 2026) | Bitkom | 2026 | 550 employed people in Germany, by telephone, February to March 2026 | Industry association survey; drawbacks named, several answers allowed | Worry about own job 22%; unclear who is responsible for AI errors 62%; unclear where data goes 59%; 49% would like an AI assistant, 45% want no AI support “22 Prozent haben Sorge um den eigenen Arbeitsplatz” | [bitkom.org ↗](https://www.bitkom.org/Presse/Presseinformation/Ein-Drittel-nutzt-KI-mindestens-einmal-pro-Woche) | | The AI-adoption divide: who benefits, who doesn't, and what it means for workers | European Commission, Directorate-General for Economic and Financial Affairs | 2026 | 21,207 responses in 18 EU countries, February and March 2026; this question asked of AI users | Ad-hoc module in the harmonized consumer surveys | Of employed AI users, 14% very and 27% somewhat concerned about losing their job to AI “14% are very concerned and 27% are somewhat concerned about potential job displacement” | [economy-finance.ec.europa.eu ↗](https://economy-finance.ec.europa.eu/economic-forecast-and-surveys/economic-forecasts/spring-2026-economic-forecast-slowdown-growth-energy-shock-drives-inflation/ai-adoption-divide-who-benefits-who-doesnt-and-what-it-means-workers_en) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **84% of Europeans want AI at work regulated (Eurobarometer).** 84% agree robots and AI require careful management, a general statement. The specific workplace rules drew 72 to 82%. - **66% of Europeans fear losing their job to AI.** 66% believe AI will destroy more jobs than it creates in general. Fear for one's own job is much lower. - **62% of European workers are positive about AI.** The Eurobarometer asked all adults about robots and AI together, in 2024. Among euro area workers asked about AI in 2026, 41% are positive (ECB). - **Half of Germans reject AI at work (Bitkom).** 45% of German workers want no AI support at work, against 49% who would like an AI assistant: a split among workers only. ## More from Europe - [F-34**5–42%**Which EU countries' companies use AI the most and least?](https://evy.so/facts/ai-use-by-eu-country/) - [F-35**9–37%**What share of workers in each EU country use generative AI for work?](https://evy.so/facts/genai-use-for-work-by-eu-country/) - [F-36**7–17 pts**How does AI use at work in Europe compare with the US?](https://evy.so/facts/ai-at-work-europe-vs-us/) - [F-38**28%**How ready are European companies for the EU AI Act?](https://evy.so/facts/eu-ai-act-readiness/) - [F-39**53–66%**How much do European firms care where their AI processes data?](https://evy.so/facts/ai-data-protection-in-europe/) - [F-40**12–14×**How does AI investment in Europe compare with the US?](https://evy.so/facts/ai-investment-europe-vs-us/) [The whole chapter](https://evy.so/facts/europe/) · [Chapter poster](https://evy.so/facts/europe/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "What share of workers in each EU country use generative AI for work? (9–37%) | The AI at Work Fact Book" description: "Between 9 percent (Romania) and 37 percent (Netherlands) of EU workers used generative AI for work in 2025, EU 22 percent. Country figures and the newer 2026 surveys." last_updated: "2026-10-24" source: "https://evy.so/facts/genai-use-for-work-by-eu-country/" -------------------------------------------------------------------------------- F-35 · Europe # What share of workers in each EU country use generative AI for work? In 2025, between 9 percent of workers (Romania) and 37 percent (Netherlands) had used generative AI for work in the previous three months; the EU figure was 22 percent. Germany was at 21 percent, France 25, Spain 26 and Portugal 28. Surveys a year later with broader questions find higher shares, up to about half of workers. 9–37% of workers used generative AI for work in 2025, from Romania to the Netherlands; EU 22% - Medium confidence - 5 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). What share of workers in each EU country use generative AI for work? The AI at Work Fact Book, F-35. https://evy.so/facts/genai-use-for-work-by-eu-country/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/genai-use-for-work-by-eu-country.json) [Markdown](https://evy.so/facts/genai-use-for-work-by-eu-country.md) ## How to read it The chart shows the top country, the EU, Germany and the lowest country from one official survey; the table above lists the others. The survey asks about generative AI in the previous three months in early 2025, a fast-moving area. Within countries the gap between office and manual workers (28% against 5%) is larger than most gaps between countries. Chapter 3 covers the US and the share of workers worldwide. ## Every study on one scale Shaded: 9–37% across EU countries One figure Chart values as text - Scale: % of workers using generative AI for work, 2025. Shaded range: 9–37% across EU countries. - Eurostat, Netherlands (top) (2025): Netherlands 37.1% - Eurostat, EU (2025): EU 21.8% - Eurostat, Germany (2025): Germany 20.8% - Eurostat, Romania (lowest) (2025): Romania 8.6% **Why medium confidence:** The country figures are official, harmonized statistics, and an independent survey correlates 0.88 with them, so the ranking is solid. The level is less certain: the data describe early 2025, and newer surveys with broader wording find higher shares. Studies without a figure on this scale are listed in the table only. ## Sources Sources for F-35 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Individuals: use of generative AI tools (dataset isoc\_ai\_iaiu) | Eurostat | 2025 | Official household survey in all EU countries, mostly second quarter of 2025 | Self-reported use of generative AI tools for work in the previous 3 months, employed people aged 16 to 74 | Netherlands 37.1%, Finland 36.9%, Denmark 35.0%, Sweden 29.9%, Portugal 27.6%, Spain 26.1%, France 24.6%, Germany 20.8%, Italy 12.5%, Romania 8.6%; Norway 43.8%; EU 21.8% “Use of generative AI tools: for professional (work) purposes” | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/databrowser/view/isoc_ai_iaiu/default/table) | | Individuals: use of generative AI tools (dataset isoc\_ai\_iaiu) | Eurostat | 2025 | Same survey, all EU countries | Same definition | 21.8% of employed people; 27.6% in non-manual jobs, 5.3% in manual jobs | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/databrowser/view/isoc_ai_iaiu/default/table) | | Individuals: use of generative AI tools (dataset isoc\_ai\_iaiu) | Eurostat | 2025 | Same survey, Germany | Same definition | Germany 20.8% for work; 36.7% for any purpose | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/databrowser/view/isoc_ai_iaiu/default/table) | | Individuals: use of generative AI tools (dataset isoc\_ai\_iaiu) | Eurostat | 2025 | Same survey, Romania | Same definition | Romania 8.6% | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/databrowser/view/isoc_ai_iaiu/default/table) | | Mind the Gap: AI Adoption in Europe and the US (St. Louis Fed Working Paper 2026-003) | Alexander Bick, Adam Blandin, David J. Deming, Nicola Fuchs-Schündeln and Jonas Jessen | 2026 | About 3,000 workers per country, January to February 2026 | Online panel weighted to labor force surveys; the same question in each country | Netherlands and Sweden 35.6%, Germany 31.5%, France 28.1%, Italy 25.6%; correlation with Eurostat country rates 0.88 “Do you use Generative AI for your job? (No/Yes)” | [doi.org ↗](https://doi.org/10.20955/wp.2026.003) | 2026 surveys ## How many European workers use AI at work in 2026? About a third to a half, in surveys with broader questions. 34.5% of employed people in 18 EU countries used AI for work in February to March 2026 (European Commission). 52% of workers in 11 euro area countries used AI in their work in 2026 (ECB, any AI tool). In Germany, 48% of workers use AI at work at any frequency, and 8% daily (Bitkom, 2026). How many European workers use AI at work in 2026? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | AI adoption and the productivity promise: what workers report (ECB Blog) | European Central Bank | 2026 | About 20,000 people a month in 11 euro area countries | Consumer Expectations Survey; any AI tool, no time window | 52% of employed workers use AI in their work (26% in 2024, 41% in 2025) “A majority of workers now report using AI for work” | [ecb.europa.eu ↗](https://www.ecb.europa.eu/press/blog/date/2026/html/ecb.blog20260826~e1c1a89999.en.html) | | Ein Drittel nutzt KI mindestens einmal pro Woche (press release, 28 April 2026) | Bitkom | 2026 | 550 employed people in Germany, by telephone, February to March 2026 | Industry association survey; any AI use at work, any frequency | 48% of German workers use AI at work: 8% daily, 18% weekly, 22% less often; 12% of users do so without their employer's knowledge “48 Prozent der Erwerbstätigen nutzen KI bei der Arbeit” | [bitkom.org ↗](https://www.bitkom.org/Presse/Presseinformation/Ein-Drittel-nutzt-KI-mindestens-einmal-pro-Woche) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **15% of Europeans use generative AI for work (Eurostat).** That counts everyone aged 16 to 74, including students and retirees. Among employed people it is 21.8%. - **Over half of working-age Europeans use generative AI for work.** Eurostat's over 50% refers to users: 55.6% of users aged 25 to 64 use it for work, which is 18.2% of everyone that age. - **Half of German workers use AI at work (Bitkom).** It counts any frequency, including 22% who use it less than weekly. Daily use is 8%; the official 2025 rate for generative AI is 20.8%. - **A third of Europeans use generative AI at work.** The 32.7% is use for any purpose by all adults. Work use is 21.8% of employed people. ## More from Europe - [F-34**5–42%**Which EU countries' companies use AI the most and least?](https://evy.so/facts/ai-use-by-eu-country/) - [F-36**7–17 pts**How does AI use at work in Europe compare with the US?](https://evy.so/facts/ai-at-work-europe-vs-us/) - [F-37**48–86%**How do European workers feel about AI at work?](https://evy.so/facts/european-attitudes-to-ai-at-work/) - [F-38**28%**How ready are European companies for the EU AI Act?](https://evy.so/facts/eu-ai-act-readiness/) - [F-39**53–66%**How much do European firms care where their AI processes data?](https://evy.so/facts/ai-data-protection-in-europe/) - [F-40**12–14×**How does AI investment in Europe compare with the US?](https://evy.so/facts/ai-investment-europe-vs-us/) [The whole chapter](https://evy.so/facts/europe/) · [Chapter poster](https://evy.so/facts/europe/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "Is there a gender gap in AI use at work, and how large is it? (5–16 pts) | The AI at Work Fact Book" description: "Men use generative AI more than women by 5 to 16 points, about 16 to 22 percent more likely. How the gap changes in the same job, and which figures are outdated." last_updated: "2026-10-24" source: "https://evy.so/facts/gender-gap-in-ai-use/" -------------------------------------------------------------------------------- F-18 · Who Uses AI at Work # Is there a gender gap in AI use at work, and how large is it? Most studies find men use generative AI more than women, by about 5 to 16 percentage points, which makes men about 16 to 22 percent more likely to use it. The gap shrinks by a quarter to a third among people in the same job, and stays open. 5–16 pts more men than women use generative AI; men are 16 to 22% more likely - Medium confidence - 5 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). Is there a gender gap in AI use at work, and how large is it? The AI at Work Fact Book, F-18. https://evy.so/facts/gender-gap-in-ai-use/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/gender-gap-in-ai-use.json) [Markdown](https://evy.so/facts/gender-gap-in-ai-use.md) ## How to read it The review pools work and personal use, students and workers, so its levels are not workforce rates; its value is the direction and the trend, with the gap narrowing to about 16% since early 2025. The US survey is the cleanest whole-workforce figure (6 points, late 2024). The Danish study is the cleanest comparison within the same job, but covers ChatGPT only, early on. Pew's strict question shows no gap, which suggests the gap is smaller for regular, job-embedded use than for any use. ## Every study on one scale Shaded: 5–16 points One figure A range in one study Chart values as text - Scale: points gap, men minus women. Shaded range: 5–16 points. - Cranney, Delecourt and Koning (2026): 8.5 points overall, 5.4 since 2025 - Bick, Blandin and Deming, US (2025): 6 points (35% against 29%) - Humlum and Vestergaard, Denmark (2025): 15.9 points (11.8 same workplace and tasks) **Why medium confidence:** A US national survey, a Danish survey linked to government records and a review of 58 sources all find men ahead. Two large surveys report no meaningful gap on their measures, and the size varies with who is asked and how use is defined. Studies without a figure on this scale are listed in the table only. ## Sources Sources for F-18 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Global Evidence on Gender Gaps and Generative AI Over Time (HBS Working Paper 25-023, May 2026) | Katelyn Cranney (Stanford), Solène Delecourt (Berkeley Haas) and Rembrand Koning (Harvard Business School) | 2026 | 76 sources from more than 100 countries, 2023 to early 2026; 318,924 people in the pooled estimate | Systematic review, weighted by sample size. Mixes work and personal use, students and workers; many sources are convenience samples. | 47.8% of men against 39.3% of women use AI (men 22% more likely); in 2025 to 2026 data, 39.8% against 34.4% (5.4 points, men 16% more likely); men ahead in 56 of 58 sources “we estimate an AI adoption rate of 47.8% for men versus 39.3% for women” | [hbs.edu ↗](https://www.hbs.edu/ris/Publication%20Files/25-023_be8fb517-3dd5-40aa-97f9-4e42e1c8e6ff.pdf) | | The Rapid Adoption of Generative AI (St. Louis Fed Working Paper 2024-027F, revised October 2025) | Alexander Bick, Adam Blandin and David Deming | 2025 | 6,935 employed US adults aged 18 to 64, August and November 2024 | Nationally representative survey; ever used generative AI for work | Men 35.0%, women 29.0%; with occupation controls the gap in last-week use falls from 7.0 to 4.8 points, and to 1.8 points once employer encouragement is added “The top panel shows genAI adoption is 6 percentage points higher for men.” | [doi.org ↗](https://doi.org/10.20955/wp.2024.027) | | The unequal adoption of ChatGPT exacerbates existing inequalities among workers (PNAS) | Anders Humlum (Chicago Booth) and Emilie Vestergaard (University of Copenhagen) | 2025 | 17,907 workers in 11 AI-exposed occupations in Denmark, late 2023 | Survey linked to government records; ChatGPT only, ever used for work | Women 15.9 points less likely than men in the same occupation; 11.8 points among coworkers in the same workplace with the same tasks “Women are 16 percentage points less likely to have used the tool for work.” | [pnas.org ↗](https://www.pnas.org/doi/10.1073/pnas.2414972121) | | U.S. Workers Are More Worried Than Hopeful About Future AI Use in the Workplace | Pew Research Center | 2025 | 5,273 employed US adults, October 2024 | Strict question: at least some of one's work is done with AI | Men are 53% of AI users and 53% of non-users: no gap on this measure “AI users are younger and more educated compared with other workers” | [pewresearch.org ↗](https://www.pewresearch.org/social-trends/2025/02/25/u-s-workers-are-more-worried-than-hopeful-about-future-ai-use-in-the-workplace/) | | Trust, attitudes and use of artificial intelligence: A global study 2025 | University of Melbourne and KPMG International | 2025 | 32,352 employed people in 47 countries | Reports only group differences above an effect-size cut-off, so a small gap may exist below it | No gender difference reported in AI use at work “there are no gender differences in AI use or attitudes toward AI at work” | [mbs.edu ↗](https://mbs.edu/-/media/PDF/Research/Trust_in_AI_Report.pdf) | Same job ## Does the gender gap close among people in the same job? It shrinks and stays open. In Denmark the gap falls from 15.9 to 11.8 points among coworkers in the same workplace doing the same tasks. In the US it falls from 7.0 to 4.8 points with occupation controls, and to 1.8 points (only weakly significant) once employer encouragement is added, which the authors say may not be a cause. Danish women more often said they needed training; men more often cited employer restrictions. ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Women are 20% (or 25%) less likely than men to use generative AI.** From 2024 drafts of the Harvard review, which reported 22% lower odds, a different quantity from 22% lower probability. The May 2026 version says men are 22% more likely overall and about 16% since early 2025. - **Women are 16 points less likely to use AI at work.** That is Denmark, ChatGPT only, in 11 exposed jobs, late 2023. US whole-workforce data from late 2024 show 6 points. - **There is no gender gap in AI use at work (KPMG).** The study reports only gaps above an effect-size cut-off and publishes no rates by gender. It rules out a large gap, not a small one. - **Only 44% of ChatGPT users are women.** It is the female share of US visits to chatgpt.com, with gender inferred from web traffic. It gives no usage rate for men or women. ## More from Who Uses AI at Work - [F-14**40–50%**What share of workers use generative AI at work?](https://evy.so/facts/share-of-workers-using-ai/) - [F-15**40–52%**What do people use AI for at work?](https://evy.so/facts/what-workers-use-ai-for/) - [F-16**78–85%**Which occupations and industries use AI the most?](https://evy.so/facts/ai-use-by-occupation-and-industry/) - [F-17**12–24 pts**Who uses AI at work more, by age, education and income?](https://evy.so/facts/ai-use-by-age-education-income/) - [F-19**50–78%**How many workers use AI tools their employer did not provide, and how many hide it?](https://evy.so/facts/bring-your-own-ai/) - [F-20**42–49%**How many employers provide AI tools, guidance or training to staff?](https://evy.so/facts/employers-providing-ai-tools/) [The whole chapter](https://evy.so/facts/who-uses-ai-at-work/) · [Chapter poster](https://evy.so/facts/who-uses-ai-at-work/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "What share of workers use generative AI at work? (40–50%) | The AI at Work Fact Book" description: "About 40 to 50 percent of US workers use generative AI for their job (2026), and 14 to 15 percent use it daily. Gallup, Pew, St. Louis Fed and EU figures, with sources." last_updated: "2026-10-24" source: "https://evy.so/facts/share-of-workers-using-ai/" -------------------------------------------------------------------------------- F-14 · Who Uses AI at Work # What share of workers use generative AI at work? About 40 to 50 percent of US workers now use generative AI for their job, depending on how the question is asked. Daily use is much smaller: about 14 to 15 percent of all workers, or a quarter to a third of the people who use it. 40–50% of US workers use generative AI for their job (2026); studies range from 38 to 52% Across all 4 studies: 38 to 52% across the three main surveys. - High confidence - 4 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). What share of workers use generative AI at work? The AI at Work Fact Book, F-14. https://evy.so/facts/share-of-workers-using-ai/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/share-of-workers-using-ai.json) [Markdown](https://evy.so/facts/share-of-workers-using-ai.md) ## How to read it The surveys ask different questions. The St. Louis Fed survey asks about generative AI for your job and about last week; Gallup asks about any AI and counts people who use it a few times a year; Pew asks whether people ever use chatbots for work tasks, and in a stricter question how much of their work is done with AI. All are self-reported. Gallup and the St. Louis Fed survey both find daily use at about 14 to 15% of workers. ## Every study on one scale Shaded: 40–50% Dashed: 38–52% One figure A range in one study Chart values as text - Scale: % of US workers using AI for work. Shaded range: 40–50%; dashed, 38–52%: 38% to 52%. - St. Louis Fed survey (2026): 39.2% last week, 45.2% for their job - Gallup (2026): 52% (a few times a year or more) - Pew (2026): 38% (chatbots, ever) - Pew, stricter question (2025): 21% (some of their work done with AI) **Why high confidence:** Three independent US surveys with different samples and wording land in the same range in the first half of 2026, and all three show the same rising trend. The width of the range comes from how each one defines use. ## Sources Sources for F-14 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Real-Time Population Survey: Generative AI Adoption Tracker (FRED series RPSGENAIUSAGESHARELWWORK) | Alexander Bick (St. Louis Fed), Adam Blandin (Vanderbilt) and David Deming (Harvard); data published by the Federal Reserve Bank of St. Louis | 2026 | Nationally representative online survey of US adults aged 18 to 64, May 2026 wave | Self-reported. Asks whether people use generative AI for their job, and on which workdays last week. | 45.2% use generative AI for their job; 39.2% used it for work last week (28.2% in August 2024); 13.7% every workday last week “Did you use Generative AI for your job LAST WEEK?” | [fred.stlouisfed.org ↗](https://fred.stlouisfed.org/series/RPSGENAIUSAGESHARELWWORK) | | Organizational AI Adoption Jumps Six Points (Q2 2026 workforce study) | Gallup | 2026 | 22,573 US employees, 6 to 20 May 2026 | Probability-based Gallup Panel web survey. Asks about AI in general, by frequency. | 52% use AI at work a few times a year or more, 30% a few times a week or more, 15% daily “As of May 2026, 15% of U.S. employees use AI daily in their role,” | [gallup.com ↗](https://www.gallup.com/workplace/712736/organizational-adoption-jumps-six-points.aspx) | | Americans and AI 2026: Chatbots, Smart Devices and Views on Impact | Pew Research Center | 2026 | 5,119 US adults, 17 to 23 February 2026 | American Trends Panel, probability-based. Asks whether people ever use AI chatbots for tasks at work. | 38% of employed adults use AI chatbots for tasks at work “38% of employed adults report using chatbots for tasks at work.” | [pewresearch.org ↗](https://www.pewresearch.org/internet/2026/06/17/americans-and-ai-2026-chatbots-smart-devices-and-views-on-impact/) | | About 1 in 5 U.S. workers now use AI in their job, up since last year | Pew Research Center | 2025 | 5,010 US workers, 2 to 8 September 2025 | Asks how much of a worker's own job is done with AI. A stricter question, so it sits below the others. | 21% say at least some of their work is done with AI, up from 16% in October 2024 “21% of U.S. workers say at least some of their work is done with AI” | [pewresearch.org ↗](https://www.pewresearch.org/short-reads/2025/10/06/about-1-in-5-us-workers-now-use-ai-in-their-job-up-since-last-year/) | How often ## How often do workers who use AI use it? A quarter to a third of them use it every workday, and a little over half use it at least weekly. In May 2026, 15% of all US employees used AI daily and 30% a few times a week or more (Gallup); 13.7% used generative AI every workday of the previous week (St. Louis Fed survey). On days they use it, about a third of US users spend an hour or more with it. The among-user shares are our division of the published figures. How often do workers who use AI use it? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | The Rapid Adoption of Generative AI (NBER Working Paper 32966) | Alexander Bick, Adam Blandin and David Deming | 2025 | 1,918 employed generative AI users, August and November 2024 | Self-reported days of use last week and minutes per day of use | 34% of work users used it every workday last week; on days used, 32% spent an hour or more “34 percent used it every workday in the previous week,” | [nber.org ↗](https://www.nber.org/papers/w32966) | | Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI (NBER Working Paper 33777) | Anders Humlum (Chicago Booth) and Emilie Vestergaard (University of Copenhagen), with Statistics Denmark | 2026 | About 25,000 workers in 11 AI-exposed occupations, late 2024 | Survey linked to Danish government records | 16% of workers use AI chatbots daily and 35% weekly; daily use is 7% where employers do nothing and 28% where they encourage use and provide tools and training “have used AI chatbots at work at least once, and 7% use them daily” | [nber.org ↗](https://www.nber.org/papers/w33777) | In the EU ## What share of workers in the EU use generative AI for work? 21.8% of employed people in the EU used generative AI tools for work in the three months before the 2025 survey, according to Eurostat. The often quoted 15% is the share of everyone aged 16 to 74, including people who are not working. What share of workers in the EU use generative AI for work? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Individuals: use of generative AI tools (dataset isoc\_ai\_iaiu) | Eurostat | 2026 | EU household ICT survey, all 27 member states, 2025 | Official statistics; self-reported use in the previous 3 months, by purpose | 21.8% of employed people used generative AI for work; 15.4% of all people aged 16 to 74 “Use of generative AI tools: for professional (work) purposes” | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/databrowser/view/isoc_ai_iaiu/default/table) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **75% of workers use AI at work (Microsoft).** Microsoft's 2024 survey covered desk-based knowledge workers in an online panel across 31 markets. It does not describe the whole workforce. - **Half of US workers use AI at work (Gallup).** Gallup's 52% counts people who use AI a few times a year or more. 30% use it a few times a week and 15% daily. - **16% of US workers use AI (Pew).** That is October 2024, on Pew's strictest question. The same question gave 21% in September 2025. - **15% of Europeans use AI for work (Eurostat).** That share counts everyone aged 16 to 74, revised to 15.4%. Among employed people it is 21.8%. - **Half of workers use AI daily.** We found no source for it. Daily use is 14 to 15% of US workers. ## More from Who Uses AI at Work - [F-15**40–52%**What do people use AI for at work?](https://evy.so/facts/what-workers-use-ai-for/) - [F-16**78–85%**Which occupations and industries use AI the most?](https://evy.so/facts/ai-use-by-occupation-and-industry/) - [F-17**12–24 pts**Who uses AI at work more, by age, education and income?](https://evy.so/facts/ai-use-by-age-education-income/) - [F-18**5–16 pts**Is there a gender gap in AI use at work, and how large is it?](https://evy.so/facts/gender-gap-in-ai-use/) - [F-19**50–78%**How many workers use AI tools their employer did not provide, and how many hide it?](https://evy.so/facts/bring-your-own-ai/) - [F-20**42–49%**How many employers provide AI tools, guidance or training to staff?](https://evy.so/facts/employers-providing-ai-tools/) [The whole chapter](https://evy.so/facts/who-uses-ai-at-work/) · [Chapter poster](https://evy.so/facts/who-uses-ai-at-work/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How many hours do small business owners work, and how much goes to admin? (45–52 h) | The AI at Work Fact Book" description: "Small business owners work about 45 to 52 hours a week, roughly 10 more than employees, and admin takes 22 to 36 percent of their time. Eurostat and owner surveys." last_updated: "2026-10-24" source: "https://evy.so/facts/small-business-owner-hours/" -------------------------------------------------------------------------------- F-33 · Small Teams # How many hours do small business owners work, and how much goes to admin? Small business owners work about 45 to 52 hours a week, roughly 10 hours more than employees: EU employers worked 46.3 hours in 2025 against 36.6 for employees. Admin takes about a fifth to a third of owner time, by vendor surveys. 45–52 h a week worked by small business owners; employees work about 37 Across all 4 studies: up to a median of 60 hours for firms with five or fewer employees. - Medium confidence - 4 primary sources - 2023 to 2025 - Last checked October 14, 2026 ## Cite this EVY (2026). How many hours do small business owners work, and how much goes to admin? The AI at Work Fact Book, F-33. https://evy.so/facts/small-business-owner-hours/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/small-business-owner-hours.json) [Markdown](https://evy.so/facts/small-business-owner-hours.md) ## How to read it Eurostat's household survey is the most comparable across countries; the owner surveys are self-reported, from small or undisclosed samples, by companies that sell software or services to owners. Eurostat counts employers of any firm size, though most run small firms. The often quoted 52 hours comes from a 2005 Gallup poll. No current official US figure exists: the Census Bureau's owner survey has no hours question. ## Every study on one scale Shaded: 45–52 h Dashed: microbusinesses One figure Chart values as text - Scale: hours a week. Shaded range: 45–52 h; dashed, microbusinesses: 45 h to 60 h. - Eurostat, EU employers (2025): 46.3 h (employees 36.6 h) - Time etc (vendor) (2023): 45.5 h - Sage (vendor) (2024): 51 h - UENI (vendor) (2024): 60 h median (5 or fewer staff) **Why medium confidence:** An official, large-sample source agrees with independent owner surveys using different methods on 45 to 52 hours for owners of employer firms. The admin share rests on two small vendor surveys, and no official source measures it. ## Sources Sources for F-33 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Average number of usual weekly hours of work in main job (dataset lfsa\_ewhuis) | Eurostat (EU Labour Force Survey) | 2025 | EU Labour Force Survey, ages 20 to 64, all 27 member states, 2025 | Official household survey of usual weekly hours. Employers are self-employed people with employees; there is no firm-size breakdown. | Employers 46.3 hours a week (full-time 47.5); self-employed without employees 39.7; employees 36.6 “Self-employed persons with employees (employers)” | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/databrowser/view/lfsa_ewhuis/default/table?lang=en) | | The Big Price Of Small Tasks | Time etc, survey by Censuswide | 2023 | 251 US owners of growing businesses, September 2023 | Online panel survey by a virtual-assistant company that sells admin outsourcing | 45.5 hours a week; 36% of the week on small admin tasks such as invoicing and data entry “the average work week for entrepreneurs at the helm of growing businesses is 45.5 hours” | [timeetc.com ↗](https://www.timeetc.com/resources/how-to-achieve-more/the-big-price-of-small-tasks-how-entrepreneurs-may-be-unwittingly-keeping-their-businesses-small) | | An extra hour of SMB productivity a week worth £7bn to UK economy | Sage | 2024 | UK small business owners; sample size not stated | Vendor survey; Sage sells accounting software | 51 hours a week on average; 51% wish they spent less time on admin “On average, they invest 51 hours weekly in their business” | [sage.com ↗](https://www.sage.com/en-gb/company/digital-newsroom/2024/04/24/an-extra-hour-of-smb-productivity-a-week-worth-7bn-to-uk-economy) | | How microbusiness owners spend their time | UENI Research | 2024 | 837 US businesses with five or fewer employees, May 2024 | Vendor survey; recruitment not stated. UENI sells websites and AI tools to microbusinesses. | Median 60 hours a week; 62% work more than 50; admin plus legal and compliance take 22.4% of owner time “The median number of hours worked per week among microbusinesses surveyed was 60 hours.” | [ueni.com ↗](https://ueni.com/en-us/small-business-research/how-microbusiness-owners-spend-their-time) | Admin time Low confidence ## How much of a small business owner's time goes to admin? About a fifth to a third, in two vendor surveys: 22% for administration plus legal and compliance among US firms with five or fewer employees (UENI, 2024), and 36% for small admin tasks such as invoicing and data entry among owners of growing businesses (Time etc, 2023). Applied to a 45-hour week that is roughly 10 to 16 hours; the surveys report only the shares. AI and the workday ## Does AI shorten small business owners' workdays? For some. In January 2026, 27% of AI-using US businesses with up to 100 employees said AI had shortened their workday and 8% that it had lengthened it, in Intuit's vendor survey. Does AI shorten small business owners' workdays? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | 2026 AI Impact Report | Intuit QuickBooks, with economists at the University of Chicago | 2026 | 2,222 US owners of businesses with 0 to 100 employees, January 2026 | Paid panel and Intuit customers. Intuit sells AI features in QuickBooks. | 27% say AI shortened their workday; 8% that it lengthened it “27% of US businesses using AI said it had shortened their workday” | [intuit.com ↗](https://www.intuit.com/blog/global-stories/ai-impact-report/) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Small business owners work 52 hours a week.** That comes from a Gallup telephone poll of 607 owners in June 2005, usually repeated without the year. Current figures run 45 to 51 hours. - **Small business owners spend 16 hours a week on admin.** Third parties derived it by multiplying a 45.5-hour week by a 36% admin share. The survey reports only those two numbers. - **Small business owners spend 23 hours a week on admin (SCORE).** We found no primary report containing it; it appears only on aggregator pages. ## More from Small Teams - [F-28**96–99%**How much of the economy do small companies make up?](https://evy.so/facts/small-companies-share-of-economy/) - [F-29**20–46%**What share of small businesses use AI, and for what?](https://evy.so/facts/small-businesses-using-ai/) - [F-30**13–38 pts**How does AI adoption differ by company size?](https://evy.so/facts/ai-adoption-by-company-size/) - [F-31**54–63%**What stops small companies from adopting AI?](https://evy.so/facts/what-stops-small-companies-adopting-ai/) - [F-32**65–78%**Do small firms that use AI see gains?](https://evy.so/facts/small-firms-gains-from-ai/) [The whole chapter](https://evy.so/facts/small-teams/) · [Chapter poster](https://evy.so/facts/small-teams/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "What share of small businesses use AI, and for what? (20–46%) | The AI at Work Fact Book" description: "20 to 46 percent of small businesses use AI, mostly for writing and marketing; vendor surveys report 66 to 77 percent. Official counts, owner surveys and what small firms pay for AI." last_updated: "2026-10-24" source: "https://evy.so/facts/small-businesses-using-ai/" -------------------------------------------------------------------------------- F-29 · Small Teams # What share of small businesses use AI, and for what? Between a fifth and a half of small businesses use AI, 20 to 46 percent in official counts and weighted owner surveys from 2025 to 2026, mostly for writing and marketing. Use is shallow: only 7 percent of AI-using US small firms have fully built it into how they work. 20–46% of small businesses use AI (official counts and weighted owner surveys; vendor surveys report 66 to 77%) Across all 8 studies: 20 to 77% counting vendor and advocacy surveys. - Medium confidence - 8 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). What share of small businesses use AI, and for what? The AI at Work Fact Book, F-29. https://evy.so/facts/small-businesses-using-ai/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/small-businesses-using-ai.json) [Markdown](https://evy.so/facts/small-businesses-using-ai.md) ## How to read it Read the figures in three tiers and do not average them: official counts for the smallest firms (20 to 29%), weighted owner surveys (24 to 46%) and vendor or advocacy surveys (66 to 77%). Official surveys ask whether the business uses AI, which owners often answer no when only individuals use a chatbot; owner surveys that count anyone at the firm, or AI features inside other software, come out higher. Online panels and program alumni lean toward tech-forward owners. ## Every study on one scale Shaded: 20–46% Dashed: with vendor surveys One figure Chart values as text - Scale: % of small businesses using AI. Shaded range: 20–46%; dashed, with vendor surveys: 20% to 77%. - US Census, 1 to 4 employees (2026): 23.9% - ONS, under 10 employees (2026): 28.7% - Statistics Canada, 1 to 4 (2026): 19.9% - NFIB (2025): 24% - OECD, seven countries (2025): 30.7% - Federal Reserve (2026): 46% - US Chamber (advocacy) (2026): 66% (vendor or advocacy) - Intuit (vendor) (2026): 77% (vendor) **Why medium confidence:** Many independent creators report figures, and they agree on how small firms use AI. The level is uncertain because official surveys ask whether the business uses AI, while owner surveys count any use by anyone at the firm, and size cutoffs differ. ## Sources Sources for F-29 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Business Trends and Outlook Survey (BTOS), employment size class estimates | US Census Bureau | 2026 | About 20,000 responses per two-week period; collection ending 4 October 2026 | Official experimental survey; AI used in any business function in the prior two weeks | 1 to 4 employees 23.9%; 5 to 9: 21.5%; 10 to 19: 22.4%; 20 to 49: 24.8% “did this business use Artificial Intelligence (AI) in any of its business functions?” | [census.gov ↗](https://www.census.gov/hfp/btos/data) | | Business insights and impact on the UK economy: 8 October 2026 | Office for National Statistics (UK) | 2026 | UK business survey, wave live 21 September to 4 October 2026 | Official statistics; current use of at least one listed AI technology | 28.7% of businesses with 0 to 9 employees and 36.1% with 10 to 49 used AI “only 29% of businesses with fewer than 10 employees reported using AI technologies” | [ons.gov.uk ↗](https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/bulletins/businessinsightsandimpactontheukeconomy/8october2026) | | Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026 | Statistics Canada | 2026 | 9,251 responding businesses, April to May 2026 | Official survey; AI used to produce goods or deliver services in the past 12 months | 19.9% of businesses with 1 to 4 employees used AI; 41.4% say it is not relevant to them “Among businesses with 1 to 4 employees, about one-fifth (19.9%) reported using AI” | [www150.statcan.gc.ca ↗](https://www150.statcan.gc.ca/n1/pub/11-621-m/11-621-m2026010-eng.htm) | | Small Business and Technology Survey (2025) | NFIB Research Center | 2025 | 521 NFIB member employers, March 2025 | Membership survey with a low response rate; run by a business advocacy group | 24% of small employers use AI; 21% of firms under 10 employees, 48% of firms with 50 or more “ranging from 21% uptake for single-digit employee firms to 48% at 50+ employee firms” | [nfib.com ↗](https://www.nfib.com/news/press-release/new-nfib-report-how-small-businesses-incorporate-tech-and-ai-advancements/) | | Generative AI and the SME Workforce: New Survey Evidence | OECD, survey by Ipsos | 2025 | 5,232 SMEs in seven countries, October to December 2024 | Random telephone survey weighted by number of firms; any staff use of generative AI | 30.7% of SMEs use generative AI; one-person companies 23.6%, firms with 50 to 249 employees 45.8% “It is in use (by the respondent or a colleague) in 31% of SMEs” | [oecd.org ↗](https://www.oecd.org/en/publications/generative-ai-and-the-sme-workforce_2d08b99d-en/full-report/component-4.html) | | 2026 Report on Employer Firms (2025 Small Business Credit Survey) | Federal Reserve Banks | 2026 | 6,525 US employer firms with 1 to 499 employees, September to November 2025 | Weighted convenience sample; counts any use by the business or its employees, including AI inside existing software | 46% of employer firms use AI; of users, 49% experimenting, 44% partly integrated, 7% fully integrated “46% of employer firms reported that their business or employees were using” | [fedsmallbusiness.org ↗](https://www.fedsmallbusiness.org/-/media/project/clevelandfedtenant/fsbsite/reports/2026/2026-report-on-employer-firms/2026-report-on-employer-firms.pdf) | | Empowering Small Business: The Impact of Technology on U.S. Small Business (2026) | US Chamber of Commerce Technology Engagement Center, with Teneo Research | 2026 | Online survey of US small businesses under 250 employees, July to August 2026 | Non-probability online survey by a business lobby group | 66% use some AI tool; 41% a generative AI chatbot “2/3 of small businesses now use AI.” | [uschamber.com ↗](https://www.uschamber.com/assets/documents/Empowering-Small-Business-The-Impact-of-Technology-on-U.S.-Small-Business-Small-Business-Report-2026.pdf) | | 2026 AI Impact Report | Intuit QuickBooks, with economists at the University of Chicago | 2026 | 2,222 US owners of businesses with 0 to 100 employees, January 2026; plus payment data on 4.3 million businesses | Paid panel and Intuit customers; any AI use at least monthly. Intuit sells AI features in QuickBooks. | 77% use AI regularly; in Intuit's payment data, 12% of US businesses paid for a dedicated AI tool between 2021 and 2025 “77% of US businesses now use AI regularly, up from 48% in July 2024.” | [intuit.com ↗](https://www.intuit.com/blog/global-stories/ai-impact-report/) | What for ## What do small businesses use AI for? Writing and marketing first, in every source. Among AI-using US employer firms, 83% use it for writing or marketing, 61% for individual productivity, 51% for planning or analysis, 38% for admin and 19% for coding (Federal Reserve, 2025). Across seven OECD countries, 92% of SME users generate text, and only 29% use generative AI in core business activities. Paying for AI Low confidence ## How many small businesses pay for AI tools, and how much do they spend? About half of the small businesses on Ramp, a spend-management platform: 50.1% of firms with 1 to 24 employees had a paid AI transaction in August 2026, up from 39.7% a year earlier. The median small firm on Ramp spent about $23 per employee per month, roughly $230 a month for a 10-person team. Ramp's customers are more tech-minded than typical small firms, so the share across all small businesses is likely lower: in Census data, only 8 to 9% of AI-using US firms under 50 employees bought software or equipment to use AI. How many small businesses pay for AI tools, and how much do they spend? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Ramp AI Index | Ramp Economics Lab | 2026 | More than 70,000 US businesses on Ramp; small means 1 to 24 employees | Measured card and bill-pay transactions to AI products. Misses free tools and personal accounts. Ramp sells corporate cards and spend software. | 50.1% of small businesses paid for AI in August 2026 (medium 62.5%, large 66.8%); median AI spend $23.23 per employee per month for small firms “Share of businesses with spending on AI models, subscriptions, tokens, and APIs” | [ramp.com ↗](https://ramp.com/data/ai-index) | | BTOS AI Supplement Table 2026 | US Census Bureau | 2026 | More than 117,000 responding US firms, November 2025 to February 2026 | Official survey of what AI-using firms changed in order to use AI | 7.6 to 8.8% of AI-using firms under 50 employees bought computing power, equipment or software; about two thirds made no changes “to use Artificial Intelligence (AI), what changes did this business make?” | [census.gov ↗](https://www.census.gov/hfp/btos/downloads/AI_Supplement_Table_2026.xlsx) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **58% of small businesses use generative AI (US Chamber of Commerce).** That was the 2025 edition. The 2026 edition says 66% use some AI tool, and 41% use a generative AI chatbot. - **77% of small businesses use AI (Intuit).** It counts any AI use at least monthly, including features inside other tools, from a paid panel and Intuit customers. Intuit's payment data show 12% paid for a dedicated AI tool. - **76% of small businesses use AI (Goldman Sachs).** The sample is alumni of a Goldman Sachs growth program for small businesses, who are larger and more growth-minded than typical firms. - **Nearly half of small businesses pay for AI.** That is Ramp's customers with 1 to 24 employees. Ramp's own chart puts the Census estimate of AI use at about 22% for the same month. - **Only 5% of small businesses use AI (Census).** That was the old question, about AI used to produce goods or services. With the current wording, 21 to 25% of firms under 50 employees use AI. ## More from Small Teams - [F-28**96–99%**How much of the economy do small companies make up?](https://evy.so/facts/small-companies-share-of-economy/) - [F-30**13–38 pts**How does AI adoption differ by company size?](https://evy.so/facts/ai-adoption-by-company-size/) - [F-31**54–63%**What stops small companies from adopting AI?](https://evy.so/facts/what-stops-small-companies-adopting-ai/) - [F-32**65–78%**Do small firms that use AI see gains?](https://evy.so/facts/small-firms-gains-from-ai/) - [F-33**45–52 h**How many hours do small business owners work, and how much goes to admin?](https://evy.so/facts/small-business-owner-hours/) [The whole chapter](https://evy.so/facts/small-teams/) · [Chapter poster](https://evy.so/facts/small-teams/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How much of the economy do small companies make up? (96–99%) | The AI at Work Fact Book" description: "Companies under 50 people are 96 to 99 percent of businesses and employ 26 percent of US employer-firm workers and 47 to 48 percent in the EU and UK. Official statistics." last_updated: "2026-10-24" source: "https://evy.so/facts/small-companies-share-of-economy/" -------------------------------------------------------------------------------- F-28 · Small Teams # How much of the economy do small companies make up? Companies with fewer than 50 people are 96 to 99 percent of all businesses in the US, EU and UK. They employ about a quarter of US employer-firm workers (26 percent) and nearly half of people working in the EU and UK business economy (47 to 48 percent, counting the self-employed). 96–99% of businesses have fewer than 50 people; they employ 26% (US) to 48% (EU) of workers - High confidence - 4 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). How much of the economy do small companies make up? The AI at Work Fact Book, F-28. https://evy.so/facts/small-companies-share-of-economy/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/small-companies-share-of-economy.json) [Markdown](https://evy.so/facts/small-companies-share-of-economy.md) ## How to read it Definitions decide the answer. The US Census count covers only firms with paid employees; the EU and UK figures include self-employed owners and one-person firms. That is why the employment share of firms under 50 runs from 26% (US employer firms) to 47 to 48% (UK and EU). The common claim that small businesses employ half of US workers uses the official US definition of small: fewer than 500 employees. ## Every study on one scale Shaded: 96–99% One figure Chart values as text - Scale: % of businesses with fewer than 50 people. Shaded range: 96–99%. - US Census, employer firms (2025): 96.1% (26% of jobs) - SBA, all US firms (2026): 99.3% incl. no-employee firms - Eurostat, EU (2026): 99.1% (48% of jobs) - UK government (2025): 99.2% (47% of jobs) **Why high confidence:** Four official sources built from business registers agree closely on the share of firms. Employment shares differ widely, but the reasons are known: whether the self-employed and one-person firms are counted, and different size cutoffs. ## Sources Sources for F-28 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | 2022 SUSB Annual Data Tables by Enterprise Employment Size | US Census Bureau (Statistics of US Businesses) | 2025 | All 6.4 million US employer firms, 2022 (released April 2025, the latest year) | Official counts from the Business Register; firms with no employees are excluded. Shares are our arithmetic from the size rows. | Firms under 50 employees: 96.1% of employer firms, 25.6% of their employment (34.8 million people); under 10 employees: 79.2% of firms | [census.gov ↗](https://www.census.gov/data/tables/2022/econ/susb/2022-susb-annual.html) | | Frequently Asked Questions About Small Business (February 2026) | US Small Business Administration, Office of Advocacy | 2026 | All US firms including 29.8 million businesses with no employees, 2022 data | Official compilation. Defines small business as fewer than 500 employees; the under-50 share is our arithmetic. | Firms under 500 employees: 99.9% of all firms and 45.9% of private-sector employees; including firms with no employees, firms under 50 are about 99.3% of all firms “45.9% of private sector employees (62.3 million workers)” | [advocacy.sba.gov ↗](https://advocacy.sba.gov/wp-content/uploads/2026/02/FINAL_FAQsAboutSmallBusiness_2026_012826.pdf) | | Enterprise statistics by size class (dataset sbs\_sc\_ovw) | Eurostat (Structural Business Statistics) | 2026 | All 33 million active enterprises in the EU business economy, 2023 | Official statistics from business registers. Persons employed include working owners. Excludes agriculture and the public sector. | Enterprises under 50 persons employed: 99.1% of enterprises, 48.3% of persons employed, 34.9% of value added | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/databrowser/view/sbs_sc_ovw/default/table) | | Business population estimates for the UK and regions 2025 | UK Department for Business and Trade | 2025 | All 5.7 million UK private-sector businesses, start of 2025 | Official statistics in development; includes sole proprietors and partnerships without employees | Businesses with 0 to 49 employees: 99.2% of businesses, 47% of employment, 34% of turnover; 75% of all businesses have no employees “there were 5.64 million small businesses (with 0 to 49 employees), 99.18%” | [gov.uk ↗](https://www.gov.uk/government/statistics/business-population-estimates-2025/business-population-estimates-for-the-uk-and-regions-2025-statistical-release) | Output ## How much do small companies produce? EU firms under 50 people produce 35% of the business economy's value added (2023), and UK businesses under 50 account for 34% of turnover (2025). Using the US definition of small business (under 500 employees), small firms produce 43.5% of US GDP, according to the SBA. ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Small businesses employ half of all American workers.** Firms under 500 employees employ 45.9% of private-sector workers (2022). Firms under 50 employees employ 25.6% of employer-firm workers. - **99.9% of US businesses are small.** It counts firms under 500 employees and includes 29.8 million businesses with no employees. Among firms with employees, 96.1% have fewer than 50. - **SMEs provide 60% of jobs.** That counts firms with up to 249 employees. For companies under 50 people it is about 26% (US employer firms) to 47 to 48% (UK, EU). ## More from Small Teams - [F-29**20–46%**What share of small businesses use AI, and for what?](https://evy.so/facts/small-businesses-using-ai/) - [F-30**13–38 pts**How does AI adoption differ by company size?](https://evy.so/facts/ai-adoption-by-company-size/) - [F-31**54–63%**What stops small companies from adopting AI?](https://evy.so/facts/what-stops-small-companies-adopting-ai/) - [F-32**65–78%**Do small firms that use AI see gains?](https://evy.so/facts/small-firms-gains-from-ai/) - [F-33**45–52 h**How many hours do small business owners work, and how much goes to admin?](https://evy.so/facts/small-business-owner-hours/) [The whole chapter](https://evy.so/facts/small-teams/) · [Chapter poster](https://evy.so/facts/small-teams/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "Do small firms that use AI see gains? (65–78%) | The AI at Work Fact Book" description: "65 to 78 percent of AI-using small firms say AI raised productivity, about 3 in 10 report higher sales, and 96 percent saw no job change. Self-reports, official data and two trials." last_updated: "2026-10-24" source: "https://evy.so/facts/small-firms-gains-from-ai/" -------------------------------------------------------------------------------- F-32 · Small Teams # Do small firms that use AI see gains? Most say yes on productivity: 65 to 78 percent of AI-using small firms report that AI raised productivity or employee performance, by their own account and with no size of gain attached. Far fewer report higher sales, about 3 in 10, and almost none report changes in jobs. 65–78% of AI-using small firms say AI raised productivity (self-reported, no size of gain) - Medium confidence - 2 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). Do small firms that use AI see gains? The AI at Work Fact Book, F-32. https://evy.so/facts/small-firms-gains-from-ai/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/small-firms-gains-from-ai.json) [Markdown](https://evy.so/facts/small-firms-gains-from-ai.md) ## How to read it Every benefit figure here is the owner's or manager's own view, and neither survey asked how large the gains were. The Federal Reserve leaves out firms answering not applicable or unsure; the OECD asks yes or no on a list of benefits and is weighted toward one-person and micro firms. Official surveys measure jobs and tasks, not productivity. No representative source measures the size of small-firm gains. ## Every study on one scale Shaded: 65–78% One figure A range in one study Chart values as text - Scale: % reporting higher productivity. Shaded range: 65–78%. - Federal Reserve (2026): 68–78% productivity - OECD, seven countries (2025): 65.1% performance **Why medium confidence:** Two independent creators with different methods agree on the self-reported share, and two official agencies agree that almost no small firm reports job changes. All the benefit figures are perceptions with no magnitude, and the only causal evidence comes from two trials in special populations that point in different directions. ## Sources Sources for F-32 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | 2026 Report on Employer Firms (2025 Small Business Credit Survey) | Federal Reserve Banks | 2026 | 1,738 to 2,011 AI-using US employer firms, September to November 2025 | Weighted convenience sample; owner-reported direction of change | Productivity up 68% to 78% by size band under 50; sales up 30% to 35%; labor costs unchanged for 75% to 78% “71% said its use led to increased productivity” | [fedsmallbusiness.org ↗](https://www.fedsmallbusiness.org/reports/survey/2026/2026-report-on-employer-firms) | | Generative AI and the SME Workforce | OECD, survey by Ipsos | 2025 | SMEs using generative AI, from 5,232 surveyed in seven countries, late 2024 | Random telephone survey; yes or no on listed benefits, with no size of gain asked | Improved employee performance 65.1%; saved money 45.2%; increased revenue 25.9% “65.1% of SMEs using generative AI report that it helped increase employee performance” | [oecd.org ↗](https://www.oecd.org/en/publications/generative-ai-and-the-sme-workforce_2d08b99d-en/full-report/component-4.html) | Jobs and tasks ## Do small firms cut jobs because of AI? Almost none, so far. 96% of AI-using US firms under 50 employees report no change in employment from AI, and AI mostly supports tasks (43 to 45% of small users) rather than taking them over (8 to 11%), according to the Census Bureau. In the UK, about 4% of AI-using firms under 50 employees report a smaller headcount. Do small firms cut jobs because of AI? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | BTOS AI Supplement Table 2026 | US Census Bureau | 2026 | More than 117,000 responding US firms, November 2025 to February 2026 | Official survey of AI-using firms' effects over the last six months | No change in total employment 95.6% to 96.3% (firms under 50); AI supplemented tasks 42.8% to 44.8%; AI performed tasks previously done by staff 8.4% to 10.9% “how did the use of Artificial Intelligence (AI) affect this business's total employment?” | [census.gov ↗](https://www.census.gov/hfp/btos/downloads/AI_Supplement_Table_2026.xlsx) | Controlled trials Low confidence ## What do controlled trials show about small firms and AI? Few trials exist, and they disagree. In Kenya, 640 small business owners given a GPT-4 business adviser saw no average change in revenue or profit; strong performers gained and weak ones did nearly 10% worse (see F-12). In a trial with 515 early-stage startups, showing founders where other firms use AI led to 1.9 times higher revenue, though only weakly significant, driven by the top tenth of firms, and in a paper not yet peer reviewed. What do controlled trials show about small firms and AI? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Mapping AI into Production: A Field Experiment on Firm Performance (INSEAD Working Paper 2026/20/STR) | Hyunjin Kim and Dahyeon Kim (INSEAD), Rembrand Koning (Harvard Business School) | 2026 | 515 pre-seed startups in an online accelerator; median team of four | Preregistered randomized experiment; working paper, not peer reviewed | Treated startups found 44% more AI use cases and had 1.9 times higher revenue (p < 0.1) “Treated firms complete 12% more tasks, are 18% more likely to acquire paying customers” | [ssrn.com ↗](https://ssrn.com/abstract=6513481) SSRN shows a browser check first; a copy is also hosted by Wharton's Mack Institute. | | The Uneven Impact of Generative Artificial Intelligence on Entrepreneurial Performance (Management Science) | Nicholas G. Otis, Rowan Clarke, Solène Delecourt, David Holtz and Rembrand Koning | 2026 | 640 small business owners in Kenya, 2023 | Randomized controlled trial, published July 2026 | No average effect on revenue or profit; low performers did nearly 10% worse “low performers did nearly 10% worse due to the AI assistant” | [doi.org ↗](https://doi.org/10.1287/mnsc.2024.06909) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **AI boosts small business revenue by 15% (Harvard Kenya study).** The trial found no average effect. Only high performers may have gained over 15%, while low performers did about 10% worse. - **AI nearly doubles startup revenue.** The 1.9 times comes from 515 accelerator startups shown where to use AI, is significant only at p < 0.1, is driven by the top 10% of firms, and is not yet peer reviewed. - **Small businesses are cutting jobs because of AI.** 96% of AI-using US firms under 50 employees report no employment change from AI (Census, 2025 to 2026). ## More from Small Teams - [F-28**96–99%**How much of the economy do small companies make up?](https://evy.so/facts/small-companies-share-of-economy/) - [F-29**20–46%**What share of small businesses use AI, and for what?](https://evy.so/facts/small-businesses-using-ai/) - [F-30**13–38 pts**How does AI adoption differ by company size?](https://evy.so/facts/ai-adoption-by-company-size/) - [F-31**54–63%**What stops small companies from adopting AI?](https://evy.so/facts/what-stops-small-companies-adopting-ai/) - [F-33**45–52 h**How many hours do small business owners work, and how much goes to admin?](https://evy.so/facts/small-business-owner-hours/) [The whole chapter](https://evy.so/facts/small-teams/) · [Chapter poster](https://evy.so/facts/small-teams/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "Small Teams: statistics on small businesses and AI | The AI at Work Fact Book" description: "Chapter 5 of The AI at Work Fact Book. How companies of 2 to 50 people work and use AI: their share of the economy, AI adoption and the gap with large firms, what stops them, what they gain and how long owners work." last_updated: "2026-10-24" source: "https://evy.so/facts/small-teams/" -------------------------------------------------------------------------------- - F-2896–99% ## [How much of the economy do small companies make up?](https://evy.so/facts/small-companies-share-of-economy/) Companies with fewer than 50 people are 96 to 99 percent of all businesses in the US, EU and UK. They employ about a quarter of US employer-firm workers (26 percent) and nearly half of people working in the EU and UK business economy (47 to 48 percent, counting the self-employed). High confidence 4 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/small-companies-share-of-economy/) - F-2920–46% ## [What share of small businesses use AI, and for what?](https://evy.so/facts/small-businesses-using-ai/) Between a fifth and a half of small businesses use AI, 20 to 46 percent in official counts and weighted owner surveys from 2025 to 2026, mostly for writing and marketing. Use is shallow: only 7 percent of AI-using US small firms have fully built it into how they work. Medium confidence 8 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/small-businesses-using-ai/) - F-3013–38 pts ## [How does AI adoption differ by company size?](https://evy.so/facts/ai-adoption-by-company-size/) Large firms (250 or more employees) are 1.4 to 3.2 times as likely to use AI as small firms, a gap of 13 to 38 percentage points in 2025 to 2026 official statistics. In the US the gap starts only at about 50 employees: firms with 1 to 49 employees all sit at 21 to 25 percent. High confidence 4 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/ai-adoption-by-company-size/) - F-3154–63% ## [What stops small companies from adopting AI?](https://evy.so/facts/what-stops-small-companies-adopting-ai/) Mostly the belief that AI does not apply to them: 54 to 63 percent of US firms under 50 employees with no AI plans say so, and 57 percent of non-adopting SMEs in seven OECD countries. Lack of knowledge and privacy worries come next, at about one in five each; cost is cited by 6 to 16 percent. High confidence 3 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/what-stops-small-companies-adopting-ai/) - F-3265–78% ## [Do small firms that use AI see gains?](https://evy.so/facts/small-firms-gains-from-ai/) Most say yes on productivity: 65 to 78 percent of AI-using small firms report that AI raised productivity or employee performance, by their own account and with no size of gain attached. Far fewer report higher sales, about 3 in 10, and almost none report changes in jobs. Medium confidence 2 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/small-firms-gains-from-ai/) - F-3345–52 h ## [How many hours do small business owners work, and how much goes to admin?](https://evy.so/facts/small-business-owner-hours/) Small business owners work about 45 to 52 hours a week, roughly 10 hours more than employees: EU employers worked 46.3 hours in 2025 against 36.6 for employees. Admin takes about a fifth to a third of owner time, by vendor surveys. Medium confidence 4 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/small-business-owner-hours/) -------------------------------------------------------------------------------- title: "Small Teams, on one sheet | The AI at Work Fact Book" description: "Small Teams on one poster: 6 numbers from The AI at Work Fact Book, each with its sources printed on the sheet. Free to reuse under CC BY 4.0." last_updated: "2026-10-24" source: "https://evy.so/facts/small-teams/sheet/" -------------------------------------------------------------------------------- 96 to 99 of 100 F-28 96–99% of businesses have fewer than 50 people US Census 2025 · Eurostat 2026 · UK government 2025 20 to 46 of 100 F-29 20–46% of small businesses use AI, mostly for writing and marketing US Census 2026 · ONS 2026 · OECD 2025 · Federal Reserve 2026 a gap of 13 to 38 points out of 100 F-30 13–38 pts more AI use in large firms than in small ones Eurostat 2025 · US Census 2026 · ONS 2026 54 to 63 of 100 F-31 54–63% of small firms with no AI plans say AI does not apply to them US Census 2026 · Federal Reserve 2026 · OECD 2025 65 to 78 of 100 F-32 65–78% of AI-using small firms say AI raised productivity Federal Reserve 2026 · OECD 2025 45 to 52 hours; the first five rows are a 40-hour week F-33 45–52 h a week worked by small business owners, against about 37 for employees Eurostat 2025 · Time etc 2023 · Sage 2024 How to read it Each 5 by 8 grid is a 40-hour week: a row per day, a square per hour. Each 10 by 10 grid is 100%. Solid orange is the low end of the range across studies, light orange runs to the high end. Methods, samples and links for every study are at **evy.so/facts**. ## The numbers on this sheet - [F-28**96–99%**of businesses have fewer than 50 people; they employ 26% (US) to 48% (EU) of workers. How much of the economy do small companies make up? Sources: US Census, employer firms (2025), SBA, all US firms (2026), Eurostat, EU (2026), UK government (2025).](https://evy.so/facts/small-companies-share-of-economy/) - [F-29**20–46%**of small businesses use AI (official counts and weighted owner surveys; vendor surveys report 66 to 77%). What share of small businesses use AI, and for what? Sources: US Census, 1 to 4 employees (2026), ONS, under 10 employees (2026), Statistics Canada, 1 to 4 (2026), NFIB (2025), OECD, seven countries (2025), Federal Reserve (2026), US Chamber (advocacy) (2026), Intuit (vendor) (2026).](https://evy.so/facts/small-businesses-using-ai/) - [F-30**13–38 pts**more AI use in large firms than small ones; large firms are 1.4 to 3.2 times as likely. How does AI adoption differ by company size? Sources: US Census (2026), ONS, UK (2026), Eurostat, EU (2025), Statistics Canada (2026).](https://evy.so/facts/ai-adoption-by-company-size/) - [F-31**54–63%**of small firms with no AI plans say AI does not apply to their business. What stops small companies from adopting AI? Sources: US Census (2026), Federal Reserve (2026), OECD, seven countries (2025).](https://evy.so/facts/what-stops-small-companies-adopting-ai/) - [F-32**65–78%**of AI-using small firms say AI raised productivity (self-reported, no size of gain). Do small firms that use AI see gains? Sources: Federal Reserve (2026), OECD, seven countries (2025).](https://evy.so/facts/small-firms-gains-from-ai/) - [F-33**45–52 h**a week worked by small business owners; employees work about 37. How many hours do small business owners work, and how much goes to admin? Sources: Eurostat, EU employers (2025), Time etc (vendor) (2023), Sage (vendor) (2024), UENI (vendor) (2024).](https://evy.so/facts/small-business-owner-hours/) Using the sheet in a post, deck or article? Credit "The AI at Work Fact Book by EVY" and link to [evy.so/facts/small-teams/](https://evy.so/facts/small-teams/). -------------------------------------------------------------------------------- title: "Which work tasks do people want AI to do on its own, and which do they want to keep? (2–12%) | The AI at Work Fact Book" description: "About 2 percent of workers want AI to work fully on its own, and up to 12 percent want it to lead. Which tasks people hand to AI and which they keep, with Stanford and KPMG data." last_updated: "2026-10-24" source: "https://evy.so/facts/tasks-workers-want-ai-to-do/" -------------------------------------------------------------------------------- F-24 · What Teams Trust # Which work tasks do people want AI to do on its own, and which do they want to keep? Very few want AI working fully on its own: about 2 percent, in two independent studies. People welcome AI help on many tasks, mostly repetitive ones, and most prefer an equal partnership with AI, while keeping decisions about people, communication and creative control. 2–12% want AI to work alone (about 2%) or to lead (up to 12%) - Medium confidence - 3 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). Which work tasks do people want AI to do on its own, and which do they want to keep? The AI at Work Fact Book, F-24. https://evy.so/facts/tasks-workers-want-ai-to-do/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/tasks-workers-want-ai-to-do.json) [Markdown](https://evy.so/facts/tasks-workers-want-ai-to-do.md) ## How to read it The Stanford study rates everyday tasks across the US workforce; the KPMG question covers only decisions about people. Stanford's 46% is a share of listed tasks, not of work time, and only 7.1% of tasks got a strong vote for automation. All of these are stated preferences, which can differ from what people actually hand to AI. The pattern across sources: people want AI to take repetitive, checkable work and to leave decisions about people, communication and creative control with them. ## Every study on one scale Shaded: 2–12% One figure A range in one study Chart values as text - Scale: % preferring AI alone or AI in the lead. Shaded range: 2–12%. - Shao et al., Stanford (2026): 1.9% of occupations: AI alone - KPMG and Melbourne (2025): 2% AI alone, 12% AI leading **Why medium confidence:** Two independent academic studies with different methods agree that only about 2% want AI acting fully alone. Beyond that the measures differ: one covers everyday tasks, the other only decisions about people, and both are stated preferences. Studies without a figure on this scale are listed in the table only. ## Sources Sources for F-24 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Future of Work with AI Agents: Auditing Automation and Augmentation Potential across the U.S. Workforce (WORKBank) | Yijia Shao, Erik Brynjolfsson, Diyi Yang and colleagues (Stanford University) | 2026 | 1,500 US workers in 104 occupations rating 844 tasks, January to May 2025 | Online survey with audio interviews; preferred level of human involvement per task. Preprint. | AI handles it entirely is the preferred level in 1.9% of occupations; equal partnership in 45.2%; workers favor automation for 46.1% of tasks “For 46.1% of tasks, workers express positive attitudes toward AI agent automation” | [arxiv.org ↗](https://arxiv.org/abs/2506.06576) | | Trust, attitudes and use of artificial intelligence: A global study 2025 | University of Melbourne and KPMG International | 2025 | 32,352 employed people in 47 countries, November 2024 to January 2025 | Most acceptable human-AI split for managerial decisions on hiring, pay, promotion and allocating work | AI 100%: 2%; AI 75%: 10%; equal: 29%; humans 75%: 45%; humans 100%: 14% “Only ten percent believe AI should dominate managerial decision-making” | [mbs.edu ↗](https://mbs.edu/-/media/PDF/Research/Trust_in_AI_Report.pdf) | | 2026 Developer Survey: AI data | Stack Overflow | 2026 | 12,547 developers answering this question, June to August 2026 | Opt-in survey; tasks developers do not use AI for and do not plan to | Writing communications 45.9%; running production systems 36.0%; writing tickets and plans 30.5%; writing tests 8.5% “46% do not use AI there and do not plan to” | [survey.stackoverflow.co ↗](https://survey.stackoverflow.co/2026/ai/data) | What people keep ## Which tasks do people most want to keep for themselves? Tasks involving people, judgment and creative control. Arts, design and media workers favored automation for only 17.1% of their tasks (Stanford). 71% of Americans opposed AI making final hiring decisions (Pew, December 2022). 46% of developers do not use AI for writing communications and do not plan to. Which tasks do people most want to keep for themselves? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | AI in Hiring and Evaluating Workers: What Americans Think | Pew Research Center | 2023 | 11,004 US adults, December 2022 | American Trends Panel, probability-based. An older wave of all adults. | 71% oppose AI making final hiring decisions; 7% favor it “they oppose AI use in making final hiring decisions by a 71%-7% margin” | [pewresearch.org ↗](https://www.pewresearch.org/internet/2023/04/20/ai-in-hiring-and-evaluating-workers-what-americans-think/) | Why automate ## Why do workers want AI to take over some tasks? Mainly to free time for higher-value work (69.4% of the reasons given in the Stanford study), and because tasks are repetitive (46.6%) or AI could improve quality (46.6%). Among workers voicing concerns, the top one was lack of trust in AI's accuracy or reliability (45.0%). ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Workers want AI to automate 46% of their jobs (Stanford).** 46.1% is the share of 844 listed tasks whose average rating leans toward automation, not weighted by time. Only 7.1% got a strong vote. - **75% of employees are comfortable working with AI agents (Workday).** The sample was 2,950 decision-makers and software implementation leaders, in a vendor survey. It does not describe the general workforce. - **Only 10% of employees trust AI to make decisions (KPMG).** The question covered only managerial decisions about people. 10% chose AI with 75% of the say and 2% AI alone; 29% chose an even split. ## More from What Teams Trust - [F-21**24–53%**How much do workers trust the output of AI tools at work?](https://evy.so/facts/trust-in-ai-output/) - [F-22**41–44%**How often do people check AI output before using it?](https://evy.so/facts/checking-ai-output/) - [F-23**17–45%**How often is AI output wrong on real work tasks?](https://evy.so/facts/ai-error-rates-at-work/) - [F-25**30–44%**How many workers use AI in ways that break company policy?](https://evy.so/facts/ai-use-against-company-policy/) - [F-26**42–48%**How often do employees put company data into public AI tools?](https://evy.so/facts/company-data-in-public-ai/) - [F-27**21–29%**How many organizations have had a security incident or data breach involving AI?](https://evy.so/facts/ai-security-breaches/) [The whole chapter](https://evy.so/facts/what-teams-trust/) · [Chapter poster](https://evy.so/facts/what-teams-trust/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How many hours a week do people spend in meetings, and how much of it is wasted? (3–18 h) | The AI at Work Fact Book" description: "Office workers spend 3 to 18 hours a week in meetings, managers about 22, and judge a third to two thirds of it unproductive. Primary sources, plus status meetings." last_updated: "2026-10-24" source: "https://evy.so/facts/time-in-meetings/" -------------------------------------------------------------------------------- F-03 · Where the Week Goes # How many hours a week do people spend in meetings, and how much of it is wasted? Studies range from about 3 to 18 hours a week for the average office worker, and managers report about 22. Workers judge roughly a third to two thirds of their meeting time unnecessary or unproductive. 3–18 h a week in meetings for the average worker; managers about 22 h - Medium confidence - 6 primary sources - 2019 to 2026 - Last checked October 13, 2026 ## Cite this EVY (2026). How many hours a week do people spend in meetings, and how much of it is wasted? The AI at Work Fact Book, F-03. https://evy.so/facts/time-in-meetings/ (last checked October 13, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/time-in-meetings.json) [Markdown](https://evy.so/facts/time-in-meetings.md) ## How to read it The studies count different meetings and different people. ActivTrak only sees time in online tools on work computers, so it is a floor. Microsoft publishes only its heaviest quarter of Teams users. Otter.ai's 18 hours comes from a senior, long-hours group with a median 50-hour week, and likely includes calls and one-to-ones. Seniority matters most: managers of 4 or more people report 22.2 hours against 13.7 for everyone else. Doodle, Otter.ai and Atlassian sell scheduling, transcription or collaboration tools. ## Every study on one scale Shaded: 3–18 h One figure A range in one study Chart values as text - Scale: hours a week in meetings. Shaded range: 3–18 h. - Doodle (2019): 3 h - ActivTrak (2025): about 3.2 h (online only) - Deming et al., Norway (2026): 4.7 h (12%) - Atlassian (2022): 5–8 h - Microsoft (top 25%) (2023): 7.5 h - Otter.ai / Rogelberg (2022): 18 h (managers 22.2 h) **Why medium confidence:** Six primary studies measure hours, and they differ by a factor of six because they count different meetings, people and methods. Every figure on wasted time is self-reported. ## Sources Sources for F-03 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | The Doodle State of Meetings Report 2019 | Doodle | 2019 | About 6,500 professionals in the UK, Germany and the US, plus data on 19 million meetings scheduled in Doodle | Survey with Nielsen plus scheduling data, 2018. Doodle sells meeting scheduling. | 3 hours a week in meetings, 2 of them considered pointless “The average professional spends three hours a week in meetings” | [assets.ctfassets.net ↗](https://assets.ctfassets.net/p24lh3qexxeo/axrPjsBSD1bLp2HYEqoij/d2f08c2aaf5a6ed80ee53b5ad7631494/Meeting_Report_2019.pdf) | | Productivity Benchmarks Report 2025 | ActivTrak Productivity Lab | 2025 | 218,900 employees at 777 companies using ActivTrak | Observed time in online meeting and chat tools on monitored work computers, 2022 to 2024. Leaves out in-person meetings and calls away from the computer. | About 38 minutes a day in digital collaboration tools (derived from the report's sector comparison) “Collaboration time (49 mins) is 11 mins longer than average” | [activtrak.com ↗](https://www.activtrak.com/blog/productivity-benchmarks/) | | Meetings (NBER Working Paper 35706) | David Deming, Katrine Løken, Alexander Willén and Yaling Xu | 2026 | About 9,000 workers in Norway, linked to employer records | Survey in 2025, self-reported hours in bands | Meetings take 12% of work hours, 4.7 hours a week “consuming an average of 12 percent of work hours” | [nber.org ↗](https://www.nber.org/papers/w35706) | | The State of Teams 2022 | Atlassian Teamwork Lab | 2022 | 1,710 knowledge workers in the US and Australia | Online survey, July 2022, self-reported | About 5 hours a week for office-only teams, about 8 hours for hybrid and distributed teams “office-only workers have roughly five hours of meetings in an average week” | [atlassian.com ↗](https://www.atlassian.com/blog/state-of-teams-2022) | | Will AI Fix Work? (2023 Work Trend Index) | Microsoft | 2023 | Anonymized Microsoft 365 usage data, commercial customers | Observed Teams meeting time. Reports only the heaviest quarter of meeting users. | 7.5 hours a week for the heaviest 25% of Teams meeting users “the heaviest meeting users (top 25%) spend 7.5 hours a week in meetings” | [microsoft.com ↗](https://www.microsoft.com/en-us/worklab/work-trend-index/will-ai-fix-work) | | The Cost of Unnecessary Meeting Attendance | Otter.ai with Steven G. Rogelberg (UNC Charlotte) | 2022 | 632 employees in 20+ industries, median 50-hour week | Voluntary online survey, summer 2022, self-reported. Otter.ai sells meeting transcription. | 18 hours a week in 17.7 meetings; non-managers 13.7 hours, managers of 4 or more people 22.2 hours “On average, employees have 17.7 meetings (totaling 18 hours) per week.” | [public.otter.ai ↗](https://public.otter.ai/reports/The_Cost_of_Unnecessary_Meeting_Attendance.pdf) | Wasted time ## How much meeting time do workers consider wasted? Roughly 30 to 67 percent of it, by workers' own judgment. In the Otter.ai study, employees said they could skip about 30% of their meetings (5.7 of 18 hours). Doodle's respondents called 2 of their 3 weekly meeting hours pointless. Asana's 2024 survey puts unproductive meetings at 5 hours a week, double the 2019 level. Widely shared figures such as 71% of senior managers calling meetings unproductive (HBR, 2017) measure opinion about meetings and give no share of time. How much meeting time do workers consider wasted? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | State of Work Innovation 2024 | Asana Work Innovation Lab | 2024 | 13,066 knowledge workers in six countries | Online surveys, February to August 2024, self-reported | 5 hours a week in unproductive meetings, double the August 2019 level; 9 hours in France and Germany “doubled to 5 hours per week since August 2019” | [asana.com ↗](https://asana.com/resources/state-of-work-innovation) | | Stop the Meeting Madness (Harvard Business Review) | Leslie A. Perlow, Constance Noonan Hadley and Eunice Eun | 2017 | 182 senior managers across industries | Authors' survey of opinions | 71% said meetings are unproductive and inefficient; 65% said meetings keep them from their own work “71 percent said meetings are unproductive and inefficient” | [hbr.org ↗](https://hbr.org/2017/07/stop-the-meeting-madness) | Status meetings Low confidence ## How much time goes to status meetings? About 4 hours a week for people who attend them, but only one study measures it, so treat this as indicative. In a 2016 Harris Poll survey for Clarizen, US workers who attend status meetings spend 4.1 hours a week in project status meetings and 3.9 hours in general ones, and those who prepare for them spend another 3.8 hours. Each average covers a different group (56 to 62 percent of respondents). In a 2015 Wrike survey, 46% said most or all of their meetings were just status updates. No study we found measures time spent writing or reading written status updates. How much time goes to status meetings? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Clarizen survey on status meetings | Clarizen, survey by The Harris Poll | 2016 | 1,120 employed US adults (plus UK and Australia samples) | Online survey, January 2016, self-reported. Clarizen sells work management software. | 4.1 hours a week in project status meetings (61% attend), 3.9 hours in general status meetings (62%), 3.8 hours preparing (56%) “spending 4.1 hours each week in these meetings” | [firmenpresse.de ↗](https://www.firmenpresse.de/pressrelease449502/clarizen-survey-says-employees-in-the-u-s-waste-up-to-30-percent-of-work-week-on-status-meetings.html) Press release as republished by Marketwired; the Clarizen site no longer hosts it. | | 2015 Work Management Survey | Wrike | 2015 | More than 1,400 office workers | Wrike's own survey, cited on its company blog. Wrike sells work management software. | 46% said most or all of their meetings were focused on status updates “most or all of their meetings were focused on just status updates” | [wrike.com ↗](https://www.wrike.com/blog/accurate-real-time-insights-with-one-click/) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Executives spend nearly 23 hours a week in meetings, up from under 10 in the 1960s.** Repeated from article to article without an original study we could find. Otter.ai and Steven Rogelberg's 2022 survey found 22.2 hours for managers of 4 or more people, which is close. - **Atlassian: 3 in 4 meetings are ineffective ("72% of the time").** The 72% averages six ratings, each the share of respondents who call meetings ineffective for one purpose, such as brainstorming. It does not count meetings. - **Clarizen: workers spend 11.8 hours a week, up to 30% of the week, on status meetings.** 11.8 adds three averages that each cover a different group of respondents. The parts are 4.1, 3.9 and 3.8 hours, for people who attend or prepare for status meetings. - **The average employee spends 6 hours a week in scheduled meetings (MIT Sloan Management Review, 2007).** The article calls this a conservative estimate from the literature and names no underlying study. ## More from Where the Week Goes - [F-01**19–25%**How much of the work week do knowledge workers spend searching for information?](https://evy.so/facts/time-spent-searching-for-information/) - [F-02**53–61%**How much of the work day goes to coordinating work rather than doing it?](https://evy.so/facts/time-on-work-about-work/) - [F-04**20–26 min**How long does it take to get back to a task after an interruption?](https://evy.so/facts/time-to-resume-after-interruption/) - [F-05**13–28%**How much of the work week do knowledge workers spend on email and chat?](https://evy.so/facts/time-on-email-and-chat/) - [F-06**3–4 h**How much time do people lose switching between apps at work?](https://evy.so/facts/time-lost-switching-apps/) - [F-07**4.6–5.7 h**How much time do people spend redoing work that already exists?](https://evy.so/facts/time-spent-duplicating-work/) [The whole chapter](https://evy.so/facts/where-the-week-goes/) · [Chapter poster](https://evy.so/facts/where-the-week-goes/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How much time do people lose switching between apps at work? (3–4 h) | The AI at Work Fact Book" description: "People lose about 3 to 4 hours a week reorienting after switching apps. How many switches and apps a day, with sources and method." last_updated: "2026-10-24" source: "https://evy.so/facts/time-lost-switching-apps/" -------------------------------------------------------------------------------- F-06 · Where the Week Goes # How much time do people lose switching between apps at work? About 3 to 4 hours a week goes to getting back on track after switching apps. Logging software (just under 4 hours) and self-reports (36 minutes a day) agree within about an hour. 3–4 h a week lost getting back on track after switching apps - Medium confidence - 2 primary sources - 2021 to 2022 - Last checked October 13, 2026 ## Cite this EVY (2026). How much time do people lose switching between apps at work? The AI at Work Fact Book, F-06. https://evy.so/facts/time-lost-switching-apps/ (last checked October 13, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/time-lost-switching-apps.json) [Markdown](https://evy.so/facts/time-lost-switching-apps.md) ## How to read it The HBR figure comes from logging software at three Fortune 500 companies, mostly in operations-style teams, so it may run high for other roles. Qatalog's is self-reported; 36 minutes a day is about 3 hours over a 5-day week. Counts of switches and counts of apps measure different things, so they are shown separately below and kept out of the range. ## Every study on one scale Shaded: 3–4 h One figure Chart values as text - Scale: hours a week lost after switching. Shaded range: 3–4 h. - Murty et al., HBR (2022): just under 4 h (9%) - Qatalog / Cornell (2021): 36 min a day (3 h) **Why medium confidence:** Two independent studies, one logged and one self-reported, agree within about an hour. Counts of switches per day differ by a factor of 50 between sources, so they are kept out of the range. ## Sources Sources for F-06 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | How Much Time and Energy Do We Waste Toggling Between Applications? (Harvard Business Review) | Rohan Narayana Murty, Sandeep Dadlani and Rajath B. Das (Soroco) | 2022 | 137 users in 20 teams at three Fortune 500 companies | Logging software for up to five weeks, observed. Soroco sells work-pattern software. | About 1,200 toggles a day; just under 4 hours a week reorienting, about 9% of time at work “just under four hours a week reorienting themselves after toggling” | [hbr.org ↗](https://hbr.org/2022/08/how-much-time-and-energy-do-we-waste-toggling-between-applications) | | Workgeist Report '21 | Qatalog with Cornell University's Ellis Idea Lab | 2021 | Three surveys of about 1,000 remote workers each, US and UK | Online surveys, March 2021, self-reported. Qatalog sells a work hub. | 36 minutes a day switching between apps; 9.5 minutes to get back into a good workflow “taking 9 and a half minutes on average to get back into a good workflow” | [qatalog.com ↗](https://qatalog.com/) [Archived copy ↗](https://web.archive.org/web/20210624191022/https://assets.qatalog.com/language.work/qatalog-2021-workgeist-report.pdf) Qatalog is now part of ClickUp and no longer hosts the report; the archived copy is the original PDF. | Switches a day ## How many times a day do people switch between apps? It depends on who counts. People report about 25 switches a day between 10 apps (Asana, 2021). Logging software records far more: 391 to 661 window switches a day (Mark and colleagues, 2014) and about 1,200 toggles between apps and websites (Harvard Business Review, 2022). Self-reported counts are likely too low. How many times a day do people switch between apps? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Anatomy of Work Index 2021 | Asana, survey by Sapio Research | 2021 | 13,123 knowledge workers in seven countries | Online survey, October 2020, self-reported | 10 apps and 25 switches between them a day (US: 13 apps, 30 switches) “on average people switch between 10 apps 25 times per day” | [asana.com ↗](https://asana.com/resources/anatomy-of-work-index) [Archived copy ↗](https://web.archive.org/web/20210301155022/https://resources.asana.com/rs/784-XZD-582/images/PDF-FY21-Global-EN-Anatomy%20of%20Work%20Report.pdf) Archived copy of the full 2021 report PDF. | | Bored Mondays and Focused Afternoons (ACM CHI 2014) | Gloria Mark, Shamsi Iqbal, Mary Czerwinski and Paul Johns | 2014 | 32 information workers at one US company, 5 days each | Logging software plus experience sampling, observed. Peer-reviewed. | 661 window switches a day on Mondays, 391 on Fridays | [dl.acm.org ↗](https://dl.acm.org/doi/abs/10.1145/2556288.2557204) [Archived copy ↗](https://web.archive.org/web/20240705042331/https://dl.acm.org/doi/abs/10.1145/2556288.2557204) | Apps in use ## How many apps does a knowledge worker use in a day? About 9 to 10, self-reported: 10 in Asana's 2021 survey and 9 in 2022, with 13 in the US. Companies deploy far more: Okta's customers average 101 apps each, and 74 in Europe, the Middle East and Africa. How many apps does a knowledge worker use in a day? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Anatomy of Work Index 2022 | Asana, survey by GWI | 2022 | 10,624 knowledge workers in seven countries | Online survey, October 2021, self-reported | 9 apps a day, down from 10 “workers are using slightly fewer apps per day (9) than last year (10)” | [businesswire.com ↗](https://www.businesswire.com/news/home/20220405005399/en/Asana-Anatomy-of-Work-Index-2022-Work-About-Work-Hampering-Organizational-Agility) [Archived copy ↗](https://web.archive.org/web/20260411014513/https://www.businesswire.com/news/home/20220405005399/en/Asana-Anatomy-of-Work-Index-2022-Work-About-Work-Hampering-Organizational-Agility) | | Businesses at Work 2025 | Okta | 2025 | Anonymized data from thousands of Okta customers | Apps connected per company in Okta's network, 2024 data. A company-level count. | 101 apps per company on average; 74 in Europe, the Middle East and Africa “The average number of apps each company uses reached 101 this year” | [okta.com ↗](https://www.okta.com/blog/2025/03/businesses-at-work-2025) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **It takes 23 minutes 15 seconds to refocus after switching apps.** That figure comes from research on interruptions, and it measures the time until people return to interrupted work. See F-04. - **The average person toggles more than 3,600 times a day.** We found no study behind it. The largest logged count we found is about 1,200 toggles a day (HBR, 2022). - **Slack: workers switch between 10 or more apps a day and lose nearly 100 minutes.** It appears on a Slack product page citing unnamed research. We could not find the study. ## More from Where the Week Goes - [F-01**19–25%**How much of the work week do knowledge workers spend searching for information?](https://evy.so/facts/time-spent-searching-for-information/) - [F-02**53–61%**How much of the work day goes to coordinating work rather than doing it?](https://evy.so/facts/time-on-work-about-work/) - [F-03**3–18 h**How many hours a week do people spend in meetings, and how much of it is wasted?](https://evy.so/facts/time-in-meetings/) - [F-04**20–26 min**How long does it take to get back to a task after an interruption?](https://evy.so/facts/time-to-resume-after-interruption/) - [F-05**13–28%**How much of the work week do knowledge workers spend on email and chat?](https://evy.so/facts/time-on-email-and-chat/) - [F-07**4.6–5.7 h**How much time do people spend redoing work that already exists?](https://evy.so/facts/time-spent-duplicating-work/) [The whole chapter](https://evy.so/facts/where-the-week-goes/) · [Chapter poster](https://evy.so/facts/where-the-week-goes/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How much of the work week do knowledge workers spend on email and chat? (13–28%) | The AI at Work Fact Book" description: "Email takes 13 to 28 percent of the work week, and Teams chat another 19 percent of measured time in Microsoft 365. Five primary studies, sources and method." last_updated: "2026-10-24" source: "https://evy.so/facts/time-on-email-and-chat/" -------------------------------------------------------------------------------- F-05 · Where the Week Goes # How much of the work week do knowledge workers spend on email and chat? Email alone takes 13 to 28 percent of the work week, depending on how it is measured. Microsoft's usage data adds Teams chat at 19 percent of measured time in its apps. 13–28% of the work week on email alone - Medium confidence - 5 primary sources - 2012 to 2025 - Last checked October 13, 2026 ## Cite this EVY (2026). How much of the work week do knowledge workers spend on email and chat? The AI at Work Fact Book, F-05. https://evy.so/facts/time-on-email-and-chat/ (last checked October 13, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/time-on-email-and-chat.json) [Markdown](https://evy.so/facts/time-on-email-and-chat.md) ## How to read it The studies fall into two groups. Workfront's self-reports and Microsoft's app data land at 13 to 15 percent; McKinsey's model, Adobe's self-reports and Microsoft Research's logs land at 26 to 29 percent. The denominators differ: the whole week, time inside Microsoft apps, or an assumed 40-hour week. Adobe's and Microsoft Research's percentages are our conversion at 40 hours, so they read high for people who work longer weeks. Microsoft Research counts whole email sessions, which is broader than Microsoft's count of deliberate actions. ## Every study on one scale Shaded: 13–28% One figure A range in one study Chart values as text - Scale: % of the work week on email. Shaded range: 13–28%. - Workfront (2019): 13–15% - Microsoft (2023): 15% (of time in apps) - Adobe (2019): 209 min a day (about 26%) - Dillon et al. (Microsoft Research) (2025): 11.7 h (about 29% of 40 h) - McKinsey Global Institute (2012): 28% **Why medium confidence:** Five primary studies, but they split into two groups (about 13 to 15 percent and about 26 to 29 percent) because they use different denominators and definitions of email time. ## Sources Sources for F-05 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | State of Work 2020 | Workfront, survey by Regina Corso Consulting | 2019 | 3,750 employees at companies with 500 or more staff in the US, UK, Germany and the Netherlands | Online survey, July 2019; people split their week into categories adding up to 100% | Email takes 15% of the week (US, UK), 14% (Netherlands), 13% (Germany) “Best guess for what percentage of your work week is taken up by each” | [business.adobe.com ↗](https://business.adobe.com/products/workfront.html) [Archived copy ↗](https://web.archive.org/web/20201003135606/https://www.workfront.com/sites/default/files/resource/file_pdf/2019-09/sow-report-2020.pdf) Workfront is now part of Adobe and no longer hosts the report; the archived copy is the original PDF. | | Will AI Fix Work? (2023 Work Trend Index) | Microsoft | 2023 | Microsoft 365 usage data from commercial customers, 28 days to March 2023 | App usage data, observed. Share of time inside Microsoft 365 apps, which is not the whole week. | Email 15%, Teams chat 19% and Teams meetings 23% of measured time “15% in email, forming 57% of the time spent communicating” | [microsoft.com ↗](https://www.microsoft.com/en-us/worklab/work-trend-index/will-ai-fix-work) | | Adobe Email Usage Study 2019 | Adobe | 2019 | About 1,000 office workers per country in the US, UK, Germany, France, Australia, India and Japan | Online survey, July to August 2019, self-reported minutes checking work email | 209 minutes per weekday on work email in the US (Germany 173, UK 207, France 216) “number of minutes you spend each weekday checking your email” | [slideshare.net ↗](https://www.slideshare.net/slideshow/2019-adobe-email-report/179705093) [Archived copy ↗](https://web.archive.org/web/20250129020143/https://www.slideshare.net/slideshow/2019-adobe-email-report/179705093) Adobe's own results deck, hosted on SlideShare. Adobe's blog post on the study covers the US only. | | Shifting Work Patterns with Generative AI (NBER Working Paper 33795) | Dillon, Jaffe, Immorlica and Stanton (Microsoft Research and Harvard Business School) | 2025 | 7,137 knowledge workers at 66 firms | Field experiment using Microsoft 365 usage data, 2023 to 2024; this is the baseline before the AI tool | 11.7 hours a week reading and responding to email “the average worker spent 11.7 hours per week reading and responding to emails” | [arxiv.org ↗](https://arxiv.org/abs/2504.11436) | | The social economy: Unlocking value and productivity through social technologies | McKinsey Global Institute, using IDC survey data | 2012 | Managers and professionals in IDC's 2009 survey; sample size not stated | Modeled estimate from McKinsey and IDC data | 28% of work time reading, writing or answering email “This includes 28 percent of work time reading, writing, or responding to e-mail” | [mckinsey.com ↗](https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-social-economy) [Archived copy ↗](https://web.archive.org/web/20260710105837/https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-social-economy) | Chat ## How much time goes to chat at work? One source measures it. In Microsoft's 2023 usage data, Teams chat takes 19% of measured time in Microsoft 365 apps, against 15% for email and 23% for Teams meetings. Email and chat together are 34% of that measured time. The heaviest quarter of email users spend 8.8 hours a week on email. ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Adobe: people spend 5 hours a day on email.** That adds work email (209 minutes a weekday, US) to personal email (143 minutes). Work email alone is about 3.5 hours a day, self-reported. - **McKinsey: email takes 13 hours a week.** The report gives 28 percent of work time. We could not find the 13-hour conversion in the report itself. - **Slack: 25% of workers spend too much time on email.** This is an opinion on whether email takes too much time. It gives no hours. ## More from Where the Week Goes - [F-01**19–25%**How much of the work week do knowledge workers spend searching for information?](https://evy.so/facts/time-spent-searching-for-information/) - [F-02**53–61%**How much of the work day goes to coordinating work rather than doing it?](https://evy.so/facts/time-on-work-about-work/) - [F-03**3–18 h**How many hours a week do people spend in meetings, and how much of it is wasted?](https://evy.so/facts/time-in-meetings/) - [F-04**20–26 min**How long does it take to get back to a task after an interruption?](https://evy.so/facts/time-to-resume-after-interruption/) - [F-06**3–4 h**How much time do people lose switching between apps at work?](https://evy.so/facts/time-lost-switching-apps/) - [F-07**4.6–5.7 h**How much time do people spend redoing work that already exists?](https://evy.so/facts/time-spent-duplicating-work/) [The whole chapter](https://evy.so/facts/where-the-week-goes/) · [Chapter poster](https://evy.so/facts/where-the-week-goes/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How much of the work day goes to coordinating work rather than doing it? (53–61%) | The AI at Work Fact Book" description: "Knowledge workers spend 53 to 61 percent of their time coordinating work instead of doing skilled work. Four independent creators, sources and method." last_updated: "2026-10-24" source: "https://evy.so/facts/time-on-work-about-work/" -------------------------------------------------------------------------------- F-02 · Where the Week Goes # How much of the work day goes to coordinating work rather than doing it? Between 53 and 61 percent, or roughly three of every five working hours. Four independent creators (Asana, McKinsey with IDC data, Microsoft and Workfront) land in that band, though each draws the line around coordination a little differently. 53–61% of working time, from 4 independent creators - High confidence - 9 primary sources - 2012 to 2025 - Last checked October 13, 2026 ## Cite this EVY (2026). How much of the work day goes to coordinating work rather than doing it? The AI at Work Fact Book, F-02. https://evy.so/facts/time-on-work-about-work/ (last checked October 13, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/time-on-work-about-work.json) [Markdown](https://evy.so/facts/time-on-work-about-work.md) ## How to read it Each study draws the line differently. Asana counts self-reported work about work, which includes searching, switching apps and chasing status. McKinsey counts everything outside role-specific tasks, including email. Microsoft compares time in communication apps with time in creation apps, so it only sees work done inside Microsoft 365. Workfront asks how much of the day goes to the job people were hired for. Asana's six editions come from one creator, so they count once toward agreement, and the 2024 change of label and panel means the series is not one clean trend. ## Every study on one scale Shaded: 53–61% One figure Chart values as text - Scale: % of working time. Shaded range: 53–61%. - Asana 2019 (2019): 60% - Asana 2021 (2021): 60% - Asana 2022 (2022): 58% - Asana 2023 (2023): 58% - Asana 2024 (2024): 53% - Asana 2025 (2025): 55% - McKinsey Global Institute (2012): 61% - Microsoft (2024): 60% (observed) - Workfront (2018): 60% **Why high confidence:** Four independent creators using surveys, adjusted survey data and app usage data converge on 53 to 61 percent. Definitions differ, and three of the four are self-reported or vendor-built. ## Sources Sources for F-02 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Anatomy of Work Index 2019 | Asana, survey by Sapio Research | 2019 | 10,223 knowledge workers in Australia, New Zealand, Germany, Japan, the UK and the US | Online survey, August to September 2019, self-reported. Asana sells work management software. | 60% work about work, 27% skilled work, 13% strategic work “The average knowledge worker devotes 60% of their time to work about work” | [resources.asana.com ↗](https://resources.asana.com/rs/784-XZD-582/images/Anatomy-of-Work-Index.pdf) | | Anatomy of Work Index 2021 | Asana, survey by Sapio Research | 2021 | 13,123 knowledge workers in seven countries | Online survey, October 2020, self-reported | 60% work about work (France 66%) “knowledge workers continue to spend 60% of their day on work about work” | [asana.com ↗](https://asana.com/resources/anatomy-of-work-index) [Archived copy ↗](https://web.archive.org/web/20210301155022/https://resources.asana.com/rs/784-XZD-582/images/PDF-FY21-Global-EN-Anatomy%20of%20Work%20Report.pdf) Archived copy of the full 2021 report PDF. | | Anatomy of Work Index 2022 | Asana, survey by GWI | 2022 | 10,624 knowledge workers in seven countries | Online survey, October 2021, self-reported | 58% work about work, 33% skilled work, 9% strategic work “individuals are dedicating 33% of their day to skilled work” | [businesswire.com ↗](https://www.businesswire.com/news/home/20220405005399/en/Asana-Anatomy-of-Work-Index-2022-Work-About-Work-Hampering-Organizational-Agility) [Archived copy ↗](https://web.archive.org/web/20260411014513/https://www.businesswire.com/news/home/20220405005399/en/Asana-Anatomy-of-Work-Index-2022-Work-About-Work-Hampering-Organizational-Agility) | | Anatomy of Work Global Index 2023 | Asana, survey by GWI | 2023 | 9,615 knowledge workers in the US, UK, Australia, France, Germany and Japan | Online survey, November 2022, self-reported | 58% work about work | [investors.asana.com ↗](https://investors.asana.com/news-releases/news-release-details/asana-anatomy-work-global-index-2023-smart-collaboration-and) [Archived copy ↗](https://web.archive.org/web/20260407093353/https://investors.asana.com/news-releases/news-release-details/asana-anatomy-work-global-index-2023-smart-collaboration-and) | | State of Work Innovation 2024 | Asana Work Innovation Lab | 2024 | 13,066 knowledge workers in the US, UK, Germany, France, Japan and Australia | Online surveys, February to August 2024, self-reported. New label (busywork) and new survey panels. | 53% of time on busywork “more than half (53%) of workers' time is swallowed up by busywork” | [asana.com ↗](https://asana.com/resources/state-of-work-innovation) | | State of AI at Work 2025 | Asana Work Innovation Lab | 2025 | 9,236 knowledge workers in the US, UK, Germany, Japan and Australia | Online surveys, February to August 2025, self-reported | 55% of time on busywork “Knowledge workers spend 55% of their time on busywork” | [asana.com ↗](https://asana.com/resources/state-of-ai-work) | | The social economy: Unlocking value and productivity through social technologies | McKinsey Global Institute, using IDC survey data | 2012 | Managers and professionals in IDC's 2009 survey; sample size not stated | Re-analysis of IDC's self-reported time-use survey | 61% of the week outside role-specific work: 28% email, 19% searching, 14% internal communication | [mckinsey.com ↗](https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-social-economy) [Archived copy ↗](https://web.archive.org/web/20260710105837/https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-social-economy) | | 2024 Work Trend Index Annual Report | Microsoft | 2024 | Anonymized usage data from commercial Microsoft 365 customers, 28 days to March 2024, outside the EU | App usage data, observed. Measures time inside Microsoft 365 apps only. | 60% of time in emails, chats and meetings; 40% in creation apps such as Word and PowerPoint “users spend 60% of their time on emails, chats, and meetings” | [assets-c4akfrf5b4d3f4b7.z01.azurefd.net ↗](https://assets-c4akfrf5b4d3f4b7.z01.azurefd.net/assets/2024/05/Microsoft_2024_Work_Trend_Index_Annual_Report_663b79bdc8f91.pdf) | | 2019 US State of Work | Workfront, survey by Regina Corso Consulting | 2018 | 2,010 US office workers at companies with 500 or more staff | Online survey, June 2018, self-reported. Workfront sells work management software. | 40% of the workday on primary tasks, so 60% on everything else “U.S. workers say they spend only 40% of their workday on their primary tasks” | [business.adobe.com ↗](https://business.adobe.com/products/workfront.html) [Archived copy ↗](https://web.archive.org/web/20190907022647/https://www.workfront.com/sites/default/files/files/2019-01/2019%20US%20State%20of%20Work.pdf) Workfront is now part of Adobe and no longer hosts the report; the archived copy is the original PDF. | Year by year ## Is work about work going down? Slightly, in Asana's series: 60% (2019), 60% (2021), 58% (2022), 58% (2023), then 53% (2024) and 55% (2025). From 2024 Asana relabeled it busywork and changed survey panels, so the last two years are not directly comparable. In 2019, respondents guessed they spent only 35% of their time on it. ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Asana 2024: 54% (or 60%) of time goes to work about work.** Asana's 2024 report says 53% of time goes to busywork. The 54% in that report is the share of workers using AI at least monthly, and 60% comes from the 2019 and 2021 editions. - **Asana's 60% quoted as this year's figure.** 60% is from the 2019 and 2021 editions. The latest figures are 53% (2024) and 55% (2025), under the new label busywork. ## More from Where the Week Goes - [F-01**19–25%**How much of the work week do knowledge workers spend searching for information?](https://evy.so/facts/time-spent-searching-for-information/) - [F-03**3–18 h**How many hours a week do people spend in meetings, and how much of it is wasted?](https://evy.so/facts/time-in-meetings/) - [F-04**20–26 min**How long does it take to get back to a task after an interruption?](https://evy.so/facts/time-to-resume-after-interruption/) - [F-05**13–28%**How much of the work week do knowledge workers spend on email and chat?](https://evy.so/facts/time-on-email-and-chat/) - [F-06**3–4 h**How much time do people lose switching between apps at work?](https://evy.so/facts/time-lost-switching-apps/) - [F-07**4.6–5.7 h**How much time do people spend redoing work that already exists?](https://evy.so/facts/time-spent-duplicating-work/) [The whole chapter](https://evy.so/facts/where-the-week-goes/) · [Chapter poster](https://evy.so/facts/where-the-week-goes/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How much time does AI save knowledge workers? (3–5.4%) | The AI at Work Fact Book" description: "Workers who use AI say it saves 3 to 5.4 percent of their work hours, about 1 to 2 hours a week. Self-reports, all-worker figures and the measured evidence, with sources." last_updated: "2026-10-24" source: "https://evy.so/facts/time-saved-by-ai-at-work/" -------------------------------------------------------------------------------- F-08 · What AI Saves # How much time does AI save knowledge workers? Workers who use generative AI say it saves them about 3 to 5.4 percent of their work hours, roughly 1 to 2 hours of a 40-hour week. The one large trial that measured time found users spent 1.4 to 2 fewer hours a week on email, with no measurable change in time in meetings or writing documents. 3–5.4% of work hours, by the estimate of people who use AI (about 1 to 2 hours a week) Across all 6 studies: about 1 to 9 hours a week, counting vendor surveys. - Medium confidence - 6 primary sources - 2024 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). How much time does AI save knowledge workers? The AI at Work Fact Book, F-08. https://evy.so/facts/time-saved-by-ai-at-work/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/time-saved-by-ai-at-work.json) [Markdown](https://evy.so/facts/time-saved-by-ai-at-work.md) ## How to read it Every figure here is a worker's own estimate of how much longer the work would have taken without AI. The surveys turn answers into numbers differently and ask different people, from national panels to a vendor's own customers. Figures for AI users run three to four times higher than figures across all workers, because in any given week most workers do not use AI. Hours assume a 40-hour week. Microsoft, Atlassian and Workday sell AI products. ## Every study on one scale Shaded: 3–5.4% of hours Dashed: with vendor surveys One figure A range in one study Chart values as text - Scale: hours a week saved (self-reported). Shaded range: 3–5.4% of hours; dashed, with vendor surveys: 1.2 h to 8.75 h. - Bick, Blandin and Deming (2025): 2.2 h (5.4%) - Humlum and Vestergaard, Denmark (2026): 1.2 h (3%) - UK civil service (2025): 26 min a day (2.2 h) - Microsoft (2024): 14 min a day (1.2 h) - Atlassian (2024): 53–105 min a day (4.4–8.75 h) - Workday (2026): 1–7 h (85% of users) **Why medium confidence:** Three independent creators with different surveys (a US national panel, a Danish survey linked to government records, and the UK civil service) agree on the range. All of them are self-reports, and the only large measured trial covers email time only. ## Sources Sources for F-08 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | The Rapid Adoption of Generative AI (NBER Working Paper 32966) | Alexander Bick (St. Louis Fed), Adam Blandin (Vanderbilt) and David Deming (Harvard) | 2025 | More than 10,000 US adults aged 18 to 64 in the Real-Time Population Survey; time question asked in November 2024 | Nationally representative online survey. Asks how many extra hours last week's work would have taken without AI. | 5.4% of work hours among people who used generative AI for work, about 2.2 hours of a 40-hour week; 1.4% across all workers “We estimate a mean time savings of 5.4 percent among all genAI work users,” | [nber.org ↗](https://www.nber.org/papers/w32966) | | Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI (NBER Working Paper 33777) | Anders Humlum (Chicago Booth) and Emilie Vestergaard (University of Copenhagen), with Statistics Denmark | 2026 | About 25,000 workers at about 7,000 workplaces in 11 occupations exposed to AI chatbots | Surveys linked to Danish payroll records. Time saved is self-reported in bands; earnings and paid hours come from the records. First circulated as "Large Language Models, Small Labor Market Effects". | About 3% of work hours for people who use AI chatbots (0.7% for teachers up to 6.1% for marketing staff); no change in earnings or paid hours in the records “On average, adopters in our sample report savings of about 3% of their work hours.” | [nber.org ↗](https://www.nber.org/papers/w33777) | | Microsoft 365 Copilot Experiment: Cross-Government Findings Report | UK Government Digital Service | 2025 | 7,115 survey responses from a trial with 20,000 civil servants | End-of-trial survey, October to December 2024, with answers in time bands averaged at their midpoints. No control group for the time question. | 26 minutes a day saved on average; 17% noticed no clear time saving “Trial participants saved an average of 26 minutes a day when using M365 Copilot.” | [assets.publishing.service.gov.uk ↗](https://assets.publishing.service.gov.uk/media/683db42bd23a62e5d32680d0/M365_Copilot_Experiment_Findings_Report.pdf) | | AI Data Drop: The 11 by 11 Tipping Point | Microsoft WorkLab | 2024 | 1,300 Microsoft 365 Copilot users, including Microsoft employees, after at least three weeks of use | Self-reported daily time saved, in answer bands. Microsoft sells Copilot. | 14 minutes a day saved on average; 30 minutes for the most efficient users “the average person saved 14 minutes a day, or nearly five hours each month” | [microsoft.com ↗](https://www.microsoft.com/en-us/worklab/ai-data-drop-the-11-by-11-tipping-point) | | AI Collaboration Report | Atlassian Teamwork Lab | 2024 | About 5,000 knowledge workers in Australia, the US, India, Germany and France | Online survey, self-estimated time saved, grouped by how advanced people's AI use is. Atlassian sells AI-enabled software. | 53 minutes a day for simple AI users, 105 minutes for the most advanced “We asked people to estimate the amount of time that AI saves them” | [atlassian.com ↗](https://www.atlassian.com/blog/ai-at-work/ai-collaboration-report) | | Beyond Productivity: Measuring the Real Value of AI | Workday, survey by Hanover Research | 2026 | 3,200 full-time employees who use AI, at organizations with $100 million or more in revenue | Online survey, November 2025, self-reported. Read in Workday's press release; the full report is gated. Workday sells AI products. | 85% say they save 1 to 7 hours a week, and that nearly 40% of the savings is lost to rework “85% of employees report saving one to seven hours per week using AI” | [newsroom.workday.com ↗](https://newsroom.workday.com/2026-01-14-New-Workday-Research-Companies-Are-Leaving-AI-Gains-on-the-Table) | Across all workers ## How much of all work hours does AI save across the economy? 1.4 to 2.2 percent of all US work hours, counting the many workers who do not use AI. The share rose from 1.41% in late 2024 to 2.17% in the second quarter of 2026, as the share of employed adults who used AI for work in a given week grew from 28% to 39%. Both figures are self-reported, from the St. Louis Fed's survey series. How much of all work hours does AI save across the economy? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Does generative AI save time at work? (FRED Blog, series RPSGENAITSALL) | Federal Reserve Bank of St. Louis, data series by Bick, Blandin and Deming | 2026 | Employed US adults aged 18 to 64, quarterly survey waves | Same self-reported question as the NBER paper, with non-users counted as zero | 1.41% of all US work hours saved (Q4 2024) rising to 2.17% (Q2 2026) “the share of work hours saved thanks to AI assistance grew from 1.6% to 2.2%.” | [fred.stlouisfed.org ↗](https://fred.stlouisfed.org/blog/2026/08/does-generative-ai-save-time-at-work/) | Measured time Low confidence ## What does measured data show about time saved? In a randomized trial of Microsoft 365 Copilot with 7,137 workers at 66 firms, people who used it spent about 2 fewer hours a week on email (1.4 hours averaged over everyone offered the tool), out of about 11.7. Time in Teams meetings and in Word did not change measurably, and they replied to as many email threads and finished as many documents. In Denmark, payroll records show no change in paid hours or earnings, which fits saved time going into other work. What does measured data show about time saved? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Shifting Work Patterns with Generative AI (NBER Working Paper 33795) | Dillon, Jaffe and Immorlica (Microsoft Research) and Stanton (Harvard Business School) | 2025 | 7,137 knowledge workers at 66 large firms | Randomized field experiment, 2023 to 2024, timed with Microsoft 365 usage data. Most authors work at Microsoft. | 1.4 fewer hours a week on email for everyone offered Copilot (12%), about 2 fewer for those who used it (17%); no significant change in meeting or document time “who used this tool spent two fewer hours on email each week” | [nber.org ↗](https://www.nber.org/papers/w33795) | Guesses and timers ## Are people's own estimates of AI time savings accurate? Often not, and they miss in both directions. In a Microsoft lab test, people guessed AI saved them 36 minutes when the measured saving was 12. In METR's 2025 trial, experienced developers felt 20% faster with AI but took 19% longer. In a 2022 GitHub Copilot test, developers guessed a 35% gain when the measured one was 55.8%. Are people's own estimates of AI time savings accurate? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Early LLM-based Tools for Enterprise Information Workers Likely Provide Meaningful Boosts to Productivity (MSR-TR-2023-43) | Microsoft Research | 2023 | Several lab studies and an early-access survey | Lab tasks timed and compared with participants' own estimates. Tasks were chosen because AI was likely to help. | Participants guessed 36 minutes saved; the measured saving was 12 minutes “They guessed 36 minutes on average, when actual time savings were 12 minutes” | [microsoft.com ↗](https://www.microsoft.com/en-us/research/wp-content/uploads/2023/12/AI-and-Productivity-Report-First-Edition.pdf) | | Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity | Joel Becker, Nate Rush, Beth Barnes and David Rein (METR) | 2025 | 16 experienced open-source developers, 246 tasks | Randomized trial at the task level, compared with developers' own estimates afterwards | Tasks took 19% longer with AI; developers believed they were 20% faster “we find that allowing AI actually increases completion time by 19%” | [arxiv.org ↗](https://arxiv.org/abs/2507.09089) | | The Impact of AI on Developer Productivity: Evidence from GitHub Copilot | Sida Peng (Microsoft Research), Eirini Kalliamvakou and Peter Cihon (GitHub), Mert Demirer (MIT Sloan) | 2023 | 95 freelance developers recruited on Upwork | Controlled experiment, 2022, with an exit survey | Developers estimated a 35% gain; the measured time saving was 55.8% “participants in both treated and control groups estimated a 35% increase in productivity” | [arxiv.org ↗](https://arxiv.org/abs/2302.06590) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Generative AI saves 5.4% of all work hours.** 5.4% is the average among people who used AI for work. Across all workers it was 1.4% in November 2024 and 2.2% by mid-2026. - **Danish study: AI's measured time savings are only 2.8%.** The 2.8% (about 3% in the 2026 revision) is what workers report in a survey. What the authors measure in payroll records is earnings and paid hours, where they find no change. - **Thomson Reuters: AI saves professionals 5 hours a week.** This is a forecast. Respondents predicted AI would save them 5 hours a week within the next year (4 hours in the 2024 edition). - **Microsoft: Copilot saves 11 minutes a day.** In Microsoft's survey, 11 minutes a day is the point at which users start to feel value. The average reported saving was 14 minutes. - **Microsoft Work Trend Index 2024: AI saves workers time, 90% say.** 90% agree AI helps them save time. The report gives no amount of time saved. - **Slack: daily AI users are 64% more productive.** Slack's 2025 survey found daily users 64% more likely to rate their productivity "very good". It is a gap in self-ratings, with no time or output measured. ## More from What AI Saves - [F-09**25–40%**How much faster do people finish writing tasks with AI, and is the work better?](https://evy.so/facts/ai-writing-speed-and-quality/) - [F-10**14–15%**How much more do customer support agents get done with AI?](https://evy.so/facts/ai-customer-support-productivity/) - [F-11**21–26%**How much faster do developers code with AI assistants?](https://evy.so/facts/ai-coding-speed/) - [F-12**27–46%**Does AI help less-experienced workers more than experienced ones?](https://evy.so/facts/ai-helps-less-experienced-workers-more/) - [F-13**14–25%**When does AI make work slower or worse?](https://evy.so/facts/when-ai-makes-work-worse/) [The whole chapter](https://evy.so/facts/what-ai-saves/) · [Chapter poster](https://evy.so/facts/what-ai-saves/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How much time do people spend redoing work that already exists? (4.6–5.7 h) | The AI at Work Fact Book" description: "Workers lose about 4.6 to 5.7 hours a week, 10 to 14 percent of the week, to duplicated work. Primary sources, country differences and misquoted figures." last_updated: "2026-10-24" source: "https://evy.so/facts/time-spent-duplicating-work/" -------------------------------------------------------------------------------- F-07 · Where the Week Goes # How much time do people spend redoing work that already exists? Workers report losing about 4.6 to 5.7 hours a week to duplicated work, or 10 to 14 percent of a 40-hour week. Every figure is self-reported, and two of the three come from the same publisher. 4.6–5.7 h a week on duplicated work, 10 to 14% of a 40-hour week - Medium confidence - 3 primary sources - 2018 to 2021 - Last checked October 13, 2026 ## Cite this EVY (2026). How much time do people spend redoing work that already exists? The AI at Work Fact Book, F-07. https://evy.so/facts/time-spent-duplicating-work/ (last checked October 13, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/time-spent-duplicating-work.json) [Markdown](https://evy.so/facts/time-spent-duplicating-work.md) ## How to read it All three figures are people's own estimates in surveys run for vendors, mostly from 2018 to 2020. Panopto's average comes from a lopsided spread: 21% report zero hours and 14% report 10 or more. Asana's 2021 hours are our conversion: 236 hours a year is 4.5 hours a week, and 13% of a 40-hour week is 5.2. No newer independent measure was found. ## Every study on one scale Shaded: 4.6–5.7 h One figure A range in one study Chart values as text - Scale: hours a week on duplicated work. Shaded range: 4.6–5.7 h. - Asana 2019 (2019): 4 h 38 min - Asana 2021 (2021): 4.5–5.2 h (13%) - Panopto (2018): 5.68 h **Why medium confidence:** Three reports agree closely, but two come from Asana with the same method, so there are only two independent creators, and every figure is self-reported in a vendor survey. ## Sources Sources for F-07 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Anatomy of Work Index 2019 | Asana, survey by Sapio Research | 2019 | 10,223 knowledge workers in Australia, New Zealand, Germany, Japan, the UK and the US | Online survey, August to September 2019, self-reported. Asana sells work management software. | 4 hours 38 minutes a week duplicating work, more than 10% of the week “waste more than 10% of their time duplicating efforts” | [resources.asana.com ↗](https://resources.asana.com/rs/784-XZD-582/images/Anatomy-of-Work-Index.pdf) | | Anatomy of Work Index 2021 | Asana, survey by Sapio Research | 2021 | 13,123 knowledge workers in seven countries | Online survey, October 2020, self-reported | 13% of time on work that has already been done, 236 hours a year “Teams are spending 13% of their time on work that's already been completed” | [asana.com ↗](https://asana.com/resources/anatomy-of-work-index) [Archived copy ↗](https://web.archive.org/web/20210301155022/https://resources.asana.com/rs/784-XZD-582/images/PDF-FY21-Global-EN-Anatomy%20of%20Work%20Report.pdf) Archived copy of the full 2021 report PDF. | | Workplace Knowledge and Productivity Report | Panopto | 2018 | 1,001 US employees with at least 5 years of work experience | Online survey, self-reported. Panopto sells video knowledge-sharing software. | 5.68 hours a week reinventing the wheel and duplicating others' work “nearly 6 hours each week "reinventing the wheel" and duplicating other people's work” | [assets.panopto.com ↗](https://assets.panopto.com/asset/fd3223fa-08dd-46ba-94ad-80d29a787d12/Social-Learning-Workplace-Knowledge-Panopto-White-Paper.pdf) | By country ## Which countries lose the most time to duplicated work? In Asana's 2019 survey, Germany reported the most at 5 hours 9 minutes a week, followed by Australia and New Zealand (5 h 7 min), the UK (5 h 5 min) and the US (4 h 55 min). Japan reported the least at 2 hours 56 minutes. How common ## How common is duplicated work across teams? Half of knowledge workers (50%) have worked on a project and only later found out that another team was doing the same thing, according to Atlassian's 2024 survey of 5,000 knowledge workers. How common is duplicated work across teams? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | The State of Teams 2024 | Atlassian Teamwork Lab | 2024 | 5,000 knowledge workers and 100 Fortune 500 executives | Online survey. Measures how many people have experienced it, with no hours. | 50% have worked on a project another team was also doing “only later found out that another team was working on the same thing” | [atlassian.com ↗](https://www.atlassian.com/blog/state-of-teams-2024) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Panopto: workers lose 5.3 hours a week waiting for information or recreating existing knowledge.** In the report, 5.31 hours is time spent waiting for colleagues' knowledge. Duplicating work is a separate 5.68 hours, and the report says the two overlap and cannot be added. - **Asana: 209 hours a year on duplicated work, 103 hours in unnecessary meetings, 352 hours talking about work.** These blog figures mix editions. 209 hours is the 2019 duplication figure (2021 says 236), and we found the 352 hours in neither the 2019 nor the 2021 report. - **IDC: reworking information takes 15% of knowledge workers' time.** The 15% is an assumption inside a cost model of a hypothetical 1,000-person company (Feldman, KMWorld, 2004). No one measured it. ## More from Where the Week Goes - [F-01**19–25%**How much of the work week do knowledge workers spend searching for information?](https://evy.so/facts/time-spent-searching-for-information/) - [F-02**53–61%**How much of the work day goes to coordinating work rather than doing it?](https://evy.so/facts/time-on-work-about-work/) - [F-03**3–18 h**How many hours a week do people spend in meetings, and how much of it is wasted?](https://evy.so/facts/time-in-meetings/) - [F-04**20–26 min**How long does it take to get back to a task after an interruption?](https://evy.so/facts/time-to-resume-after-interruption/) - [F-05**13–28%**How much of the work week do knowledge workers spend on email and chat?](https://evy.so/facts/time-on-email-and-chat/) - [F-06**3–4 h**How much time do people lose switching between apps at work?](https://evy.so/facts/time-lost-switching-apps/) [The whole chapter](https://evy.so/facts/where-the-week-goes/) · [Chapter poster](https://evy.so/facts/where-the-week-goes/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How much of the work week do knowledge workers spend searching for information? (19–25%) | The AI at Work Fact Book" description: "Knowledge workers spend 19 to 25 percent of the work week, about 9 to 10 hours, searching for information. Five primary studies, sources and method." last_updated: "2026-10-24" source: "https://evy.so/facts/time-spent-searching-for-information/" -------------------------------------------------------------------------------- F-01 · Where the Week Goes # How much of the work week do knowledge workers spend searching for information? The most-cited studies put it at 19 to 25 percent of the work week, or about 9 to 10 hours. Across all five studies the estimates run from about 5 to 15 hours a week, because each one defines searching a little differently. 19–25% of the work week, about 9 to 10 hours Across all 5 studies: 5–15 hours a week. - Medium confidence - 5 primary sources - 2012 to 2025 - Last checked October 13, 2026 ## Cite this EVY (2026). How much of the work week do knowledge workers spend searching for information? The AI at Work Fact Book, F-01. https://evy.so/facts/time-spent-searching-for-information/ (last checked October 13, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/time-spent-searching-for-information.json) [Markdown](https://evy.so/facts/time-spent-searching-for-information.md) ## How to read it All five figures are people's own estimates; none comes from time tracking. They also count different things: McKinsey includes tracking down colleagues, Qatalog counts only searching inside work apps, and Atlassian gives no definition. McKinsey re-used IDC's 2009 survey, so its number is a 2009 measurement published in 2012. Three of the five creators sell search or work-hub software. Hours assume a 5-day week; McKinsey's share uses the 46.5-hour week its respondents reported. ## Every study on one scale Shaded: 19–25% of the week Dashed: all five studies One figure Chart values as text - Scale: hours a week. Shaded range: 19–25% of the week; dashed, all five studies: 4.9 h to 15 h. - McKinsey Global Institute (2012): 8.8 h (19%) - Atlassian (2025): 10 h (25% of 40 h) - Glean / Harris Poll (2022): 2 h a day (10 h) - Qatalog / Cornell (2021): 59 min a day (4.9 h) - Coveo (2025): 3 h a day (15 h) **Why medium confidence:** Three independent studies agree on 19 to 25 percent, but all five are self-reported surveys, and two vendor surveys sit well below and above that cluster. ## Sources Sources for F-01 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | The social economy: Unlocking value and productivity through social technologies | McKinsey Global Institute, using IDC survey data | 2012 | Managers and professionals in IDC's 2009 survey; sample size not stated | Re-analysis of IDC's self-reported time-use survey | 19% of the week (8.8 of 46.5 hours) “19 percent of working hours trying to track down information needed to complete tasks” | [mckinsey.com ↗](https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-social-economy) [Archived copy ↗](https://web.archive.org/web/20260710105837/https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-social-economy) | | The State of Teams 2025 | Atlassian Teamwork Lab | 2025 | 12,000 knowledge workers and 200 Fortune 1000 executives in the US, UK, Australia, India, Germany and France | Online survey, self-reported | 25% of the workweek “Executives and teams alike spend a quarter of the workweek searching for information” | [atlassian.com ↗](https://www.atlassian.com/blog/state-of-teams-2025) | | Hybrid Workplace Habits & Hangups | Glean, survey by The Harris Poll | 2022 | 1,043 employed US adults | Online survey, December 2021, self-reported. Glean sells enterprise search. | At least 2 hours a day, or 25% of the workweek, looking for documents, information or people “spend, on average, at least two hours a day” | [businesswire.com ↗](https://www.businesswire.com/news/home/20220208005510/en/Hybrid-Workplace-Habits-Hangups-Report-Frustrated-Employees-Spend-a-Quarter-of-Workweek-Searching-for-Information-Needed-to-Do-Their-Jobs) [Archived copy ↗](https://web.archive.org/web/20260615090357/https://www.businesswire.com/news/home/20220208005510/en/Hybrid-Workplace-Habits-Hangups-Report-Frustrated-Employees-Spend-a-Quarter-of-Workweek-Searching-for-Information-Needed-to-Do-Their-Jobs) | | Workgeist Report '21 | Qatalog with Cornell University's Ellis Idea Lab | 2021 | Three surveys of about 1,000 remote-working professionals each, US and UK | Online surveys, March 2021, self-reported. Counts only searching inside work apps. Qatalog sells a work hub. | 59 minutes a day finding information trapped in apps and tools “up to 5 hours a week, just trying to find what we need” | [qatalog.com ↗](https://qatalog.com/) [Archived copy ↗](https://web.archive.org/web/20210624191022/https://assets.qatalog.com/language.work/qatalog-2021-workgeist-report.pdf) Qatalog is now part of ClickUp and no longer hosts the report; the archived copy is the original PDF. | | EX Relevance Report 2025: The Search for Relevance | Coveo, survey by Arlington Research | 2025 | 4,000 people who use a computer for work at companies with 5,000 or more staff, US and UK | Online survey, self-reported. Coveo sells AI search. | 3 hours a day searching for information (3.6 hours in the 2022 edition) “an average of three hours daily searching for information” | [coveo.com ↗](https://www.coveo.com/en/company/news-releases/2025/coveo-ex-relevance-report) | Related finding ## How many hours a day do people spend looking for information? About 1.8 to 2 hours a day in the main cluster: McKinsey's 8.8 hours a week over 5 days, and Glean's at least 2 hours a day. The full spread runs from about 1 hour a day (Qatalog, inside work apps only) to 3 hours a day (Coveo). ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **McKinsey: 19.8% of the week is spent searching for information.** The report says 19 percent, and "nearly 20 percent" in its summary. The figure 19.8% does not appear in it. - **IDC: knowledge workers spend 15 to 35% of their time searching, or 2.5 hours a day.** The 2.5 hours was a working assumption IDC used to model costs in 2001. The 15 to 35% rolls up older studies that are not named, so neither figure traces back to a measurement. - **Gartner: professionals spend 47% of their time searching for information.** Gartner's 2023 survey found that 47% of digital workers struggle to find the information they need. It counts people and gives no figure for time. - **Microsoft: 62% of time goes to searching for information.** In Microsoft's 2023 Work Trend Index, 62% of people say they struggle with too much time spent searching. It is a share of people and gives no figure for time. ## More from Where the Week Goes - [F-02**53–61%**How much of the work day goes to coordinating work rather than doing it?](https://evy.so/facts/time-on-work-about-work/) - [F-03**3–18 h**How many hours a week do people spend in meetings, and how much of it is wasted?](https://evy.so/facts/time-in-meetings/) - [F-04**20–26 min**How long does it take to get back to a task after an interruption?](https://evy.so/facts/time-to-resume-after-interruption/) - [F-05**13–28%**How much of the work week do knowledge workers spend on email and chat?](https://evy.so/facts/time-on-email-and-chat/) - [F-06**3–4 h**How much time do people lose switching between apps at work?](https://evy.so/facts/time-lost-switching-apps/) - [F-07**4.6–5.7 h**How much time do people spend redoing work that already exists?](https://evy.so/facts/time-spent-duplicating-work/) [The whole chapter](https://evy.so/facts/where-the-week-goes/) · [Chapter poster](https://evy.so/facts/where-the-week-goes/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How long does it take to get back to a task after an interruption? (20–26 min) | The AI at Work Fact Book" description: "It takes about 20 to 26 minutes to get back to an interrupted task, from observed field studies. Where the 23-minute figure comes from, and how often people are interrupted." last_updated: "2026-10-24" source: "https://evy.so/facts/time-to-resume-after-interruption/" -------------------------------------------------------------------------------- F-04 · Where the Week Goes # How long does it take to get back to a task after an interruption? Field studies that watched people at work found it takes about 20 to 26 minutes to return to an interrupted task, usually after handling about two other things first. The often quoted "23 minutes" comes from this research and counts that whole detour. 20–26 min until back on the interrupted task, observed in the field - Medium confidence - 3 primary sources - 2005 to 2007 - Last checked October 13, 2026 ## Cite this EVY (2026). How long does it take to get back to a task after an interruption? The AI at Work Fact Book, F-04. https://evy.so/facts/time-to-resume-after-interruption/ (last checked October 13, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/time-to-resume-after-interruption.json) [Markdown](https://evy.so/facts/time-to-resume-after-interruption.md) ## How to read it Three different things get called the cost of an interruption: how often attention breaks, how long until people are back on the task, and stress or mistakes measured in the lab. This range covers the second only. It is the elapsed time to return, which includes dealing with the interruption and doing other work first. The Iqbal and Horvitz range is our sum of the two phases they report. All three studies observed mostly US office workers. ## Every study on one scale Shaded: 20–26 min One figure A range in one study Chart values as text - Scale: minutes until back on the task. Shaded range: 20–26 min. - Mark et al., 2005 (2005): 25 min 26 s - Mark, Gallup interview (2006): 23 min 15 s - Iqbal and Horvitz, 2007 (2007): about 20–25 min **Why medium confidence:** Two independent research groups observed workers in the field and land at about 20 to 26 minutes. Both studies are from 2005 to 2007, the samples are small, and the groups define getting back slightly differently. ## Sources Sources for F-04 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | No task left behind? Examining the nature of fragmented work (ACM CHI 2005) | Gloria Mark, Victor M. Gonzalez and Justin Harris, UC Irvine | 2005 | 24 information workers in two teams, over 700 hours of observation | Researchers shadowed each person for 3 days with second-by-second timing. Peer-reviewed. | 25 minutes 26 seconds to resume interrupted work, after an average of 2.26 other tasks “it took an average length of time of 25 min. 26 sec” | [ics.uci.edu ↗](https://ics.uci.edu/~gmark/CHI2005.pdf) | | Too Many Interruptions at Work? (interview with Gloria Mark) | Gallup Business Journal | 2006 | 36 managers, analysts, developers, engineers and project leaders, shadowed 3 days each | The researcher's own account of her shadowing study, in an interview | 23 minutes 15 seconds to resume interrupted work, with about two tasks in between “it was resumed, on average, in 23 minutes and 15 seconds” | [news.gallup.com ↗](https://news.gallup.com/businessjournal/23146/too-many-interruptions-work.aspx) | | Disruption and recovery of computing tasks: field study, analysis, and directions (ACM CHI 2007) | Shamsi T. Iqbal and Eric Horvitz, Microsoft Research | 2007 | 27 computer users logged for two weeks | Logging software for windows, apps, email and chat alerts, plus interviews. Peer-reviewed. | About 10 minutes on the diversion after an email or chat alert, then 10 to 15 minutes more to return to the task “spent on average nearly 10 minutes on switches caused by alerts” | [erichorvitz.com ↗](https://erichorvitz.com/CHI_2007_Iqbal_Horvitz.pdf) | How often ## How often are knowledge workers interrupted? Every few minutes. Microsoft's 2025 usage data counts a meeting, email or chat ping every 2 minutes during core hours for its busiest fifth of users, or 275 a day. In Gloria Mark's shadowing study, people spent about 3 minutes on a task before switching, and interrupted themselves about 44% of the time. A 2016 logging study found a median of 40 seconds on one screen before switching (mean 47 seconds). How often are knowledge workers interrupted? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Breaking down the infinite workday (2025 Work Trend Index) | Microsoft | 2025 | Anonymized Microsoft 365 usage data, 28 days to February 2025, outside the EU; top 20% of users by ping volume | App usage data, observed. Counts incoming pings, which do not always break focus. | Interrupted every 2 minutes during core work hours, 275 times a day “interrupted every 2 minutes by a meeting, email, or notification” | [microsoft.com ↗](https://www.microsoft.com/en-us/worklab/work-trend-index/breaking-down-infinite-workday) | | Neurotics Can't Focus: An in situ Study of Online Multitasking in the Workplace (ACM CHI 2016) | Gloria Mark, Shamsi Iqbal, Mary Czerwinski, Paul Johns and Akane Sano | 2016 | 40 information workers at one US technology company | Window-logging software over two work weeks. Peer-reviewed. | Median 40 seconds of focus on one screen; mean 47 seconds “The median duration of online screen focus was 40 seconds.” | [ics.uci.edu ↗](https://ics.uci.edu/~gmark/Home_page/Publications_files/CHI%2016%20Multitasking%20and%20Focus.pdf) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **It takes 23 minutes to refocus after an interruption.** The 23 minutes 15 seconds (Gloria Mark, in a 2006 Gallup interview) is the time until people returned to the interrupted work, after about two other tasks. It says nothing about how long concentration takes to come back. - **The 23 minutes comes from Mark, Gudith and Klocke, "The Cost of Interrupted Work" (2008).** That paper is a lab experiment on email tasks. Interrupted people finished faster but with more stress, and the paper contains no 23-minute figure. - **Workers are interrupted 275 times a day (Microsoft).** Microsoft's count covers its busiest 20% of users by ping volume and runs over 24 hours. It does not describe the typical worker. - **Attention spans fell from 150 seconds in 2004 to 47 seconds today.** The 47 seconds is in Mark and colleagues' 2016 paper. The earlier values come from talks and essays; we did not open the underlying papers, so only the 2016 figure is used here. ## More from Where the Week Goes - [F-01**19–25%**How much of the work week do knowledge workers spend searching for information?](https://evy.so/facts/time-spent-searching-for-information/) - [F-02**53–61%**How much of the work day goes to coordinating work rather than doing it?](https://evy.so/facts/time-on-work-about-work/) - [F-03**3–18 h**How many hours a week do people spend in meetings, and how much of it is wasted?](https://evy.so/facts/time-in-meetings/) - [F-05**13–28%**How much of the work week do knowledge workers spend on email and chat?](https://evy.so/facts/time-on-email-and-chat/) - [F-06**3–4 h**How much time do people lose switching between apps at work?](https://evy.so/facts/time-lost-switching-apps/) - [F-07**4.6–5.7 h**How much time do people spend redoing work that already exists?](https://evy.so/facts/time-spent-duplicating-work/) [The whole chapter](https://evy.so/facts/where-the-week-goes/) · [Chapter poster](https://evy.so/facts/where-the-week-goes/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "How much do workers trust the output of AI tools at work? (24–53%) | The AI at Work Fact Book" description: "24 to 53 percent of workers trust AI output at work, depending on the definition; only 7 percent of developers trust it for important decisions. KPMG, Stack Overflow and Google DORA." last_updated: "2026-10-24" source: "https://evy.so/facts/trust-in-ai-output/" -------------------------------------------------------------------------------- F-21 · What Teams Trust # How much do workers trust the output of AI tools at work? A quarter to a half of workers trust AI output at work, depending on how strictly trust is defined: 24 to 33 percent clearly trust its quality or accuracy, and 53 percent of employees across 47 countries are willing to trust AI for work. Very few trust it with important decisions: 7 percent of developers in 2026. 24–53% of workers trust AI output at work: 24 to 33% firmly, 53% willing to trust it for work - Medium confidence - 4 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). How much do workers trust the output of AI tools at work? The AI at Work Fact Book, F-21. https://evy.so/facts/trust-in-ai-output/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/trust-in-ai-output.json) [Markdown](https://evy.so/facts/trust-in-ai-output.md) ## How to read it Each source asks about trust differently: the quality of AI output (DORA), its accuracy (Stack Overflow 2025), willingness to trust AI for work (KPMG) and when people trust it (Stack Overflow 2026). Where you draw the line matters: counting DORA's somewhat trust moves the figure from 24% to 70%. DORA and Stack Overflow survey software workers, who can test AI output; KPMG covers all occupations. All are self-reports. ## Every study on one scale Shaded: 24–53% One figure Chart values as text - Scale: % who trust AI output at work. Shaded range: 24–53%. - Google DORA (2025): 24% a lot or a great deal - Stack Overflow 2025 (2025): 33% (46% distrust) - KPMG and Melbourne (2025): 53% willing to trust for work **Why medium confidence:** Three independent creators show the same shape: a minority with firm trust, a large group with conditional trust, and little trust for high-stakes use. The range is wide because the questions differ, and two of the three survey software workers only. Studies without a figure on this scale are listed in the table only. ## Sources Sources for F-21 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | State of AI-assisted Software Development 2025 (DORA Report) | Google Cloud DORA | 2025 | Nearly 5,000 technology professionals, June to July 2025 | Online survey: how much did you trust the quality of AI-generated output at work. Google sells AI coding tools. | 24% trust it a lot or a great deal, 46% somewhat, 30% a little or not at all “currently report little to no trust in the code generated by AI” | [dora.dev ↗](https://dora.dev/research/2025/dora-report/) | | 2025 Developer Survey: AI | Stack Overflow | 2025 | 33,244 developers answering the trust question, mid-2025 | Opt-in online survey: how much do you trust the accuracy of AI output in your workflow | 33% trust its accuracy (3.1% highly), 46% distrust it “More developers actively distrust the accuracy of AI tools (46%) than trust it (33%)” | [survey.stackoverflow.co ↗](https://survey.stackoverflow.co/2025/ai) | | Trust, attitudes and use of artificial intelligence: A global study 2025 | University of Melbourne and KPMG International | 2025 | 32,352 employed people in 47 countries, November 2024 to January 2025 | Online survey; willing to trust AI for work purposes, including somewhat willing | 53% trust AI tools for work, from 31% in Japan to 81% in India and Nigeria; 46% of all people are willing to trust AI in general “About half (53%) report trusting AI tools for work purposes” | [mbs.edu ↗](https://mbs.edu/-/media/PDF/Research/Trust_in_AI_Report.pdf) | | 2026 Developer Survey: AI data | Stack Overflow | 2026 | 14,304 developers answering this question, June to August 2026 | New 2026 question: when do you trust AI output in your workflow | 48% trust it when they can easily verify it; 6.6% trust it for important work decisions; 12.3% do not trust it for most work tasks “48% trust AI when they can easily verify it” | [survey.stackoverflow.co ↗](https://survey.stackoverflow.co/2026/ai/data) | Conditional trust ## Do workers trust AI with important decisions? Rarely. In 2026, 48% of developers trust AI output when they can easily verify it, and only 6.6% trust it for important work decisions (4% among early-career developers). The pattern holds outside software: people trust AI most where the result is easy to check. Trust in AI in general ## Is trust in AI rising or falling? Falling a little in the countries tracked longest. Across 47 countries, 46% of people are willing to trust AI in general, while 65% believe it gives accurate output, so trust in AI's answers runs ahead of trust in AI overall. In the 17 countries surveyed in both 2022 and 2024, willingness to rely on AI fell from 52% to 43%. ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Only 3% of developers trust AI (Stack Overflow).** 3.1% highly trust the accuracy of AI output. Including those who somewhat trust it, 33% trust it, and 46% distrust it. - **Most workers don't trust AI (KPMG).** The 54% who are wary combines unwilling and ambivalent people, asked about AI in general. Among employees asked about AI for work, 53% are willing to trust it. - **30% of developers don't trust AI-generated code (DORA).** The 30% is a little trust (23%) plus none at all (7%). Only 7% have no trust at all. ## More from What Teams Trust - [F-22**41–44%**How often do people check AI output before using it?](https://evy.so/facts/checking-ai-output/) - [F-23**17–45%**How often is AI output wrong on real work tasks?](https://evy.so/facts/ai-error-rates-at-work/) - [F-24**2–12%**Which work tasks do people want AI to do on its own, and which do they want to keep?](https://evy.so/facts/tasks-workers-want-ai-to-do/) - [F-25**30–44%**How many workers use AI in ways that break company policy?](https://evy.so/facts/ai-use-against-company-policy/) - [F-26**42–48%**How often do employees put company data into public AI tools?](https://evy.so/facts/company-data-in-public-ai/) - [F-27**21–29%**How many organizations have had a security incident or data breach involving AI?](https://evy.so/facts/ai-security-breaches/) [The whole chapter](https://evy.so/facts/what-teams-trust/) · [Chapter poster](https://evy.so/facts/what-teams-trust/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "What AI Saves: statistics on time saved with AI at work | The AI at Work Fact Book" description: "Chapter 2 of The AI at Work Fact Book. What AI changes in knowledge work, measured in trials and surveys: time saved, writing, customer support, coding, who gains most and when AI makes work worse." last_updated: "2026-10-24" source: "https://evy.so/facts/what-ai-saves/" -------------------------------------------------------------------------------- - F-083–5.4% ## [How much time does AI save knowledge workers?](https://evy.so/facts/time-saved-by-ai-at-work/) Workers who use generative AI say it saves them about 3 to 5.4 percent of their work hours, roughly 1 to 2 hours of a 40-hour week. The one large trial that measured time found users spent 1.4 to 2 fewer hours a week on email, with no measurable change in time in meetings or writing documents. Medium confidence 6 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/time-saved-by-ai-at-work/) - F-0925–40% ## [How much faster do people finish writing tasks with AI, and is the work better?](https://evy.so/facts/ai-writing-speed-and-quality/) In controlled experiments, professionals using ChatGPT or GPT-4 finished defined writing and consulting tasks 25 to 40 percent faster, and graders rated the work 18 to 34 percent better. In a six-month field trial of everyday work, time spent writing documents did not change measurably. Medium confidence 3 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/ai-writing-speed-and-quality/) - F-1014–15% ## [How much more do customer support agents get done with AI?](https://evy.so/facts/ai-customer-support-productivity/) In the largest real-world study, chat agents at a Fortune 500 software firm resolved 14 to 15 percent more issues per hour with an AI assistant, and the least skilled gained 30 to 36 percent. A larger randomized trial at Alibaba found much smaller speed gains, with chats 1 to 8 percent faster. Low confidence 2 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/ai-customer-support-productivity/) - F-1121–26% ## [How much faster do developers code with AI assistants?](https://evy.so/facts/ai-coding-speed/) Randomized trials inside companies find gains of about 21 to 26 percent. Across all studies the results run from 19 percent slower, for experienced developers in their own mature codebases, to 56 percent faster on one simple lab task, so the evidence conflicts. Medium confidence 5 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/ai-coding-speed/) - F-1227–46% ## [Does AI help less-experienced workers more than experienced ones?](https://evy.so/facts/ai-helps-less-experienced-workers-more/) Mostly yes, on well-defined tasks: the newest or weakest workers gained about 27 to 46 percent in support, consulting and software studies, against 0 to 17 percent for the strongest. On open-ended business decisions the pattern reversed: in Kenya, strong entrepreneurs gained from an AI adviser while weak ones did nearly 10 percent worse. Medium confidence 5 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/ai-helps-less-experienced-workers-more/) - F-1314–25% ## [When does AI make work slower or worse?](https://evy.so/facts/when-ai-makes-work-worse/) On tasks the AI handles poorly, controlled studies find results 14 to 25 percent worse: consultants using GPT-4 gave fewer correct answers, and experienced developers took 19 percent longer. No study measures how often this happens across all AI-assisted work. Medium confidence 3 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/when-ai-makes-work-worse/) -------------------------------------------------------------------------------- title: "What AI Saves, on one sheet | The AI at Work Fact Book" description: "What AI Saves on one poster: 6 numbers from The AI at Work Fact Book, each with its sources printed on the sheet. Free to reuse under CC BY 4.0." last_updated: "2026-10-24" source: "https://evy.so/facts/what-ai-saves/sheet/" -------------------------------------------------------------------------------- about 1 to 2 of 40 hours in the week F-08 3–5.4% of work hours saved, by AI users' own estimate St. Louis Fed · Denmark · UK civil service 25 to 40 of every 100 minutes saved F-09 25–40% less time on a defined writing task with AI Noy and Zhang 2023 · Dell'Acqua et al. 2026 14 to 15 more issues per 100 F-10 14–15% more issues resolved per hour by support agents with AI Brynjolfsson, Li and Raymond 2025 · Alibaba 2026 21 to 26 of every 100 minutes saved F-11 21–26% faster in company trials; all studies run from 19% slower to 56% faster Google 2024 · Microsoft, Accenture 2025 · METR 2025 27 to 46 more per 100 for the weakest F-12 27–46% gain for the newest or weakest workers; the strongest gain 0 to 17% Brynjolfsson et al. 2025 · Dell'Acqua et al. 2023 · Cui et al. 2026 14 to 25 points worse out of 100 F-13 14–25% worse results on tasks the AI handles poorly Dell'Acqua et al. 2026 · METR 2025 · UK DBT 2025 How to read it Each 5 by 8 grid is a 40-hour week: a row per day, a square per hour. Each 10 by 10 grid is 100%. Solid orange is the low end of the range across studies, light orange runs to the high end. Methods, samples and links for every study are at **evy.so/facts**. ## The numbers on this sheet - [F-08**3–5.4%**of work hours, by the estimate of people who use AI (about 1 to 2 hours a week). How much time does AI save knowledge workers? Sources: Bick, Blandin and Deming (2025), Humlum and Vestergaard, Denmark (2026), UK civil service (2025), Microsoft (2024), Atlassian (2024), Workday (2026).](https://evy.so/facts/time-saved-by-ai-at-work/) - [F-09**25–40%**less time on a defined writing or consulting task, with quality up 18 to 34%. How much faster do people finish writing tasks with AI, and is the work better? Sources: Noy and Zhang (2023), Dell'Acqua et al., BCG (2026), Dillon et al. (2025).](https://evy.so/facts/ai-writing-speed-and-quality/) - [F-10**14–15%**more issues resolved per hour, in one firm's study. How much more do customer support agents get done with AI? Sources: Brynjolfsson, Li and Raymond (2025), Ni et al., Alibaba (2026).](https://evy.so/facts/ai-customer-support-productivity/) - [F-11**21–26%**faster, or more tasks done, in randomized trials inside companies. How much faster do developers code with AI assistants? Sources: Peng et al., lab task (2023), Cui et al., three companies (2025), Paradis et al., Google (2024), METR, 2025 (2025), METR, late-2025 rerun (2026).](https://evy.so/facts/ai-coding-speed/) - [F-12**27–46%**gain for the newest or weakest workers; the strongest gain 0 to 17%. Does AI help less-experienced workers more than experienced ones? Sources: Brynjolfsson et al., support (2025), Dell'Acqua et al., BCG (2023), Cui et al., software (2026), Noy and Zhang (2023), Ni et al., Alibaba (2026).](https://evy.so/facts/ai-helps-less-experienced-workers-more/) - [F-13**14–25%**worse on tasks the AI handles poorly (points fewer correct answers, or percent more time). When does AI make work slower or worse? Sources: Dell'Acqua et al., BCG (2026), METR (2025), UK DBT, Excel task (2025).](https://evy.so/facts/when-ai-makes-work-worse/) Using the sheet in a post, deck or article? Credit "The AI at Work Fact Book by EVY" and link to [evy.so/facts/what-ai-saves/](https://evy.so/facts/what-ai-saves/). -------------------------------------------------------------------------------- title: "What stops small companies from adopting AI? (54–63%) | The AI at Work Fact Book" description: "54 to 63 percent of small firms with no AI plans say AI does not apply to them; knowledge and privacy follow, and cost is cited by 6 to 16 percent. Census, Federal Reserve, OECD and Eurostat." last_updated: "2026-10-24" source: "https://evy.so/facts/what-stops-small-companies-adopting-ai/" -------------------------------------------------------------------------------- F-31 · Small Teams # What stops small companies from adopting AI? Mostly the belief that AI does not apply to them: 54 to 63 percent of US firms under 50 employees with no AI plans say so, and 57 percent of non-adopting SMEs in seven OECD countries. Lack of knowledge and privacy worries come next, at about one in five each; cost is cited by 6 to 16 percent. 54–63% of small firms with no AI plans say AI does not apply to their business - High confidence - 3 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). What stops small companies from adopting AI? The AI at Work Fact Book, F-31. https://evy.so/facts/what-stops-small-companies-adopting-ai/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/what-stops-small-companies-adopting-ai.json) [Markdown](https://evy.so/facts/what-stops-small-companies-adopting-ai.md) ## How to read it Each source asks a different group. The Census Bureau and the Federal Reserve ask firms with no plans to use AI; the OECD asks non-users; Eurostat (below) asks only non-users that had considered AI, which is why its shares run much higher. Across creators the order holds: AI seems irrelevant, then knowledge gaps and privacy worries, with cost and regulation minor for small firms. Small firms cite privacy less often than large ones (20% against 30% in the Census data). ## Every study on one scale Shaded: 54–63% One figure A range in one study Chart values as text - Scale: % saying AI does not apply to their business. Shaded range: 54–63%. - US Census (2026): 58–63% - Federal Reserve (2026): 54–60% - OECD, seven countries (2025): 57.3% **Why high confidence:** Three independent creators with different methods agree that 'does not apply to us' leads by a wide margin. The ranking of the smaller barriers is less stable, because answer lists differ and one survey asks only firms that had considered AI. ## Sources Sources for F-31 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | BTOS AI Supplement Table 2026 | US Census Bureau | 2026 | More than 117,000 responding US firms, November 2025 to February 2026 | Official survey of firms not planning to use AI in the next six months; several answers allowed | AI not applicable: 63.3% (1 to 4 employees) to 58.1% (20 to 49), 42.2% at 250+; lack of knowledge 21 to 23%; privacy 20 to 22%; too expensive 6.7 to 7.3% “AI is not applicable to this business” | [census.gov ↗](https://www.census.gov/hfp/btos/downloads/AI_Supplement_Table_2026.xlsx) | | 2026 Report on Employer Firms (2025 Small Business Credit Survey) | Federal Reserve Banks | 2026 | 1,855 US employer firms with no plans to use AI, September to November 2025 | Weighted convenience sample; several answers allowed | AI not applicable: 54% to 60% by size band under 50; prefer not to use it 21 to 33%; cost 6 to 16% “AI is not applicable to my business” | [fedsmallbusiness.org ↗](https://www.fedsmallbusiness.org/reports/survey/2026/2026-report-on-employer-firms) | | Generative AI and the SME Workforce | OECD, survey by Ipsos | 2025 | SMEs not using generative AI, from 5,232 surveyed in seven countries, late 2024 | Random telephone survey weighted by number of firms | Not suited to the company's work 57.3%; legal or copyright concerns 54.1%; data concerns 52.5%; staff lack skills 49.8%; value for money 21% “The most common reason for not using generative AI is that it is not suited” | [oecd.org ↗](https://www.oecd.org/en/publications/generative-ai-and-the-sme-workforce_2d08b99d-en/full-report/component-6.html) | Firms that looked into AI ## What stops small firms that have looked into AI? Expertise, legal uncertainty and data worries. Among EU firms with 10 to 49 employees that considered AI and did not adopt it, 71% cite lack of expertise, 55% unclear legal consequences, 52% data protection worries and 39% cost (2025). But only about 1 in 8 small EU non-users had ever considered AI, so as shares of all small firms these are 4 to 7%. What stops small firms that have looked into AI? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Artificial intelligence by size class of enterprise (dataset isoc\_eb\_ai) | Eurostat | 2026 | EU enterprises with 10 or more employees, 2025 survey | Official statistics; asked only of non-users that had considered AI | Small firms that considered AI: lack of expertise 70.87%, legal uncertainty 54.79%, data protection 52.49%, cost 38.82%; 12.65% of small non-users had considered AI “Enterprises do not use AI technologies, because of a lack of relevant expertise” | [ec.europa.eu ↗](https://ec.europa.eu/eurostat/databrowser/view/isoc_eb_ai/default/table) | No barrier ## Do most small firms face barriers to AI at all? Many say they do not. In the UK in June 2026, 52% of businesses under 10 employees and 42% of those with 10 to 49 reported no barrier to adopting AI, and no single barrier reached 14% among firms with 10 to 49 employees. In Canada, 41% of businesses with 1 to 4 employees say AI is not relevant to them. Do most small firms face barriers to AI at all? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Artificial intelligence in UK businesses: 2023 to 2026 | Office for National Statistics (UK) | 2026 | 38,637 responding UK businesses, June 2026 | Official statistics in development; barriers in the last three months | No barriers: 51.6% (0 to 9 employees), 42.0% (10 to 49); expertise 7.0% and 13.3%; cost 6.5% and 11.3% “41% of UK businesses with 10 or more employees report no barriers to adoption.” | [ons.gov.uk ↗](https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Cost is the biggest barrier stopping small businesses from using AI.** Official surveys don't support it: cost is cited by 6.7 to 7.3% of small US firms with no AI plans (Census) and 6 to 16% (Federal Reserve). It ranks high only among EU firms that had considered AI. - **70.89% of EU companies cite lack of expertise as the main barrier (2025).** 70.89% is Eurostat's 2024 value shown under a 2025 label; the 2025 value is 70.31%. It is asked only of non-users that had considered AI. As a share of all EU enterprises, lack of expertise is 7.8%. - **Census: 65% of businesses say AI is not applicable to them.** That appears in a working paper's text. Its own chart and the published table give 61.6% of firms not planning to use AI. - **Lack of skills is the top barrier for SMEs.** True in Japan and Korea and among EU firms that considered AI. Across the seven OECD countries and in US official data, the top reason is that AI does not fit the business. ## More from Small Teams - [F-28**96–99%**How much of the economy do small companies make up?](https://evy.so/facts/small-companies-share-of-economy/) - [F-29**20–46%**What share of small businesses use AI, and for what?](https://evy.so/facts/small-businesses-using-ai/) - [F-30**13–38 pts**How does AI adoption differ by company size?](https://evy.so/facts/ai-adoption-by-company-size/) - [F-32**65–78%**Do small firms that use AI see gains?](https://evy.so/facts/small-firms-gains-from-ai/) - [F-33**45–52 h**How many hours do small business owners work, and how much goes to admin?](https://evy.so/facts/small-business-owner-hours/) [The whole chapter](https://evy.so/facts/small-teams/) · [Chapter poster](https://evy.so/facts/small-teams/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "What Teams Trust: statistics on trust in AI at work | The AI at Work Fact Book" description: "Chapter 4 of The AI at Work Fact Book. What people trust AI to do, and what they check: trust in AI output, checking, error rates, what people hand over, policy breaches, company data and AI-related breaches." last_updated: "2026-10-24" source: "https://evy.so/facts/what-teams-trust/" -------------------------------------------------------------------------------- - F-2124–53% ## [How much do workers trust the output of AI tools at work?](https://evy.so/facts/trust-in-ai-output/) A quarter to a half of workers trust AI output at work, depending on how strictly trust is defined: 24 to 33 percent clearly trust its quality or accuracy, and 53 percent of employees across 47 countries are willing to trust AI for work. Very few trust it with important decisions: 7 percent of developers in 2026. Medium confidence 4 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/trust-in-ai-output/) - F-2241–44% ## [How often do people check AI output before using it?](https://evy.so/facts/checking-ai-output/) About four in ten AI users regularly use AI output without checking it, by their own account: 44 percent of AI-using employees across 47 countries do not verify its accuracy most of the time, and knowledge workers report no critical thinking on 41 percent of their AI-assisted tasks. 56 percent of AI-using employees say AI has caused a mistake in their work at least once. Medium confidence 4 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/checking-ai-output/) - F-2317–45% ## [How often is AI output wrong on real work tasks?](https://evy.so/facts/ai-error-rates-at-work/) In independent tests on realistic professional questions, 17 to 45 percent of AI answers contained a hallucination or significant error (2024 to 2025). Rates fall sharply when the AI only summarizes a document it is given, and rise sharply for whole projects, and they change with every model release. Medium confidence 3 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/ai-error-rates-at-work/) - F-242–12% ## [Which work tasks do people want AI to do on its own, and which do they want to keep?](https://evy.so/facts/tasks-workers-want-ai-to-do/) Very few want AI working fully on its own: about 2 percent, in two independent studies. People welcome AI help on many tasks, mostly repetitive ones, and most prefer an equal partnership with AI, while keeping decisions about people, communication and creative control. Medium confidence 3 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/tasks-workers-want-ai-to-do/) - F-2530–44% ## [How many workers use AI in ways that break company policy?](https://evy.so/facts/ai-use-against-company-policy/) About three to four in ten workers have used AI in ways that break company policy: 44 percent of AI-using employees say so themselves, and IT leaders estimate about 30 percent. Missing rules are a big part of it: 56 percent have used AI without knowing whether it was allowed, and only 38 to 44 percent of organizations have a formal AI policy. Medium confidence 3 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/ai-use-against-company-policy/) - F-2642–48% ## [How often do employees put company data into public AI tools?](https://evy.so/facts/company-data-in-public-ai/) A little under half of workers say they have put company information into public AI tools: 48 percent of AI-using employees, and 42 percent of privacy and security professionals themselves. Monitoring software confirms it happens routinely, and a third to a half of AI use at work runs through personal accounts. Medium confidence 2 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/company-data-in-public-ai/) - F-2721–29% ## [How many organizations have had a security incident or data breach involving AI?](https://evy.so/facts/ai-security-breaches/) About one in five to three in ten organizations report a breach of, or attack on, their own AI systems in the past year: 21 percent of breached organizations in IBM's 2026 study and 29 percent in Gartner's 2025 survey. Employees' unapproved AI was involved in 43 percent of the breaches IBM studied, up from 20 percent. Medium confidence 3 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/ai-security-breaches/) -------------------------------------------------------------------------------- title: "What Teams Trust, on one sheet | The AI at Work Fact Book" description: "What Teams Trust on one poster: 7 numbers from The AI at Work Fact Book, each with its sources printed on the sheet. Free to reuse under CC BY 4.0." last_updated: "2026-10-24" source: "https://evy.so/facts/what-teams-trust/sheet/" -------------------------------------------------------------------------------- 24 to 53 of 100 F-21 24–53% of workers trust AI output at work Google DORA 2025 · Stack Overflow 2025 · KPMG and Melbourne 2025 41 to 44 of 100 F-22 41–44% of AI users, or their AI tasks, skip routine checking KPMG and Melbourne 2025 · Microsoft and CMU 2025 17 to 45 of 100 F-23 17–45% of AI answers had a serious error on realistic professional questions Stanford 2025 · EBU and BBC 2025 2 want AI alone, up to 12 want it to lead, out of 100 F-24 2% want AI deciding fully on its own; most prefer an equal partnership Stanford 2026 · KPMG and Melbourne 2025 30 to 44 of 100 F-25 30–44% of workers have used AI against company policy KPMG and Melbourne 2025 · GoTo 2026 42 to 48 of 100 F-26 42–48% of workers say they have put company data into public AI tools KPMG and Melbourne 2025 · Cisco 2025 21 to 29 of 100 F-27 21–29% of organizations report a breach of, or attack on, their own AI IBM and Ponemon 2026 · Gartner 2025 How to read it Each 10 by 10 grid is 100%. Solid orange is the low end of the range across studies, light orange runs to the high end. Methods, samples and links for every study are at **evy.so/facts**. ## The numbers on this sheet - [F-21**24–53%**of workers trust AI output at work: 24 to 33% firmly, 53% willing to trust it for work. How much do workers trust the output of AI tools at work? Sources: Google DORA (2025), Stack Overflow 2025 (2025), KPMG and Melbourne (2025), Stack Overflow 2026 (2026).](https://evy.so/facts/trust-in-ai-output/) - [F-22**41–44%**of AI users, or of their AI tasks, go without routine checking (self-reported). How often do people check AI output before using it? Sources: KPMG and Melbourne (2025), Lee et al., Microsoft and CMU (2025), Stack Overflow 2026 (2026), Noy and Zhang, experiment (2023).](https://evy.so/facts/checking-ai-output/) - [F-23**17–45%**of AI answers had a hallucination or significant error, in independent tests (2024 to 2025). How often is AI output wrong on real work tasks? Sources: Stanford, legal research tools (2025), EBU and BBC, news answers (2025), Tow Center, news citations (2025).](https://evy.so/facts/ai-error-rates-at-work/) - [F-24**2–12%**want AI to work alone (about 2%) or to lead (up to 12%). Which work tasks do people want AI to do on its own, and which do they want to keep? Sources: Shao et al., Stanford (2026), KPMG and Melbourne (2025), Stack Overflow 2026 (2026).](https://evy.so/facts/tasks-workers-want-ai-to-do/) - [F-25**30–44%**of workers have used AI against company policy (44% self-reported, about 30% by IT leaders' estimate). How many workers use AI in ways that break company policy? Sources: KPMG and Melbourne (employees) (2025), GoTo (IT leaders) (2026), Software AG (stated intention) (2024).](https://evy.so/facts/ai-use-against-company-policy/) - [F-26**42–48%**of workers say they have put company information into public AI tools (2024 to 2025). How often do employees put company data into public AI tools? Sources: KPMG and Melbourne (employees) (2025), Cisco (privacy professionals) (2025).](https://evy.so/facts/company-data-in-public-ai/) - [F-27**21–29%**of organizations report a breach of, or attack on, their own AI systems (2025 to 2026). How many organizations have had a security incident or data breach involving AI? Sources: IBM and Ponemon (2026), Gartner (2025), Microsoft (2026).](https://evy.so/facts/ai-security-breaches/) Using the sheet in a post, deck or article? Credit "The AI at Work Fact Book by EVY" and link to [evy.so/facts/what-teams-trust/](https://evy.so/facts/what-teams-trust/). -------------------------------------------------------------------------------- title: "What do people use AI for at work? (40–52%) | The AI at Work Fact Book" description: "About half of AI-using workers use it for writing or editing, and about half for search; coding is 13 to 27 percent. Surveys and usage logs from OpenAI, Anthropic and Microsoft." last_updated: "2026-10-24" source: "https://evy.so/facts/what-workers-use-ai-for/" -------------------------------------------------------------------------------- F-15 · Who Uses AI at Work # What do people use AI for at work? Mostly for writing and finding information: about half of AI-using workers use it for writing or editing, and about half for research or search. Coding is a minority use outside tech, named by 13 to 27 percent of AI users in surveys. 40–52% of AI users use it for writing or editing, the most common use - High confidence - 5 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). What do people use AI for at work? The AI at Work Fact Book, F-15. https://evy.so/facts/what-workers-use-ai-for/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/what-workers-use-ai-for.json) [Markdown](https://evy.so/facts/what-workers-use-ai-for.md) ## How to read it Surveys ask workers what they use AI for, and the answers depend on the list offered and on whether the question asks about any use or the most helpful use. Usage logs classify real conversations, but each covers one product, counts messages instead of people, mixes work and personal use, and comes from the vendor. On general chatbots both kinds agree: writing and finding information lead. About two thirds of ChatGPT writing requests ask it to change text the user wrote. ## Every study on one scale Shaded: 40–52% One figure Chart values as text - Scale: % using AI for writing or editing. Shaded range: 40–52%. - Gallup (survey) (2026): 51% writing and editing - Pew (survey) (2025): 52% editing - Bick, Blandin and Deming (survey) (2025): 39.5% rank writing top two - OpenAI (usage logs) (2025): 40% of work messages **Why high confidence:** Three US surveys and two usage-log studies, using different methods, agree that writing and finding information are the top work uses of general AI chatbots. Claude's logs lean toward coding, which reflects its user base and is shown separately. Studies without a figure on this scale are listed in the table only. ## Sources Sources for F-15 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Organizational AI Adoption Jumps Six Points (Q2 2026 workforce study) | Gallup | 2026 | US employees who use AI, from 22,573 surveyed in May 2026 | Self-reported uses, several answers allowed | Writing and editing 51%, search or research 49%, general help or problem-solving 39%, data analysis 18%, coding 16% “the most common uses of AI are writing and editing (51%), search or research (49%)” | [gallup.com ↗](https://www.gallup.com/workplace/712736/organizational-adoption-jumps-six-points.aspx) | | U.S. Workers Are More Worried Than Hopeful About Future AI Use in the Workplace | Pew Research Center | 2025 | Workers who have used AI chatbots at work, from 5,273 employed US adults, October 2024 | Self-reported uses ever, several answers allowed | Research or finding information 57%, editing written content 52%, drafting 47%, summarizing 40%, analyzing data or writing code 27% “Most common uses for AI chatbots at work are doing research and editing written content” | [pewresearch.org ↗](https://www.pewresearch.org/social-trends/2025/02/25/workers-experience-with-ai-chatbots-in-their-jobs/) | | The Rapid Adoption of Generative AI (NBER Working Paper 32966) | Alexander Bick, Adam Blandin and David Deming | 2025 | 4,698 employed generative AI users, August and November 2024 | Users ranked the tasks AI helped with most last week (top two) | Writing 39.5%, administrative tasks 25.6%, translating or summarizing 22.7%, searching for facts 18.0%, coding 12.9% “The highest ranked tasks at work were writing (39.5 percent), administrative tasks” | [nber.org ↗](https://www.nber.org/papers/w32966) | | How People Use ChatGPT (NBER Working Paper 34255) | Aaron Chatterji, David Deming and colleagues (OpenAI, Duke, Harvard) | 2025 | Random samples of consumer ChatGPT messages, May 2024 to June 2025; business plans excluded | Automated classification of messages. Counts messages, not people. Authors include OpenAI staff. | Writing is 40% of work-related messages; 27% of all consumer messages were work-related in June 2025; programming is 4.2% of all messages “Writing is the most common use case at work, accounting for 40% of work-related messages” | [nber.org ↗](https://www.nber.org/papers/w34255) | | Working with AI: Measuring the Applicability of Generative AI to Occupations | Kiran Tomlinson, Sonia Jaffe, Will Wang, Scott Counts and Siddharth Suri (Microsoft Research) | 2025 | About 200,000 anonymized US Bing Copilot conversations, January to September 2024 | Conversations matched to work activities by an AI classifier. Authors are Microsoft employees. | Most common user goals: learning, communicating, explaining and writing “the most common and successful AI-assisted work activities involve information work” | [arxiv.org ↗](https://arxiv.org/abs/2507.07935) | Claude's logs ## Why do some studies say a third of AI use is coding? That comes from Claude's usage logs, where tasks tied to computer and mathematical jobs were 35% of Claude.ai conversations in February 2026, because Claude's users lean toward software work. In ChatGPT, programming is 4.2% of messages, and in worker surveys 13 to 27% of AI users name coding. In November 2025, Claude.ai chats were 52% augmentation (learning, iterating, checking) and 45% automation, while API traffic was about three quarters automation. Why do some studies say a third of AI use is coding? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Anthropic Economic Index reports (January and March 2026) | Anthropic | 2026 | 1 million sampled conversations each from Claude.ai and the API, February 2026; a one-week sample in November 2025 | Privacy-preserving classification of conversations, mapped to occupational tasks. Anthropic sells Claude. | Computer and mathematical tasks 35% of Claude.ai conversations (February 2026); Claude.ai 52% augmentation and 45% automation (November 2025) “The share of conversations classified as augmented jumped 5pp to 52%” | [anthropic.com ↗](https://www.anthropic.com/research/anthropic-economic-index-january-2026-report) | Work or not ## Is most chatbot use for work? No. In June 2025, 73% of consumer ChatGPT messages were not about work, up from 53% a year earlier. Business and enterprise plans were not in that sample, so work use is understated. On Claude.ai, 46% of conversations were work in November 2025. ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Most ChatGPT use is for work.** 73% of consumer ChatGPT messages were not about work in June 2025. - **A third or more of AI use is coding.** True for Claude only. Programming is 4.2% of ChatGPT messages, and 13 to 27% of AI-using workers name coding in surveys. - **57% of AI use is augmentation, 43% automation (Anthropic).** That described Claude.ai in January 2025. By November 2025 it was 52% augmentation and 45% automation on Claude.ai, and API traffic was about three quarters automation. - **Therapy and companionship is the top use of generative AI.** OpenAI's study of ChatGPT messages found relationships and personal reflection at 1.9% of messages. ## More from Who Uses AI at Work - [F-14**40–50%**What share of workers use generative AI at work?](https://evy.so/facts/share-of-workers-using-ai/) - [F-16**78–85%**Which occupations and industries use AI the most?](https://evy.so/facts/ai-use-by-occupation-and-industry/) - [F-17**12–24 pts**Who uses AI at work more, by age, education and income?](https://evy.so/facts/ai-use-by-age-education-income/) - [F-18**5–16 pts**Is there a gender gap in AI use at work, and how large is it?](https://evy.so/facts/gender-gap-in-ai-use/) - [F-19**50–78%**How many workers use AI tools their employer did not provide, and how many hide it?](https://evy.so/facts/bring-your-own-ai/) - [F-20**42–49%**How many employers provide AI tools, guidance or training to staff?](https://evy.so/facts/employers-providing-ai-tools/) [The whole chapter](https://evy.so/facts/who-uses-ai-at-work/) · [Chapter poster](https://evy.so/facts/who-uses-ai-at-work/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "When does AI make work slower or worse? (14–25%) | The AI at Work Fact Book" description: "On tasks AI handles poorly, controlled studies find results 14 to 25 percent worse. Rework, software delivery and misquoted figures, with sources." last_updated: "2026-10-24" source: "https://evy.so/facts/when-ai-makes-work-worse/" -------------------------------------------------------------------------------- F-13 · What AI Saves # When does AI make work slower or worse? On tasks the AI handles poorly, controlled studies find results 14 to 25 percent worse: consultants using GPT-4 gave fewer correct answers, and experienced developers took 19 percent longer. No study measures how often this happens across all AI-assisted work. 14–25% worse on tasks the AI handles poorly (points fewer correct answers, or percent more time) - Medium confidence - 3 primary sources - 2025 to 2026 - Last checked October 14, 2026 ## Cite this EVY (2026). When does AI make work slower or worse? The AI at Work Fact Book, F-13. https://evy.so/facts/when-ai-makes-work-worse/ (last checked October 14, 2026). CC BY 4.0. Copy citation [Data as JSON](https://evy.so/facts/when-ai-makes-work-worse.json) [Markdown](https://evy.so/facts/when-ai-makes-work-worse.md) ## How to read it Three kinds of worse get mixed up: wrong answers, slower work, and extra effort to check and fix output. This range covers the first two, from controlled studies, on tasks where AI was weak or people trusted it too much. Units differ: the BCG figure is percentage points of correct answers, the others percent more time. The same studies show gains on other tasks, and the UK observation had only 3 people per group. ## Every study on one scale Shaded: 14–25 One figure A range in one study Chart values as text - Scale: how much worse (points fewer correct, or % more time). Shaded range: 14–25. - Dell'Acqua et al., BCG (2026): 13.9–24.5 points fewer correct - METR (2025): 19% more time - UK DBT, Excel task (2025): about 22% more time **Why medium confidence:** Three independent creators with different designs each found a penalty in this band on some tasks. Each measures a different outcome, one study has only 3 people per group, and none answers how often it happens across all work. ## Sources Sources for F-13 | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Navigating the Jagged Technological Frontier (Organization Science) | Fabrizio Dell'Acqua, Ethan Mollick, Karim R. Lakhani and colleagues, with Boston Consulting Group | 2026 | 758 BCG consultants; the outside-the-frontier task was run with a subset | Preregistered randomized experiment. One business case was chosen so GPT-4 would give a plausible but wrong answer. | 84.5% correct without AI, 70.6% and 60% with AI: 13.9 and 24.5 points fewer correct answers, while spending less time | [doi.org ↗](https://doi.org/10.1287/orsc.2025.21838) | | Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity | Joel Becker, Nate Rush, Beth Barnes and David Rein (METR) | 2025 | 16 experienced developers, 246 tasks in mature open-source projects | Randomized at the task level, February to June 2025, with screen recordings | 19% more time per task with AI; developers accepted under 44% of AI suggestions and spent more time prompting, waiting and reviewing “we find that allowing AI actually increases completion time by 19%” | [arxiv.org ↗](https://arxiv.org/abs/2507.09089) | | The Evaluation of the M365 Copilot Pilot in the Department for Business and Trade | UK Department for Business and Trade | 2025 | 11 civil servants observed, 3 per group on the Excel task | Timed and blind-graded fictional tasks with matched colleagues as controls, late 2024. The authors warn the samples are very small. | Excel analysis took 25:01 with Copilot against 20:33 without, with lower accuracy (1.5 against 2.7 out of 5) “M365 Copilot users completed Excel data analysis more slowly and to a worse quality” | [assets.publishing.service.gov.uk ↗](https://assets.publishing.service.gov.uk/media/68adbe409e1cebdd2c96a19d/dbt-microsoft-365-copilot-evaluation.pdf) | Rework Low confidence ## How much of the time AI saves goes into fixing its output? Nearly 40 percent, by AI users' own account in one vendor survey: Workday's November 2025 survey of 3,200 AI users at large firms. Only 14% said they consistently get clearly positive results. In a 2024 Upwork survey, 77% of employees who use AI said it had added to their workload in some way, and 39% spend more time reviewing AI output. How much of the time AI saves goes into fixing its output? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | Beyond Productivity: Measuring the Real Value of AI | Workday, survey by Hanover Research | 2026 | 3,200 full-time employees who use AI, at organizations with $100 million or more in revenue | Online survey, November 2025, self-reported. Read in Workday's press release; the full report is gated. Workday sells AI products. | Nearly 40% of AI time savings lost to rework; 14% consistently get clear net-positive results “Nearly 40% of AI time savings are lost to rework” | [newsroom.workday.com ↗](https://newsroom.workday.com/2026-01-14-New-Workday-Research-Companies-Are-Leaving-AI-Gains-on-the-Table) | | From Burnout to Balance: AI-Enhanced Work Models | Upwork Research Institute with Workplace Intelligence | 2024 | 2,500 respondents in the US, UK, Australia and Canada, including 625 full-time employees | Online survey, April to May 2024, self-reported. Read in Upwork's press release. Upwork sells freelance talent. | 77% of employees who use AI say it added to their workload; 39% spend more time reviewing AI content “77% of employees using AI say these tools have added to their workload” | [investors.upwork.com ↗](https://investors.upwork.com/news-releases/news-release-details/upwork-study-finds-employee-workloads-rising-despite-increased-c) | Software delivery ## Does AI make software delivery less stable? Google's DORA surveys link wider AI adoption to less stable software delivery, in both 2024 and 2025. In 2025 the link to throughput turned positive while the stability link held. These are statistical associations from surveys of technology professionals and do not show cause. Does AI make software delivery less stable? | Study | Creator | Year | Sample | Method | Figure | Link | | --- | --- | --- | --- | --- | --- | --- | | 2025 DORA Report: State of AI-assisted Software Development | Google Cloud DORA | 2025 | Nearly 5,000 technology professionals | Annual survey with statistical modeling. Read in Google Cloud's announcement posts for 2024 and 2025. | AI adoption linked to lower software delivery stability; 30% report little or no trust in AI-generated code “AI adoption does continue to have a negative relationship with software delivery stability” | [cloud.google.com ↗](https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report) | ## Numbers you may have seen These figures circulate widely. Here is what the original source says. - **Upwork: 77% of workers say AI has decreased their productivity.** The survey says 77% of employees who use AI said it added to their workload, in a group of 625 full-time employees. It does not measure productivity. - **Workday: workers lose 40% of their time fixing AI.** The survey says nearly 40% of the time AI saves is lost to rework. As a share of the whole work week it is far smaller. - **AI makes consultants 19% worse (Dell'Acqua et al.).** The drop applies to one task chosen to sit outside GPT-4's abilities. On 18 other tasks the same consultants did better with AI. - **DORA: every 25% increase in AI adoption cuts delivery throughput by 1.5% and stability by 7.2%.** These are survey associations and do not show cause. In 2025 the throughput link turned positive; only the stability link held. ## More from What AI Saves - [F-08**3–5.4%**How much time does AI save knowledge workers?](https://evy.so/facts/time-saved-by-ai-at-work/) - [F-09**25–40%**How much faster do people finish writing tasks with AI, and is the work better?](https://evy.so/facts/ai-writing-speed-and-quality/) - [F-10**14–15%**How much more do customer support agents get done with AI?](https://evy.so/facts/ai-customer-support-productivity/) - [F-11**21–26%**How much faster do developers code with AI assistants?](https://evy.so/facts/ai-coding-speed/) - [F-12**27–46%**Does AI help less-experienced workers more than experienced ones?](https://evy.so/facts/ai-helps-less-experienced-workers-more/) [The whole chapter](https://evy.so/facts/what-ai-saves/) · [Chapter poster](https://evy.so/facts/what-ai-saves/sheet/) · [All of The AI at Work Fact Book](https://evy.so/facts/) -------------------------------------------------------------------------------- title: "Where the Week Goes: statistics on lost work time | The AI at Work Fact Book" description: "Chapter 1 of The AI at Work Fact Book. How knowledge workers lose time before the real work starts: searching, coordinating, meetings, interruptions, email, app switching and duplicated work." last_updated: "2026-10-24" source: "https://evy.so/facts/where-the-week-goes/" -------------------------------------------------------------------------------- - F-0119–25% ## [How much of the work week do knowledge workers spend searching for information?](https://evy.so/facts/time-spent-searching-for-information/) The most-cited studies put it at 19 to 25 percent of the work week, or about 9 to 10 hours. Across all five studies the estimates run from about 5 to 15 hours a week, because each one defines searching a little differently. Medium confidence 5 sources Last checked October 13, 2026 [Sources and method →](https://evy.so/facts/time-spent-searching-for-information/) - F-0253–61% ## [How much of the work day goes to coordinating work rather than doing it?](https://evy.so/facts/time-on-work-about-work/) Between 53 and 61 percent, or roughly three of every five working hours. Four independent creators (Asana, McKinsey with IDC data, Microsoft and Workfront) land in that band, though each draws the line around coordination a little differently. High confidence 9 sources Last checked October 13, 2026 [Sources and method →](https://evy.so/facts/time-on-work-about-work/) - F-033–18 h ## [How many hours a week do people spend in meetings, and how much of it is wasted?](https://evy.so/facts/time-in-meetings/) Studies range from about 3 to 18 hours a week for the average office worker, and managers report about 22. Workers judge roughly a third to two thirds of their meeting time unnecessary or unproductive. Medium confidence 6 sources Last checked October 13, 2026 [Sources and method →](https://evy.so/facts/time-in-meetings/) - F-0420–26 min ## [How long does it take to get back to a task after an interruption?](https://evy.so/facts/time-to-resume-after-interruption/) Field studies that watched people at work found it takes about 20 to 26 minutes to return to an interrupted task, usually after handling about two other things first. The often quoted "23 minutes" comes from this research and counts that whole detour. Medium confidence 3 sources Last checked October 13, 2026 [Sources and method →](https://evy.so/facts/time-to-resume-after-interruption/) - F-0513–28% ## [How much of the work week do knowledge workers spend on email and chat?](https://evy.so/facts/time-on-email-and-chat/) Email alone takes 13 to 28 percent of the work week, depending on how it is measured. Microsoft's usage data adds Teams chat at 19 percent of measured time in its apps. Medium confidence 5 sources Last checked October 13, 2026 [Sources and method →](https://evy.so/facts/time-on-email-and-chat/) - F-063–4 h ## [How much time do people lose switching between apps at work?](https://evy.so/facts/time-lost-switching-apps/) About 3 to 4 hours a week goes to getting back on track after switching apps. Logging software (just under 4 hours) and self-reports (36 minutes a day) agree within about an hour. Medium confidence 2 sources Last checked October 13, 2026 [Sources and method →](https://evy.so/facts/time-lost-switching-apps/) - F-074.6–5.7 h ## [How much time do people spend redoing work that already exists?](https://evy.so/facts/time-spent-duplicating-work/) Workers report losing about 4.6 to 5.7 hours a week to duplicated work, or 10 to 14 percent of a 40-hour week. Every figure is self-reported, and two of the three come from the same publisher. Medium confidence 3 sources Last checked October 13, 2026 [Sources and method →](https://evy.so/facts/time-spent-duplicating-work/) -------------------------------------------------------------------------------- title: "Where the Week Goes, on one sheet | The AI at Work Fact Book" description: "Where the Week Goes on one poster: 7 numbers from The AI at Work Fact Book, each with its sources printed on the sheet. Free to reuse under CC BY 4.0." last_updated: "2026-10-24" source: "https://evy.so/facts/where-the-week-goes/sheet/" -------------------------------------------------------------------------------- about 8 to 10 of 40 hours in the week F-01 19–25% of the work week goes to searching for information McKinsey 2012 · Atlassian 2025 · Glean 2022 about 21 to 24 of 40 hours in the week F-02 53–61% of the work day goes to coordinating work Asana · McKinsey · Microsoft · Workfront 3 to 18 of 40 hours in the week F-03 3–18 h a week in meetings; managers report about 22 Otter.ai 2022 · Atlassian 2022 · Doodle 2019 · NBER 2026 20 to 26 of 60 minutes F-04 20–26 min to get back to an interrupted task Mark et al. 2005 · Iqbal and Horvitz 2007 about 5 to 11 of 40 hours in the week F-05 13–28% of the work week goes to email alone McKinsey 2012 · Microsoft 2023 · Workfront 2019 · Adobe 2019 3 to 4 of 40 hours in the week F-06 3–4 h a week lost getting back on track after switching apps HBR 2022 · Qatalog 2021 about 5 to 6 of 40 hours in the week F-07 4.6–5.7 h a week spent redoing work that already exists Asana 2019, 2021 · Panopto 2018 How to read it Each 5 by 8 grid is a 40-hour week: a row per day, a square per hour. The clock is one hour. Solid orange is the low end of the range across studies, light orange runs to the high end. Methods, samples and links for every study are at **evy.so/facts**. ## The numbers on this sheet - [F-01**19–25%**of the work week, about 9 to 10 hours. How much of the work week do knowledge workers spend searching for information? Sources: McKinsey Global Institute (2012), Atlassian (2025), Glean / Harris Poll (2022), Qatalog / Cornell (2021), Coveo (2025).](https://evy.so/facts/time-spent-searching-for-information/) - [F-02**53–61%**of working time, from 4 independent creators. How much of the work day goes to coordinating work rather than doing it? Sources: Asana 2019 (2019), Asana 2021 (2021), Asana 2022 (2022), Asana 2023 (2023), Asana 2024 (2024), Asana 2025 (2025), McKinsey Global Institute (2012), Microsoft (2024), Workfront (2018).](https://evy.so/facts/time-on-work-about-work/) - [F-03**3–18 h**a week in meetings for the average worker; managers about 22 h. How many hours a week do people spend in meetings, and how much of it is wasted? Sources: Doodle (2019), ActivTrak (2025), Deming et al., Norway (2026), Atlassian (2022), Microsoft (top 25%) (2023), Otter.ai / Rogelberg (2022).](https://evy.so/facts/time-in-meetings/) - [F-04**20–26 min**until back on the interrupted task, observed in the field. How long does it take to get back to a task after an interruption? Sources: Mark et al., 2005 (2005), Mark, Gallup interview (2006), Iqbal and Horvitz, 2007 (2007).](https://evy.so/facts/time-to-resume-after-interruption/) - [F-05**13–28%**of the work week on email alone. How much of the work week do knowledge workers spend on email and chat? Sources: Workfront (2019), Microsoft (2023), Adobe (2019), Dillon et al. (Microsoft Research) (2025), McKinsey Global Institute (2012).](https://evy.so/facts/time-on-email-and-chat/) - [F-06**3–4 h**a week lost getting back on track after switching apps. How much time do people lose switching between apps at work? Sources: Murty et al., HBR (2022), Qatalog / Cornell (2021).](https://evy.so/facts/time-lost-switching-apps/) - [F-07**4.6–5.7 h**a week on duplicated work, 10 to 14% of a 40-hour week. How much time do people spend redoing work that already exists? Sources: Asana 2019 (2019), Asana 2021 (2021), Panopto (2018).](https://evy.so/facts/time-spent-duplicating-work/) Using the sheet in a post, deck or article? Credit "The AI at Work Fact Book by EVY" and link to [evy.so/facts/where-the-week-goes/](https://evy.so/facts/where-the-week-goes/). -------------------------------------------------------------------------------- title: "Who Uses AI at Work: statistics on AI use at work | The AI at Work Fact Book" description: "Chapter 3 of The AI at Work Fact Book. Who uses AI at work, how often and for what: adoption, uses, jobs and industries, age and education, gender, bringing your own AI, and what employers provide." last_updated: "2026-10-24" source: "https://evy.so/facts/who-uses-ai-at-work/" -------------------------------------------------------------------------------- - F-1440–50% ## [What share of workers use generative AI at work?](https://evy.so/facts/share-of-workers-using-ai/) About 40 to 50 percent of US workers now use generative AI for their job, depending on how the question is asked. Daily use is much smaller: about 14 to 15 percent of all workers, or a quarter to a third of the people who use it. High confidence 4 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/share-of-workers-using-ai/) - F-1540–52% ## [What do people use AI for at work?](https://evy.so/facts/what-workers-use-ai-for/) Mostly for writing and finding information: about half of AI-using workers use it for writing or editing, and about half for research or search. Coding is a minority use outside tech, named by 13 to 27 percent of AI users in surveys. High confidence 5 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/what-workers-use-ai-for/) - F-1678–85% ## [Which occupations and industries use AI the most?](https://evy.so/facts/ai-use-by-occupation-and-industry/) Tech workers, by a wide margin: about 8 in 10 US workers in computer jobs or the tech industry use AI for work, against about 2 in 10 in hands-on service jobs. The same order shows up in the EU, among EU firms and in AI companies' usage logs. High confidence 5 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/ai-use-by-occupation-and-industry/) - F-1712–24 pts ## [Who uses AI at work more, by age, education and income?](https://evy.so/facts/ai-use-by-age-education-income/) College graduates use AI at work 12 to 24 percentage points more than workers without a degree, and workers 50 and older trail younger colleagues by 4 to 20 points in the US. The income gap is the widest: across 47 countries, 83 percent of high-income employees use AI at work against 34 percent of low-income employees. High confidence 5 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/ai-use-by-age-education-income/) - F-185–16 pts ## [Is there a gender gap in AI use at work, and how large is it?](https://evy.so/facts/gender-gap-in-ai-use/) Most studies find men use generative AI more than women, by about 5 to 16 percentage points, which makes men about 16 to 22 percent more likely to use it. The gap shrinks by a quarter to a third among people in the same job, and stays open. Medium confidence 5 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/gender-gap-in-ai-use/) - F-1950–78% ## [How many workers use AI tools their employer did not provide, and how many hide it?](https://evy.so/facts/bring-your-own-ai/) Half to three quarters of workers who use AI use at least some tools their employer did not provide, and most of them also use approved tools. About half are reluctant to tell their manager or have avoided revealing their AI use. Medium confidence 3 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/bring-your-own-ai/) - F-2042–49% ## [How many employers provide AI tools, guidance or training to staff?](https://evy.so/facts/employers-providing-ai-tools/) About half of workers say their employer has brought in AI tools, 42 to 49 percent in 2024 to 2026, but only 25 to 40 percent report a clear AI plan or policy. Counted by firms instead of workers, adoption is lower: 20 percent of EU firms and 24 percent of US businesses use AI, because most firms are small. Medium confidence 4 sources Last checked October 14, 2026 [Sources and method →](https://evy.so/facts/employers-providing-ai-tools/) -------------------------------------------------------------------------------- title: "Who Uses AI at Work, on one sheet | The AI at Work Fact Book" description: "Who Uses AI at Work on one poster: 7 numbers from The AI at Work Fact Book, each with its sources printed on the sheet. Free to reuse under CC BY 4.0." last_updated: "2026-10-24" source: "https://evy.so/facts/who-uses-ai-at-work/sheet/" -------------------------------------------------------------------------------- 40 to 50 of 100 F-14 40–50% of US workers use generative AI for their job Gallup 2026 · Pew 2026 · St. Louis Fed 2026 40 to 52 of 100 F-15 40–52% of AI-using workers use it for writing or editing Gallup 2026 · Pew 2025 · OpenAI 2025 78 to 85 of 100 F-16 78–85% of US tech workers use AI for work; hands-on service jobs about 20% St. Louis Fed 2026 · Gallup 2026 · Eurostat 2025 a gap of 12 to 24 points out of 100 F-17 12–24 pts more AI use at work among college graduates Pew 2025 · St. Louis Fed 2025 · KPMG and Melbourne 2025 a gap of 5 to 16 points out of 100 F-18 5–16 pts more men than women use generative AI Cranney et al. 2026 · St. Louis Fed 2025 · Denmark 2025 50 to 78 of 100 F-19 50–78% of AI-using workers use tools their employer did not provide Microsoft 2024 · KPMG and Melbourne 2025 · Software AG 2024 42 to 49 of 100 F-20 42–49% of workers say their employer has brought in AI tools Gallup 2026 · KPMG and Melbourne 2025 How to read it Each 10 by 10 grid is 100%. Solid orange is the low end of the range across studies, light orange runs to the high end. Methods, samples and links for every study are at **evy.so/facts**. ## The numbers on this sheet - [F-14**40–50%**of US workers use generative AI for their job (2026); studies range from 38 to 52%. What share of workers use generative AI at work? Sources: St. Louis Fed survey (2026), Gallup (2026), Pew (2026), Pew, stricter question (2025).](https://evy.so/facts/share-of-workers-using-ai/) - [F-15**40–52%**of AI users use it for writing or editing, the most common use. What do people use AI for at work? Sources: Gallup (survey) (2026), Pew (survey) (2025), Bick, Blandin and Deming (survey) (2025), OpenAI (usage logs) (2025), Microsoft Research (usage logs) (2025).](https://evy.so/facts/what-workers-use-ai-for/) - [F-16**78–85%**of US tech workers use AI for work; hands-on service jobs about 15 to 25%. Which occupations and industries use AI the most? Sources: St. Louis Fed survey, computer jobs (2026), Gallup, technology industry (2026), Eurostat, ICT professionals (EU) (2026), Eurostat, ICT firms (EU) (2025), Humlum and Vestergaard, Denmark (2024).](https://evy.so/facts/ai-use-by-occupation-and-industry/) - [F-17**12–24 pts**more AI use at work among college graduates than among workers without a degree. Who uses AI at work more, by age, education and income? Sources: Pew (2025), Bick, Blandin and Deming (2025), KPMG and Melbourne, 47 countries (2025), Pew, ChatGPT (2025), Humlum and Vestergaard, Denmark (2025).](https://evy.so/facts/ai-use-by-age-education-income/) - [F-18**5–16 pts**more men than women use generative AI; men are 16 to 22% more likely. Is there a gender gap in AI use at work, and how large is it? Sources: Cranney, Delecourt and Koning (2026), Bick, Blandin and Deming, US (2025), Humlum and Vestergaard, Denmark (2025), Pew (2025), KPMG and Melbourne (2025).](https://evy.so/facts/gender-gap-in-ai-use/) - [F-19**50–78%**of AI-using workers use at least some tools their employer did not provide (2024). How many workers use AI tools their employer did not provide, and how many hide it? Sources: Microsoft and LinkedIn (2024), KPMG and Melbourne (2025), Software AG (2024).](https://evy.so/facts/bring-your-own-ai/) - [F-20**42–49%**of workers say their employer has brought in AI tools; 25 to 40% report a clear plan or policy. How many employers provide AI tools, guidance or training to staff? Sources: Gallup, AI tools (2026), Gallup, clear plan (2026), KPMG and Melbourne (2025), Gallup, guidelines (2025).](https://evy.so/facts/employers-providing-ai-tools/) Using the sheet in a post, deck or article? Credit "The AI at Work Fact Book by EVY" and link to [evy.so/facts/who-uses-ai-at-work/](https://evy.so/facts/who-uses-ai-at-work/). -------------------------------------------------------------------------------- title: "Pricing: EVY is free | EVY" description: "EVY is free for you and your whole team. Its agents run on the Claude or ChatGPT plan you already pay for, so there is no second AI bill." last_updated: "2026-10-24" source: "https://evy.so/pricing/" -------------------------------------------------------------------------------- Pricing # EVY is free Free for you and your whole team. EVY's agents run on the Claude or ChatGPT plan you already pay for, so there is no second AI bill. €0 for the whole team [Get EVY free for Mac](https://evy.so/download/) - To-do for people and agents in one list - Direction: goals and initiatives for each space - A space for every company or project you run - Agents on your Mac: Claude Code and Codex, with your plan and your context - Company knowledge kept as files any agent can read - Talk to EVY: hold ⌥⌘ and speak Paying for Notion AI today? [See what a team of 12 pays in Notion and in EVY](https://evy.so/compare/evy-vs-notion/#price). ## Questions about price Is EVY really free? Yes. EVY is free for you and your whole team. There is no paid plan and no card to enter. Do I need a Claude or ChatGPT subscription? To hand tasks to agents, yes. EVY runs Claude Code on your Claude plan and Codex on your ChatGPT plan, so the AI you already pay for does the work and there is no second AI bill. How does that compare with Notion AI? Notion's full AI set needs the Business plan at €19.50 per member a month billed yearly, so a team of 12 pays €234 a month before agent credits. EVY is €0 for the same team. The full comparison is at evy.so/compare/evy-vs-notion/. Does EVY work on Windows or in the browser? Not yet. EVY is a Mac app for now. Do I need to create an account? No. You set EVY up inside the desktop app, without a sign-up. -------------------------------------------------------------------------------- title: "Privacy Policy | EVY" description: "How EVY handles your data: what stays on your Mac, what EVY's servers in Germany store, which AI providers run EVY's features, and your rights." last_updated: "2026-10-24" source: "https://evy.so/privacy-policy/" -------------------------------------------------------------------------------- # Privacy policy Last updated: 21 October 2026 This policy explains what data EVY collects, why, where it goes and what you can do about it. It covers the website evy.so, the EVY desktop app for macOS, the EVY web app at app.evy.so and the EVY phone app for iPhone and Android (listed in the stores as “EVY: AI todo list” or “EVY”). **The short version** - The files in your space folders stay on your Mac. EVY’s servers store your plans, tasks, notes and chats so EVY can plan with you. - EVY’s servers are in Germany. - Agents such as Claude Code and Codex run on your Mac with your own Claude or ChatGPT plan. - EVY’s own AI features use the providers listed below. We do not use your content to train AI models. - The phone app shows no ads, has no advertising ID and does not track you across other apps. - We never sell your data. Email [help@evy.so](mailto:help@evy.so) to get a copy of your data or have it deleted. ## Who is responsible EVY is operated by Helix Studio OÜ, Pärnu mnt. 139c, 11317 Tallinn, Estonia (registry code 14249878). Helix Studio OÜ is the controller of your personal data under the GDPR. Contact us at [help@evy.so](mailto:help@evy.so). ## What we collect ### On the website - **Access requests.** Your name, company email, team size, how you heard about EVY if you tell us, and where you came from (for example a search engine or an AI assistant). We use this to bring you onto EVY and to see which channels work. - **Connection data.** Your IP address and browser details pass through Cloudflare when you visit. - **Analytics cookies**, only if you agree in the cookie banner. With the same consent, your browser tab remembers which site sent you, so an access request made a few pages later still counts that source. This note is cleared when you close the tab. ### Your account - When you open the desktop app for the first time, EVY creates an account for you without an email or password. A key is stored in your Mac’s Keychain. Our server keeps only a fingerprint of that key and your Mac’s name. - You can add an email later. You need one to invite people. - On the web app you sign in with an email and password, or with Google (we receive your email address and name). - The phone app signs in when you scan the Connect your phone code in the desktop app. There is no password on the phone. Its sign-in key sits in the phone’s secure storage (the Keychain on iPhone, the Keystore on Android). ### What you put into EVY Your spaces, goals, initiatives, plans, tasks, notes, assets, comments, your chat with EVY and the transcripts of what you say to EVY. This is stored on EVY’s servers so EVY can plan with you and so your team can work in the same space. ### From your Mac - **Folders you link to a space:** the folder path, the number of files, up to 25 file names, the first lines of a README, and the git remote. For files inside space folders: size, dates and a checksum. The contents of your files stay on your Mac. - **Projects on your Mac:** the names, folders, remotes and last commit dates of code projects in common project folders, so EVY can suggest spaces. - **Agent work:** the tasks you hand to agents, their results, their questions, your answers, and how much each run used. - **Agent conversations in linked folders:** when you work with Claude Code or Codex in a folder linked to EVY, EVY’s plugin can send your messages and the agent’s answers (shortened, without tool output) so EVY can keep your space up to date. We delete these excerpts once they are summarised. ### On your phone The phone app shows your To-do, the questions agents ask you and voice notes from your team. It keeps only what it needs to show the screen you are on, and a recording until it has been sent. - **Microphone:** for voice notes to teammates, Talk to EVY and tasks you speak. The audio goes to EVY and is turned into text in the same way as voice input on your Mac (see Voice below). - **Camera:** only to read the Connect your phone code when you sign in. No picture is stored or sent. - **Push notifications:** if you allow them, EVY stores your phone’s push token and the phone’s name. EVY sends a push only when an agent asks you a question, when a voice note arrives, or for a To-do reminder you set. Pushes travel through Expo’s push service and then Apple’s or Google’s (listed below). When you sign out on the phone, EVY removes its token. - **Crash reports:** when the app crashes, it sends a report to Sentry in the EU, without personal data and without screen recordings. - **No ads or tracking:** the phone app has no ads, reads no advertising ID and does not track you across other companies’ apps or websites. It does not record which screens you open. ### Voice When you hold the shortcut, use Talk to EVY or speak a task on your phone, the audio goes to our server and is turned into text by OpenAI. We keep the text and discard the audio. Voice notes you send to teammates are kept for 24 hours, then deleted. ### Calendar If you connect Google Calendar, EVY reads it with read-only access and keeps the title, start, end and busy status of events from two days back to three weeks ahead. Your access token is stored encrypted. ### How you use the app In the web app and the desktop app we record which screens you open, what you click and which steps you take, such as finishing setup, making a note or starting a task, linked to your EVY account. We also keep short screen recordings of your sessions to see where EVY is hard to use. In these recordings every piece of text and everything you type is hidden, so your notes, tasks and messages never reach us this way; button labels such as the names of your spaces and folders can still appear. Nothing for this is stored on your device, so it needs no cookies. To have your account left out, email [help@evy.so](mailto:help@evy.so). ### Technical data Error reports, crash reports from the phone app, server health data, and the app’s update checks, which tell us which version you run. ## How AI is used **Agents on your Mac.** Claude Code, Codex and other agents run on your Mac under your own accounts with Anthropic, OpenAI or the agent’s provider. What they send to their provider is governed by your agreement with that provider, which EVY does not control. **EVY’s own features.** Some features run on EVY’s servers and call AI providers: answers when you talk to EVY, the morning check-in, planning suggestions, transcription, search and sorting. These providers receive only what the feature needs and are listed below. We use them through their business APIs, and we do not use your content to train AI models. ## Why we use your data (legal basis) - **To provide EVY** (contract): your account, your content, agent work, voice, calendar, push notifications and the emails you need. - **To keep EVY working and secure** (legitimate interest): error and crash reports, server monitoring, abuse prevention. - **To understand where people come from** (legitimate interest): the source of access requests. - **To improve EVY** (legitimate interest): usage analytics and screen recordings in the app, with all text hidden. - **Analytics cookies** (consent), which you can withdraw at any time in the cookie settings. ## Who receives data We share data only with the providers that help run EVY, under data processing agreements, and never sell it. | Provider | Location | What for | | --- | --- | --- | | **Hetzner Online** | Germany (Nuremberg) | Hosts EVY’s servers and database: your account, spaces, plans, tasks, notes and chat with EVY. | | **Amazon Web Services (S3)** | EU or USA | Stores uploaded files, images and voice notes, and encrypted database backups. | | **Cloudflare** | Global network | Delivers the website and the app, and protects them from attacks. Sees connection data such as IP addresses. | | **Supabase** | Germany (Frankfurt) | Stores what you send through the forms on evy.so, such as access requests. | | **OpenRouter, routing to Anthropic and OpenAI** | USA | Runs EVY’s own AI features: Talk to EVY answers, the morning check-in, planning help and summaries. Receives the content these features need, such as your goals, plan, tasks and calendar entries. | | **OpenAI** | USA | Turns your voice into text, powers search across your notes, and helps file notes and suggest folders. | | **TypeSafe** | USA | Sorts voice commands, notes and assets. | | **Cartesia** | USA | Reads EVY’s answers aloud when you talk to EVY. | | **Groq** | USA | Writes short summaries of text. | | **Google** | USA | Sign in with Google, and Google Calendar if you connect it. | | **Expo (650 Industries)** | USA | Passes EVY’s push notifications on to Apple and Google for delivery to your phone. Receives your phone’s push token and the text of the notification. | | **Apple Push Notification service** | USA | Delivers EVY’s push notifications to iPhones. | | **Google Firebase Cloud Messaging** | USA | Delivers EVY’s push notifications to Android phones. | | **Loops** | USA | Sends EVY’s emails, such as confirmations and invitations. Receives your email address and name. | | **Sentry** | USA | Error reports from the website, the web app and the desktop app, which can include your IP address and user id. | | **Sentry** | EU (Germany, Frankfurt) | Crash reports from the phone app, without personal data and without screen recordings. | | **Datadog** | EU | Monitors the health of EVY’s servers. | | **Slack** | USA | Notifies our team about new sign-ups and access requests. | | **PostHog** | EU (Germany) | Usage analytics in the web app and the desktop app: the screens you open, what you click and screen recordings in which all text and everything you type is hidden. Keeps nothing on your device. | | **Google Analytics** | USA | Usage analytics on the website and web app, only after you agree in the cookie banner. Off in the desktop app. | Some of these providers process data outside the European Economic Area, mainly in the USA. Where they do, we rely on the safeguards the GDPR provides for this, such as the EU Standard Contractual Clauses or the EU-US Data Privacy Framework. ## How long we keep it - Your account and content: as long as you use EVY. When you ask us to delete your account, we delete it within 30 days, and it leaves our backups as they rotate. - Agent conversation excerpts: deleted once summarised. - Voice notes: 24 hours. Voice audio for transcription: not kept. - Screen recordings of app sessions: 30 days. - Calendar copy: two days back to three weeks ahead, refreshed as time moves. - Phone push tokens: until you sign out on that phone or your account is deleted. - Access requests: until you are onboarded or ask us to remove them, and at most 12 months. ## Your rights You can ask us for a copy of your data, to correct it, to delete it, to restrict or object to how we use it, and to receive it in a portable format. Email [help@evy.so](mailto:help@evy.so) and we will answer within 30 days. To delete your account, follow the steps on [Delete your EVY account](https://evy.so/delete-account/). You can also complain to a data protection authority, for example the Estonian Data Protection Inspectorate (Andmekaitse Inspektsioon) or the authority where you live. ## Security Traffic to EVY is encrypted. Keys on your Mac sit in the macOS Keychain, and the server stores only fingerprints of them. The phone app keeps its sign-in in the phone’s secure storage. Calendar tokens and your own API keys are stored encrypted, and database backups are encrypted. ## Children EVY is for people aged 18 and over. ## Changes When we change this policy, we update the date at the top. If a change matters for how we use your data, we tell you in the app or by email before it takes effect. -------------------------------------------------------------------------------- title: "Support | EVY" description: "Get help with EVY. Email help@evy.so for setup, agents, integrations or your data. A small team answers its own email." last_updated: "2026-10-24" source: "https://evy.so/support/" -------------------------------------------------------------------------------- # Support We are a small team and we answer our own email. ## Get in touch Email [help@evy.so](mailto:help@evy.so) with anything: a question, a bug, something you wish EVY did, or help getting set up. We aim to reply within one business day. It helps if you tell us what you were trying to do, what happened instead, and add a screenshot if something looked wrong. ## Getting EVY - EVY is a Mac app and needs macOS 14 or later. - [Download EVY](https://evy.so/download/) after a short form with your name, company email and team size. - EVY sets up your first space with you when you open it. You don’t need an account or a card. ## Agents EVY hands tasks to agents that run on your Mac with the AI plan you already have: - **Claude Code**, with a Claude plan - **Codex**, with a ChatGPT plan If neither is installed, EVY offers to install it for you. Agents only work in the folders you give them, and they ask you when they get stuck. ## Integrations EVY connects to other services only when you ask it to. You can disconnect any of them at any time in the app under **Settings → Integrations**, and it takes effect immediately. ### Google Calendar Read-only. EVY sees when you are busy so it can plan your day around your meetings. It cannot create, change or delete your events. ### Slack Lets EVY post in a channel you choose and message the people in it. EVY can read the list of channels and the names of their members so it knows where to post. **EVY cannot read your team’s messages.** It does not request message history, file access or email addresses. You can remove EVY from your workspace at any time in Slack’s app management. ### LinkedIn Lets EVY publish posts to your profile when you choose. EVY does not read your feed, your connections or your messages. ## Your data Our [privacy policy](https://evy.so/privacy-policy/) explains what we store and why. To get a copy of your data, email [help@evy.so](mailto:help@evy.so) and we will send it within 30 days. ## Delete your EVY account Email [help@evy.so](mailto:help@evy.so) from your account’s address and ask us to delete it. [Delete your EVY account](https://evy.so/delete-account/) lists what we delete and what we keep. -------------------------------------------------------------------------------- title: "Terms of Service | EVY" description: "The terms for using EVY, the free workspace for people and agents: your account, your content, agents on your Mac and your own AI plans." last_updated: "2026-10-24" source: "https://evy.so/terms-of-service/" -------------------------------------------------------------------------------- # Terms of service Last updated: 21 October 2026 These terms apply when you use EVY: the desktop app for macOS, the phone app for iPhone and Android, the web app at app.evy.so and the website evy.so. By using EVY you agree to them. If you use EVY for a company, you agree on its behalf and confirm you may do so. ## 1\. Who we are EVY is provided by Helix Studio OÜ, Pärnu mnt. 139c, 11317 Tallinn, Estonia, registry code 14249878 (“we”, “us”). Contact: [help@evy.so](mailto:help@evy.so). ## 2\. What EVY is EVY plans work with you, hands tasks to people and to AI agents, and keeps your company’s knowledge in one place. EVY is new and changes often, and we may add, change or remove features as we go. ## 3\. Price EVY is free. If we ever offer paid features, we will tell you in advance, and nothing will cost you money unless you agree to it. ## 4\. Your account The desktop app creates your account on your Mac and keeps its key in your Keychain. Keep your Mac and your sign-in details safe, because anyone using them can act in your EVY account. Tell us at [help@evy.so](mailto:help@evy.so) if you think someone else has access. You must be at least 18 to use EVY. ## 5\. Agents and your own AI plans - Agents such as Claude Code and Codex run on your Mac under your own accounts with Anthropic, OpenAI or the agent’s provider. Your agreement with that provider applies to that use, including its fees and limits. EVY is not affiliated with these providers. - You choose what you hand to an agent and which folders it may work in. Agents can create, change and delete files in those folders and can make mistakes. Review their work before you rely on it, and keep backups of anything important. - Answers from EVY and from agents are generated by AI and can be wrong. Check anything that matters, especially legal, financial or medical questions. ## 6\. Your content What you put into EVY stays yours: plans, tasks, notes, files and everything agents produce for you. You give us permission to store, process and show it only as needed to run EVY for you and the people you share a space with. Our [privacy policy](https://evy.so/privacy-policy/) explains how we handle it. You are responsible for having the right to use the content you add. ## 7\. Fair use Do not use EVY to break the law, to harm others, to send spam, to get into systems or data you have no right to, or to disrupt EVY or attempt to work around its limits. We may limit or suspend access when someone does. ## 8\. Other services EVY connects to services such as Google Calendar, and agents may use other tools. Those services have their own terms, and we are not responsible for them. ## 9\. Availability We work to keep EVY running, but we provide it as it is and as available, without a promise that it will always be free of errors or interruptions. We may change or stop EVY. If we stop it, we will give you reasonable notice and a way to get your content out. ## 10\. Liability EVY is free, so to the extent the law allows, we are not liable for indirect losses, lost profits or lost data, and our total liability is limited to EUR 100. Nothing in these terms limits liability that cannot be limited by law, such as liability for intent, gross negligence or injury to life and health, or your mandatory rights as a consumer. ## 11\. Ending You can stop using EVY at any time and ask us to delete your account at [help@evy.so](mailto:help@evy.so) (see [Delete your EVY account](https://evy.so/delete-account/)). We may end or suspend your access if you break these terms. ## 12\. Changes to these terms We may update these terms as EVY changes. We will update the date at the top, and we will tell you in the app or by email before a significant change takes effect. If you keep using EVY after that, the new terms apply. ## 13\. Law Estonian law applies, and the courts of Estonia decide disputes. If you are a consumer, you keep the protection of the mandatory laws of the country where you live and may also bring a claim there. -------------------------------------------------------------------------------- title: "Free tools | EVY" description: "Free tools from EVY, like the voice file generator that turns your writing into a voice.md any AI can follow." last_updated: "2026-10-24" source: "https://evy.so/tools/" -------------------------------------------------------------------------------- - [Free tool **Voice file generator** Paste a few things you wrote and get a voice.md any AI can follow.](https://evy.so/voice/) -------------------------------------------------------------------------------- title: "Trust: where your data lives and who sees it | EVY" description: "No sign-up, your notes as Markdown files on your Mac, servers in Germany, agents on your own Claude or ChatGPT plan, and no training on your content. Each answer with its source." last_updated: "2026-10-24" source: "https://evy.so/trust/" -------------------------------------------------------------------------------- - [No sign-up EVY makes an account for your Mac. No email, no password.](#sign-up) - [Notes as Markdown Every note sits on your Mac as a plain file any tool can read.](#markdown) - [Servers in Germany Hetzner, Nuremberg. The company is in Tallinn, Estonia.](#storage) - [Your own AI plan Agents work on the Claude or ChatGPT plan you already pay for.](#ai) - [No training on your content We never train AI on it, and we never sell your data.](#training) Where your data lives ## Three places, and what each one holds On your Mac ### Yours, and stays there - The files in your space folders - Every note as Markdown, with its history - Claude Code or Codex, on your own plan notes copied to your Mac EVY’s servers · Germany ### What EVY needs to plan with you - Plans, tasks, notes and chats, at Hetzner in Nuremberg - Uploads and voice notes, with Amazon S3 - For links to folders: names, sizes, dates only what a feature needs AI providers · USA ### Anthropic and OpenAI - Your agents: under your own plan and terms - EVY’s features: through business APIs - No training on your content Every provider, its location and what it receives is listed in the [privacy policy](https://evy.so/privacy-policy/). Plain files ## Open your notes in anything EVY’s helper writes each space’s notes into its folder on your Mac and keeps them current. No export, no lock-in: the copy is already there. - **Knowledge** holds the space’s shared notes, **Private notes** your own, meetings included. - Each file starts with its title, when it changed and who changed it, so an agent knows what it reads. - For now the copy runs from EVY to your Mac. Edits you make there are kept on a separate git branch, so nothing is lost. Northwind › Knowledge › Clients - Knowledge - Clients - fernway-onboarding.md - brightpath-renewal.md - Playbooks - investor-update.md - Private notes - weekly-planning.md ``` --- title: Fernway onboarding updated_at: 2026-10-14 09:12 updated_by: Mara --- # Fernway onboarding Kick-off moves to Thursday. Ana owns the CSV export, due 29 September. ## Open questions - Single sign-on before launch? ``` Questions, answered ## The details, with sources ### Do I need to sign up? No. The first time you open EVY on your Mac, it creates an account for that Mac without an email or password. A key stays in your Mac’s Keychain, and our server keeps only a fingerprint of it. You add an email later, when you want to invite people. - [Privacy policy, “Your account”](https://evy.so/privacy-policy/) ### Can I take my notes with me? They are already on your Mac. EVY writes every note as a plain Markdown file into your space folder, in the same folders you see in EVY, and updates the copy every few minutes with full history in git. Any editor, search tool or AI agent can read it, and it stays yours if you stop using EVY. - [How EVY’s helper syncs your notes](#files) ### Where is my data stored? Your files stay on your Mac. EVY’s servers, at Hetzner in Nuremberg, Germany, store your plans, tasks, notes and chats so EVY can plan with you and your team can share a space. Uploaded files and voice notes are stored with Amazon S3. From folders you link to a space, EVY reads only the path, file names, sizes and dates. - [Privacy policy, “What you put into EVY”](https://evy.so/privacy-policy/) ### Which AI sees my data? Agents such as Claude Code and Codex run on your Mac under your own Claude or ChatGPT plan, so that work goes to Anthropic or OpenAI under your agreement with them. EVY’s own features, such as Talk to EVY, the morning check-in, transcription and search, call AI providers in the USA, mainly Anthropic and OpenAI, with only the content each feature needs. - [Privacy policy, “How AI is used”](https://evy.so/privacy-policy/) - [EU AI data residency research](https://evy.so/compare/eu-ai-data-residency/) ### Is my data used to train AI models? We do not use your content to train AI models. EVY calls its providers through their business APIs, and OpenAI and Anthropic state that they do not train on data sent that way by default. - [OpenAI, enterprise privacy](https://openai.com/enterprise-privacy/) - [Anthropic, commercial terms](https://www.anthropic.com/legal/commercial-terms) ### Who is behind EVY? EVY is built by a small team in Europe and operated by Helix Studio OÜ, a company registered in Tallinn, Estonia, under EU law. You can reach the people who build it at help@evy.so. - [Imprint](https://evy.so/imprint/) - [About EVY](https://evy.so/about/) ### What happens to my data if EVY stops? Your notes are already on your Mac as Markdown with their history, your space folders never left it, and your agents run on your own AI plan. Nothing you need is locked inside EVY. If you want your account gone, email help@evy.so and we delete it within 30 days. - [Privacy policy, “How long we keep it”](https://evy.so/privacy-policy/) ### What does EVY record about how I use it? In the app we record which screens you open and what you click, and keep short screen recordings with every piece of text and everything you type hidden. Error reports help us fix bugs. On the website, analytics run only if you agree in the cookie banner. - [Privacy policy, “How you use the app”](https://evy.so/privacy-policy/) ## Something we didn’t answer? Email the people who build EVY. If your question shows a gap on this page, we add the answer here. [help@evy.so](mailto:help@evy.so) - Helix Studio OÜ - Tallinn, Estonia - EU law -------------------------------------------------------------------------------- title: "Make Any AI Write Like You: Free voice.md Generator · EVY" description: "Paste a few things you've written and get a free, downloadable voice.md file: your personal writing voice plus EVY's anti-AI humanization layer, so ChatGPT, Claude, or any AI stops sounding like a robot." last_updated: "2026-10-24" source: "https://evy.so/voice/" -------------------------------------------------------------------------------- What goes in your file ## Your file, built to sound human Your distilled voice tells the AI how you sound. On top of that, EVY adds an anti-AI layer that kills the tells every model leaves behind. Toggle what goes in your file. Your voice profile Always included, distilled from your samples below Writing rules 12 rules that read like a human, not a model Formatting rules Clean structure, and never an em dash Banned AI phrases 40+ dead giveaways the model must never use ### Paste a few things you've written Three or more short pieces work best: articles, LinkedIn posts, emails, landing-page copy. The more it sounds like the real you, the sharper your profile. Pre-AI writing is gold. Add another sample Generate voice.md 0/200 words We read your samples to build your profile. We don't store the text you paste. Distilling your voice… Mapping rhythm, vocabulary, and your signature tics. Your voice archetype voice.md Copy Download Your full voice.md is ready Drop your email to unlock the download. We'll send you a couple more ways to make AI work like you. Unlock & download Stop pasting it in by hand Keep your voice.md in EVY and every agent you hand writing to, Claude Code or Codex, gets it as part of your context. Free, on the AI plan you already pay for. [Get EVY free for Mac →](https://evy.so/download/?utm_source=voice-md) ← Start over with new samples ## What is a voice.md file? A voice.md file is a short, portable set of instructions that teaches an AI to write in your voice. You paste it into ChatGPT, Claude, or Gemini and the drafts come back sounding like you, not like a generic model. It is plain markdown, so it works anywhere you can paste text. AI writing sounds robotic because models reach for the same safe words and the same predictable shapes. Your file replaces that default with how you actually write, and bans the dead giveaways: LeverageDelveGame-changerIt's worth notingSuperchargeLet that sink inThis isn't X, it's Y and 30+ more ## How to use your voice.md 1 ### Distil your voice Paste a few things you've written and download your voice.md. Takes about a minute. 2 ### Paste it into your AI Drop the file into ChatGPT, Claude, or Gemini, as a system prompt or at the top of your chat. 3 ### Write as usual Ask for a post, an email, a reply. It comes back in your voice, not the model's default. ## Frequently asked questions ### What is a voice.md file? A voice.md file is a short, portable set of instructions that teaches an AI to write in your voice. You paste it into ChatGPT, Claude, or Gemini and the drafts come back sounding like you instead of like a generic model. It is plain markdown, so it works anywhere you can paste text. ### How do I make ChatGPT or Claude write like me? Give the model your own writing turned into clear rules. This tool reads a few things you've written, distils your rhythm, vocabulary, and phrasing, adds an anti-AI humanization layer, and hands you a voice.md file to paste into any AI. The model then pattern-matches to you instead of its default style. ### Why does AI writing sound robotic? Models reach for the same safe phrases (leverage, delve, 'it's worth noting') and the same predictable structures. Readers, and increasingly platform algorithms, can spot it. A voice.md file replaces that default with your voice and bans the dead giveaways outright. ### Is it free, and do you store my writing? Distilling your voice and previewing it is free; you enter an email to download the file. We read your samples to build your profile and do not store the text you paste. ### Which AI tools does it work with? Any text AI: ChatGPT, Claude, Gemini, and most others. The file is plain markdown you paste into the chat or set as a system or custom instruction. ### What makes this different from other voice generators? Most tools just describe your style. Every EVY voice.md also ships a humanization layer: writing rules, formatting rules, and 40+ banned AI phrases that kill the tells. You choose which sections go in your file. ## Keep going [ Pick your model ### The best LLMs for writing You picked your voice. Now pick the model that writes it best, ranked. ](https://evy.so/compare/best-llms-for-writing/)[ See the numbers ### What AI saves at work Time saved on writing and other work, measured in trials and surveys. ](https://evy.so/facts/what-ai-saves/) ## Your voice, in every agent you use A voice.md works wherever you paste it. The catch is pasting it every time, into every chat and every agent. EVY is the free workspace for people and agents. Keep your voice.md in a space's knowledge, and Claude Code or Codex get it as part of your context whenever you hand them writing, on the Claude or ChatGPT plan you already have. [Get EVY free for Mac →](https://evy.so/download/?utm_source=voice-md) -------------------------------------------------------------------------------- title: "Comparisons and rankings | EVY" description: "AI models and tools compared side by side, including the best LLMs for writing, ranked weekly on three benchmarks." last_updated: "2026-10-24" source: "https://evy.so/compare/" -------------------------------------------------------------------------------- - [Ranking **Best AI for business work in 2026** One score for knowledge work, built from GDPval-AA, AA-Briefcase, Vals Finance and Vals Legal. Updated 24 October 2026](https://evy.so/compare/best-ai-for-business/) - [Ranking **Best AI workspace for small teams in 2026** Nine AI workspaces scored on price, agents, planning, knowledge, voice, team work and platforms. Change the weights to fit your team. Updated 24 October 2026](https://evy.so/compare/best-ai-workspace/) - [Comparison **EVY vs Notion AI** Price for a team of 12, whose AI does the work, agents, planning and team voice notes, side by side. Updated 12 October 2026](https://evy.so/compare/evy-vs-notion/) - [Ranking **Best LLMs for writing in 2026** Claude, GPT and Gemini scored for creative writing and copywriting, updated every week.](https://evy.so/compare/best-llms-for-writing/) - [Comparison **EU AI data residency** Where each major AI tool processes and stores your data.](https://evy.so/compare/eu-ai-data-residency/) -------------------------------------------------------------------------------- title: "Best AI for Business Work in 2026: Ranked on Real Task Benchmarks" description: "The best AI for business and knowledge work in 2026, ranked by one score built from GDPval-AA, AA-Briefcase, the Vals Finance Index and the Vals Legal Index. Cost per task included." last_updated: "2026-10-24" source: "https://evy.so/compare/best-ai-for-business/" -------------------------------------------------------------------------------- ## The short answer Updated October 2026 Claude Opus 5.5 (Anthropic) and Muse Wonder Full (Meta) are effectively tied at the top, with Claude Sonnet 5.5 next. The top two are separated by a gap of 1.0 points, inside the margin of error, so treat them as tied. For the best result per dollar, **DeepSeek V4.1 Flash 1017** (DeepSeek) scores 62 at $0.26 per task, about 23× cheaper than Claude Opus 5.5. Full ranking below. 1. 1 Claude Opus 5.5 Anthropic Work Score 72 2. 2 Muse Wonder Full Meta Work Score 71 3. 3 Claude Sonnet 5.5 Anthropic Work Score 70 ## How the EVY Work Score works Each benchmark tests a different slice of professional work, and none of them has tested every model. Averaging the scores a model happens to have would reward models that skipped the hard tests. The Work Score fixes that in three steps: 1. **One scale.** Elo ratings and percentages are converted to the same units, so they can sit side by side. 2. **Difficulty adjustment.** The calculation learns how hard each benchmark is and how well it separates strong models from weak ones. Doing well on a hard benchmark counts for more. 3. **Gaps filled by estimate.** A model's score is the ability level that best explains the results it has. Models with fewer results are pulled toward the average and shown with a wider margin, so one lucky number cannot put a model on top. The benchmarks agree with each other: on models that have both, their scores correlate between 0.78 and 0.96. That agreement is why one combined number is meaningful. A score of about 50 is average across the models tracked here, and the ± after each score is one standard error. Gaps smaller than that margin are ties. Show benchmark details Hide benchmark details - GDPval-AA [Artificial Analysis](https://artificialanalysis.ai/evaluations/gdpval-aa) runs OpenAI's GDPval tasks: real work from 44 occupations in 9 industries. Two anonymised answers are compared by an LLM judge and aggregated into an Elo rating. 27 models. About 96% of its variation is explained by general work ability. - AA-Briefcase [Artificial Analysis](https://artificialanalysis.ai/evaluations/aa-briefcase) runs a private test of long, multi-step knowledge work: realistic business workflows that end in deliverables such as spreadsheets, presentations and memos. Each model is scored at its best effort setting. 33 models, matched to the models tracked here. About 94% explained by general work ability. - Vals Finance [Vals AI](https://www.vals.ai/benchmarks/finance_index) combines Finance Agent, EMB and Tax Agent tests into one accuracy score. 30 models. About 87% explained by general work ability. - Vals Legal [Vals AI](https://www.vals.ai/benchmarks/legal_index) combines case-law questions across practice areas with legal research and agent-style legal tasks. 27 models. About 75% explained by general work ability. Known limits: GDPval-AA and AA-Briefcase come from the same evaluator and agree closely (r = 0.96), so the general-work part of the score leans on one method. The Finance and Legal indices both include Vals' Tax Agent test, a small overlap we accept. Muse Wonder Full has a Work Score but no public API, so it has no price or cost per task. Benchmarks are run by third parties on current model versions and can change when they update their tasks. A dash (–) means no score exists, never an estimated one. Prices come from Artificial Analysis and are standard API prices per 1M tokens. [Related **The model is half the answer.** See how the AI workspaces that run these models compare. →](https://evy.so/compare/best-ai-workspace/) -------------------------------------------------------------------------------- title: "Best AI workspace for small teams in 2026, scored and ranked" description: "Nine AI workspaces for small teams scored on price, agents, planning, knowledge, voice, team work, platforms and integrations. Change the weights to match your team. Every score has its reason." last_updated: "2026-10-24" source: "https://evy.so/compare/best-ai-workspace/" -------------------------------------------------------------------------------- Ranking · updated 24 October 2026 # Best AI workspace for small teams Nine tools that want to be the one place a small team plans and works with AI, scored on nine things. EVY, ChatGPT Business and Notion Business come out level at the top. Change what matters to you and the order changes with it. By [Eugen Petersen](https://www.linkedin.com/in/eugenpetersen), co-founder of EVY. We make EVY, so every tool gets the same rubric and every score shows its reason. Prices checked 12 October 2026. ## The short answer 30-second read Pick Notion if your team's work lives in docs, wikis and databases, or you need Windows and phones. Pick ChatGPT Business or Claude Team if you want one strong AI for everyone, with agents and coding included. Choose Claude when agents and coding matter most. Pick EVY if you run more than one thing, work on Macs and already pay for Claude or ChatGPT. EVY plans the day and hands the work to that plan, for nothing extra. Pick ClickUp, Google or Microsoft if your day runs on tasks and sprints (ClickUp), or on Gmail (Google) or Outlook and Office files (Microsoft). Price for 10 people From **$0** (EVY) to **$320** a month (Microsoft 365 Copilot), billed yearly with the AI included. $200 is the going rate for Notion, ChatGPT Business and Claude Team. The scorecard ## Nine tools, nine criteria Each score runs from 0 to 4 against a published rubric. Pick the weighting closest to your team, or set your own, and the ranking updates. BalancedSeveral projects, own AI planDocs and wikisAgents firstLowest costWorking together all dayWindows or mixed devices Set your own weights Cost 1 Your AI plan 1 Agents 1 Planning 1 Knowledge 1 Voice & meetings 1 Team 1 Platforms 1 Integrations 1 Every criterion counts the same. Nine AI workspaces scored from 0 to 4 on nine criteria, ranked by weighted score | Tool | Score | Cost | Your AI plan | Agents | Planning | Knowledge | Voice & meetings | Team | Platforms | Integrations | 10 people | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | 1 e **EVY**Mac users who run several projects on their own Claude or ChatGPT plan | **75** | | | | | | | | | | $0 | | 2 C **ChatGPT Business**One AI everyone on the team uses, coding included | **72** | | | | | | | | | | $200 | | 3 N **Notion Business**Teams whose work lives in docs, wikis and databases | **72** | | | | | | | | | | $200 | | 4 C **Claude Team**Teams that hand real work to agents | **67** | | | | | | | | | | $200 | | 5 K **ClickUp with Brain AI**Teams that run on tasks, sprints and dashboards | **67** | | | | | | | | | | $160 | | 6 G **Google Workspace with Gemini**Teams already living in Gmail, Docs and Meet | **67** | | | | | | | | | | $140 | | 7 M **Microsoft 365 Copilot**Microsoft shops | **64** | | | | | | | | | | $320 | | 8 T **Taskade**A whole team on AI agents for one flat price | **61** | | | | | | | | | | $25 | | 9 m **Motion**Teams whose problem is the calendar | **58** | | | | | | | | | | $190 | ← Scroll sideways for all columns → Hover a dot for the reason behind it. Every reason is listed under the tool further down. Scores reflect what each tool ships today. On equal weights EVY ranks 1st of 9 with 75 points, level with the leaders. It ranks 1st when you weigh "Several projects, own AI plan" and 9th when you weigh "Windows or mixed devices". Both are true, which is why the weights are yours to change. ### Where EVY lands under each weighting | Weighting | EVY ranks | Leader | | --- | --- | --- | | BalancedEvery criterion counts the same. | **1st** of 9, 75 points | EVY | | Several projects, own AI planYou run more than one thing, already pay for Claude or ChatGPT and want the AI to plan and do the work. | **1st** of 9, 95 points | EVY | | Docs and wikisYour team's work lives in shared documents and databases. | **9th** of 9, 60 points | Notion Business | | Agents firstYou want to hand real jobs to AI and get finished work back. | **1st** of 9, 89 points | EVY | | Lowest costEvery dollar counts. | **1st** of 9, 92 points | EVY | | Working together all daySeveral people edit, comment and hand work to each other in the same place. | **9th** of 9, 56 points | Notion Business | | Windows or mixed devicesYour team is not all on Macs. | **9th** of 9, 48 points | Notion Business | What EVY does ## The part a scorecard can't show A score compresses a lot. Here is what sits behind EVY's Planning, Agents, Voice and Knowledge marks, shown from the app. ### Plans the day across everything you run Most workspaces store tasks. EVY decides what comes first each morning, splits the day between your companies and projects, and moves work when something runs over. It is the part the Planning column rewards. [How the AI chief of staff works →](https://evy.so/ai-chief-of-staff/) EVY Morning. Fernway gets half the day: the Olea pitch is on Thursday, so Write the pitch story comes first, and Claude takes Pull three hotel case studies. Saltmarsh needs half an hour: Approve the October menu before it goes to print. Kiln & Co. can run on Auto today. **Your day is set**4 added · 1 movedOpen To-do Keep Wednesday afternoon free, I’m with the kids. EVY Done. Wednesday’s two Fernway tasks moved to the morning, and I told Mara. Tell EVY what matters today ### Hands the work to the AI you already pay for Tasks go to Claude Code, Codex or Hermes on your Mac, with the plan and your company knowledge loaded. Each agent works in the open, asks when it is stuck and hands back a result you check. There is no second AI bill. [How agents work through your To-do →](https://evy.so/ai-to-do-list/) **@codex**Loopline · Beta launch Needs input #### Sort 240 beta sign-ups 1. 10:09Read the 240 sign-ups from the waitlist sheet 2. 10:11Grouped them by team size and what they want to fix 3. 10:1238 of them use a calendar tool we don’t connect to yet **@codex asks:** Invite those 38 first, or keep the waitlist order? Invite them firstKeep the order **@hermes**Brightwire Digital · Clients Done #### Weekly ad reports for six clients 1. 07:30Pulled last week’s ad results for six clients 2. 07:33Compared each one to its monthly budget and goal 3. 07:36Wrote a one-page report per client in the agency’s template 4. 07:38Added two follow-ups to To-do: two clients are over budget 6 reports readyView ### Keeps what your company knows in plain files Notes, decisions and context live as Markdown in a folder per space. Agents read them before every task, and any other AI or tool can read them too. If you stop using EVY, the files stay. [How company knowledge works →](https://evy.so/ai-knowledge-base/) **Loopline** Talk to EVY⌥⌘ Direction To-do Assets Private Meetings Seed round Shared Customers Product Sales How we work Socials Ideas Posts **Customers**14New **Brightpath: single sign-on options**12m Start with Okta. Brightpath uses it, and so do two of the five leads… **Brightpath check-in**3h Ana, Ravi and Mia from Brightpath. CSV export is done, sign-on next… **Halden Labs onboarding plan**Yesterday Three workshops in November, one for each team, starting with support… **Nordwave renewal**Oct 12 Renews on 1 December. They want annual billing and two more seats… **Churn interviews, September**Oct 9 Six calls. Four left because setup took too long for a small team… **How we run customer calls**Oct 7 Agenda goes in the invite. EVY files the notes here after the call… **Pricing for bigger teams**Oct 3 Above 50 seats we offer annual only, with sign-on included… **Brightpath: what they bought**Oct 2 40 seats on the annual plan, signed 14 March by their head of product… Created by **Claude** · Edited by **Ravi**, 12m ago••• From the task Compare single sign-on providers ### Brightpath: single sign-on options ###### Recommendation Start with Okta. Brightpath uses it, and so do two of the five leads in the pipeline. Google Workspace is the quicker build and can follow in December. ###### What Brightpath asked for - CSV export of all feedback - Single sign-on before their support team joins, 25 more seats ###### Effort - Okta: about four days, including user sync - Google Workspace: about two days - Microsoft Entra: about five days, no customer is asking yet Sources: Brightpath check-in on 29 September, the Okta and Google developer docs. ### Everything EVY does today All of it free for you and your whole team. Mac only for now. - **Morning check-in**EVY asks what matters today and sets your To-do. - **Goals and spaces**One space per company or project, each with goals and initiatives. - **A day split across everything**Each space gets its share of the day, and work moves when plans change. - **Tasks for people or agents**Claude Code, Codex, Hermes or Aside take a task with the plan and your knowledge. - **Auto**Agents work down the list in plan order, a few at a time. - **Agents that ask**A stuck agent stops and asks at the top of your To-do, with buttons to answer. - **Repeat work**Weekly reports and checks come back on their own day. - **Talk to EVY**Hold ⌥⌘ anywhere on the Mac and speak. Dictation is built in. - **Meeting transcripts**Record a meeting and get the transcript and the gist. - **Voice notes for teammates**Say who it is for and talk. Urgent notes sit apart from the rest. - **Shared notes**Private and Shared notes. Shared ones edit live with everyone in them. - **Knowledge in plain files**Company knowledge is Markdown in a folder per space. Any AI can read it, and you can leave with it. On the way_Windows__A phone companion app__Ask EVY from Slack or Telegram_ The nine, in detail ## What each tool is good at, and where it falls short A free Mac app, for you and your whole team. Each space (one per company or project) holds goals and initiatives, and EVY plans the day with you at a morning check-in. Tasks go to people or to agents: Claude Code, Codex or Hermes do them on the AI plan you already have, with the plan and your company's knowledge handed over. **Price:** $0 for the whole team (free, on AI plans you already have). **Falls short:** Mac only for now, with Windows coming. No custom databases or big wikis. The youngest tool here, with a fraction of the integrations and track record of the others. You need a Claude or ChatGPT plan, or a Hermes agent, to put agents to work. [How agents work through your To-do →](https://evy.so/ai-to-do-list/) All nine scores and why - **Cost** 4 of 4. $0. No paid plan, for the whole team. - **Your AI plan** 4 of 4. Agents run on your Claude or ChatGPT plan. - **Agents** 4 of 4. Claude Code, Codex or Hermes get the plan and your knowledge. Auto works down the list and agents ask when stuck. - **Planning** 4 of 4. Goals, initiatives, a morning check-in and a day split across spaces. - **Knowledge** 3 of 4. Notes and company knowledge as plain files any agent reads. No databases. - **Voice & meetings** 4 of 4. Hold ⌥⌘ to talk, dictation, meeting transcripts and voice notes to teammates. - **Team** 2 of 4. Shared notes edit live, with a shared plan and voice notes. No roles or comments. - **Platforms** 1 of 4. Mac only. Windows is coming. - **Integrations** 1 of 4. The youngest tool here, with few integrations so far. Every seat gets ChatGPT, ChatGPT Work and Codex on web, desktop and mobile. It connects to Microsoft 365, Google Drive, Slack, GitHub, Linear and Figma, and teams can build plugins on their company knowledge. **Price:** $200 a month for 10 people ($20 a user, billed yearly). **Falls short:** An AI first and a workspace second. There is no team task list or goals, so planning still lives elsewhere. All nine scores and why - **Cost** 1 of 4. $200 a month for 10 people. $20 a user, billed yearly, from 2 users. - **Your AI plan** 4 of 4. It is the plan. - **Agents** 4 of 4. ChatGPT Work and Codex on every seat. - **Planning** 0 of 4. No team task list or goals. - **Knowledge** 3 of 4. Company knowledge and shared projects. - **Voice & meetings** 3 of 4. Record transcribes and summarises meetings in the Mac app. - **Team** 3 of 4. Shared workspace with admin controls. - **Platforms** 4 of 4. Web, desktop and mobile. - **Integrations** 4 of 4. Connects to Microsoft 365, Drive, Slack, GitHub, Linear and Figma. Still the shared brain most small teams reach for. The Business plan carries Notion Agent, AI Meeting Notes and Enterprise Search, with Custom Agents on top at $10 per 1,000 credits. **Price:** $200 a month for 10 people ($20 a member, billed yearly). **Falls short:** The AI is Notion's and billed by Notion, on top of whatever Claude or ChatGPT plan the team has. You build your own planning system from templates and keep it alive yourself. [EVY vs Notion AI →](https://evy.so/compare/evy-vs-notion/) All nine scores and why - **Cost** 1 of 4. $200 a month for 10 people. $20 a member, billed yearly. - **Your AI plan** 1 of 4. Notion bills the AI, and Custom Agents run on credits. - **Agents** 3 of 4. Notion Agent and Custom Agents act inside Notion. - **Planning** 2 of 4. You build the planning system from templates. - **Knowledge** 4 of 4. Docs, wikis, databases and Enterprise Search. - **Voice & meetings** 3 of 4. AI Meeting Notes on the Business plan. - **Team** 4 of 4. Comments, permissions and live editing. - **Platforms** 4 of 4. Web, Mac, Windows, iOS and Android. - **Integrations** 4 of 4. Years at scale with a large ecosystem. Claude Code and Claude Cowork on every seat, plus Claude Design, Slides and Docs, enterprise search across the organisation and the Microsoft 365 integration. Claude's pricing page now says "Claude Cowork is now just Claude", so agent mode is folding into the main app. **Price:** $200 a month for 10 people ($20 a seat, billed yearly). **Falls short:** Like ChatGPT, it has no shared plan or task list, so the plan lives somewhere else. All nine scores and why - **Cost** 1 of 4. $200 a month for 10 people. $20 a seat, billed yearly. Premium seats with 5x the usage cost $100. - **Your AI plan** 4 of 4. It is the plan. - **Agents** 4 of 4. Claude Code and Cowork on every seat. - **Planning** 0 of 4. No shared plan or task list. - **Knowledge** 3 of 4. Projects and enterprise search. - **Voice & meetings** 2 of 4. Dictation on mobile. No meeting recording found. - **Team** 3 of 4. Shared projects, with standard and premium seats. - **Platforms** 4 of 4. Web, desktop and mobile. - **Integrations** 3 of 4. Microsoft 365 integration and enterprise search. Tasks, docs, chat and dashboards in one app. The Brain AI add-on brings Brain Assistant, the @Brain Agent and AI chat with Claude, ChatGPT and Gemini models inside ClickUp. Super Agents and the AI Notetaker come with the Everything AI add-on at $28 a user. **Price:** $160 a month for 10 people (Unlimited $7 plus Brain AI $9, per user, billed yearly). **Falls short:** Two price lists (the plan and the AI) plus credits for agent work. It is a lot of app for a team of four. All nine scores and why - **Cost** 1 of 4. $160 a month for 10 people. $7 for Unlimited plus $9 for Brain AI, per user. - **Your AI plan** 0 of 4. Brain AI is an add-on, and agent work runs on credits. - **Agents** 3 of 4. @Brain Agent, and Super Agents on credits. - **Planning** 3 of 4. Tasks, sprints and dashboards. - **Knowledge** 3 of 4. Tasks, docs and dashboards the AI can read. - **Voice & meetings** 2 of 4. AI Notetaker comes with the $28 Everything AI add-on. - **Team** 4 of 4. Comments, roles, dashboards and chat. - **Platforms** 4 of 4. Web, desktop and mobile. - **Integrations** 4 of 4. Long track record and a large app ecosystem. Every Business plan now includes Gemini: in Gmail, as a chat assistant and in Google Vids. For a team that already pays for Workspace email, the AI comes at no extra seat cost, which makes it the cheapest AI here per person. **Price:** $140 a month for 10 people (Business Standard $14 a user). **Falls short:** Gemini helps inside each app but does not run the team's plan, and agents that finish work across apps are thin next to Claude or ChatGPT. All nine scores and why - **Cost** 2 of 4. $140 a month for 10 people. Business Standard at $14 a user. - **Your AI plan** 2 of 4. Gemini is in the base plan at no extra cost, but it is Gemini only. - **Agents** 2 of 4. Gemini helps inside each app. - **Planning** 1 of 4. Calendar and Tasks only. - **Knowledge** 3 of 4. Docs and Drive. - **Voice & meetings** 3 of 4. Gemini takes notes in Meet on Business Standard. - **Team** 4 of 4. Shared docs, comments and admin controls. - **Platforms** 3 of 4. Web and mobile. - **Integrations** 4 of 4. Gmail, Docs, Meet and a large add-on ecosystem. Copilot Business adds Copilot to Word, Excel, Outlook and Teams, with Copilot Studio for building agents. Copilot Business is $21 a user a month billed yearly (on offer for $18 right now), on top of a Microsoft 365 plan. **Price:** $320 a month for 10 people (Business Premium with Copilot, $32 a user). **Falls short:** The most expensive option here, and it only pays off if the team already works in Office files. Anthropic models offered inside Copilot run outside Microsoft's EU Data Boundary. All nine scores and why - **Cost** 0 of 4. $320 a month for 10 people. Business Premium with Copilot at $32 a user. - **Your AI plan** 0 of 4. Copilot is a $21 add-on on top of the Microsoft 365 plan. - **Agents** 3 of 4. Copilot Studio agents act on Microsoft 365 data. - **Planning** 1 of 4. Planning lives in separate Microsoft apps. - **Knowledge** 4 of 4. Office files, Teams and SharePoint. - **Voice & meetings** 3 of 4. Copilot works inside Teams. - **Team** 4 of 4. Shared files, Teams and admin controls. - **Platforms** 4 of 4. Web, Windows, Mac and mobile. - **Integrations** 4 of 4. The default business suite at scale. Projects, docs, AI agents, automations and small AI apps, with a Business plan that includes unlimited users. That makes it the cheapest team plan here by a distance. **Price:** $25 a month for 10 people (Business, unlimited users). **Falls short:** The AI runs on Taskade's credits, and heavy agent use means buying packs. A smaller company with a smaller ecosystem than everyone above it. All nine scores and why - **Cost** 3 of 4. $25 a month for 10 people. Business at $25 a month with unlimited users. - **Your AI plan** 1 of 4. AI runs on monthly credits, with packs from $5. - **Agents** 3 of 4. AI agents and automations, on credits. - **Planning** 2 of 4. Projects and tasks, without a goals layer. - **Knowledge** 2 of 4. Projects, docs and AI apps. - **Voice & meetings** 1 of 4. No native meeting recording confirmed. - **Team** 3 of 4. Unlimited users on Business. - **Platforms** 4 of 4. Web, desktop and mobile. - **Integrations** 3 of 4. A smaller company with a smaller ecosystem. The AI task planner puts the team's tasks into their calendars, next to meetings. It also has projects, docs, a wiki and AI chat, with 7,500 AI credits per seat a month on Pro AI. **Price:** $190 a month for 10 people (Pro AI $19 a seat, billed yearly). **Falls short:** Priced per seat like Notion, with its own AI credits. The planning is about time slots, and agents that do the work are limited. [Motion alternatives →](https://evy.so/compare/motion-alternatives/) All nine scores and why - **Cost** 1 of 4. $190 a month for 10 people. Pro AI at $19 a seat, billed yearly. - **Your AI plan** 1 of 4. AI sits in a higher tier and runs on 7,500 credits a seat. - **Agents** 2 of 4. AI planner and chat. Agents that do the work are limited. - **Planning** 3 of 4. The AI planner schedules tasks into calendars. - **Knowledge** 2 of 4. Docs and a wiki. - **Voice & meetings** 2 of 4. AI Notetaker is included on Pro AI and runs on credits. - **Team** 3 of 4. Team calendars and shared projects. - **Platforms** 4 of 4. Web, desktop and mobile. - **Integrations** 3 of 4. Established, with a smaller ecosystem. ### Also considered - **Linear:** the best issue tracker for software teams, with agents built in. Too narrow for a whole small company. $10 to $16 a user. - **Coda:** a strong doc-and-app tool, now part of Superhuman. It rarely came up in AI answers in 2026. - **Slack and Asana:** both have AI, and a team still needs a workspace next to them. How we scored ## The rubric, in the open We make EVY, so the method is published and the order is computed, never placed by hand. ### What keeps it fair - One rubric for all nine tools, shown below. - Every score has its reason next to it, taken from the vendor's own pages. - Equal weights until you change them. The ranking is computed from the scores, so no tool is placed by hand. - EVY's weakest criteria are scored as they are: Platforms 1 of 4, Integrations 1 of 4, Team 2 of 4. ### What we did not do - Test answer quality. It depends on the model behind each tool, and the models change monthly. - Score enterprise plans. Everything here is a small-team plan at list price. - Count discounts, usage credits or taxes. - Spot something out of date? Write to [support](https://evy.so/support/) and we fix it. Show the full rubric | Criterion | 4 | 3 | 2 | 1 | 0 | | --- | --- | --- | --- | --- | --- | | CostWhat does a 10-person team pay each month, with the AI included? | $0 | Up to $50 | Up to $150 | Up to $250 | Over $250 | | Your AI planDoes it work with the Claude or ChatGPT plan you already pay for? | Runs on your Claude or ChatGPT plan, or is that plan | AI is in the base price with no add-on or credits, but it is the vendor's own model | AI is in a higher tier and metered by credits | AI is a separate paid add-on | AI is a separate paid add-on and metered by credits | | AgentsCan it hand a multi-step job to an agent and get finished work back? | Claude Code or Codex class agents that work across files and apps | Built-in agents that act on the product's own data, often metered | An assistant that helps inside each app | Chat and writing help only | No AI that acts | | PlanningDoes it plan the work across several projects, not only store tasks? | Goals, a daily plan and a check-in across several projects | Strong task planning, with scheduling or dashboards | Build your own planning from projects or templates | Tasks and reminders only | No planning layer | | KnowledgeCan the AI read your company's docs, notes and files? | Docs, wikis, databases or Office files with AI search across all of them | Shared notes, projects or a drive the AI can read | A lighter doc or wiki layer | Files you upload by hand | None | | Voice & meetingsCan you talk to it, and does it capture meetings? | Talk to it anywhere, dictate, record and transcribe meetings, and send voice notes to teammates | Records and summarises meetings on the plan | Dictation, or a notetaker that is limited or an add-on | Nothing confirmed | None | | TeamHow well does a team work together in it? | Comments, roles and permissions, and live editing | Shared workspaces with admin controls | Shared notes and a shared plan, without roles or comments | Mostly one person per account | Single user | | PlatformsWhere does it run? | Web, Mac, Windows and phone | Web and phone | Desktop on more than one system, or one desktop system plus phone | One desktop system only | None | | IntegrationsHow mature is it, and what does it connect to? | Years at scale and a large app ecosystem | Established, with the main connectors | Smaller company or ecosystem | New, with few connectors | Very early | Choosing ## How to pick in five minutes 1. **Most of your work is docs and wikis?**Notion. 2. **You want one strong AI and your other tools stay as they are?**ChatGPT Business or Claude Team. Pick Claude if agents and coding matter most. 3. **You run several projects, work on Macs and already pay for Claude or ChatGPT?**Try EVY first. It costs nothing and puts that plan to work on your goals. 4. **You run on tasks and sprints?**ClickUp. 5. **Gmail or Outlook is the center of your day?**Google Workspace or Microsoft 365, whichever you already pay for. 6. **Your team uses Windows or mixes devices?**Anything on this list except EVY for now. Windows is coming. ## Questions What is an AI workspace? An AI workspace is the app where a team keeps its plans, docs and tasks, with AI that can read all of it and act on it. Notion, ClickUp, Google Workspace and Microsoft 365 added AI to documents and tasks. ChatGPT Business and Claude Team start from the AI and give a team shared agents. EVY plans the work and hands tasks to Claude Code, Codex or Hermes. What is the best AI workspace for a small team? EVY, ChatGPT Business and Notion Business score within 3.5 points of each other on equal weights, so the right pick depends on the team. Notion suits teams that live in docs and databases, ChatGPT Business or Claude Team suit teams that want one strong AI, and EVY suits Mac teams that run several projects on their own Claude or ChatGPT plan. Is there a free AI workspace for teams? EVY is free for the whole team, with no paid plan, and runs its agents on your own Claude or ChatGPT plan, or on a Hermes agent. It is a Mac app, with Windows coming. Notion and ClickUp have free plans with a limited AI trial, and Taskade has a free plan too. Is a plain AI assistant like ChatGPT or Claude enough, or do we need a workspace? For a team of 2 or 3 that mostly writes, researches and codes, the best AI assistant on a team plan is often enough. Once you have goals, a task list and company knowledge that the AI should know about, you need somewhere to keep them, which is the job of a workspace like Notion, ClickUp or EVY. Which team collaboration tools have AI built in? Notion (Notion Agent, AI Meeting Notes), ClickUp (Brain AI add-on, $9 a user a month), Google Workspace (Gemini in every Business plan), Microsoft 365 (Copilot Business add-on, $21 a user a month), Motion (AI planner and calendar) and Taskade (AI agents and automations). Slack and Asana also have AI, and they work best next to one of these. Can AI agents really do work for a small team in 2026? Yes, for well-scoped tasks. Claude Code, Claude Cowork, Codex and ChatGPT Work research, write, edit files and run multi-step jobs on a computer. The hard part is giving them the plan and the context, which is why the scorecard weighs planning and knowledge next to the agents themselves. We make EVY. Why should we trust this ranking? You shouldn't have to. Every tool is scored on the same rubric, every score has its reason next to it, and the weights are equal until you change them. EVY ranks 9th of 9 when you weigh Windows, integrations and team features, and the page says where EVY is the wrong pick. Spot a score that is out of date and we will fix it. How often is this ranking updated? Prices and plan facts were checked on 12 October 2026, and meeting and voice facts on 24 October 2026. We recheck when a vendor changes its pricing page, and the date at the top of the page moves with every update. Sources ## Where the facts come from Prices and plan facts checked on 12 October 2026, unless a source says otherwise. - [Notion pricing](https://www.notion.com/pricing) - [ChatGPT pricing, Business and Enterprise](https://chatgpt.com/pricing/?type=team) - [OpenAI Help: Record in ChatGPT](https://help.openai.com/en/articles/11487532-record-mode-in-chatgpt) read 24 Oct 2026 - [Claude pricing](https://claude.com/pricing) - [ClickUp pricing](https://clickup.com/pricing) - [Google Workspace pricing](https://workspace.google.com/pricing) - [Google Workspace Updates: Take notes for me, in-person meetings](https://workspaceupdates.googleblog.com/2026/08/take-notes-with-me-for-in-person-meetings-is-now-available.html) read 24 Oct 2026 - [Microsoft 365 Copilot for business](https://www.microsoft.com/en-us/microsoft-365/copilot/business) - [Motion pricing](https://www.usemotion.com/pricing) - [Motion AI Notetaker overview](https://help.usemotion.com/motion-ai-notetaker-and-motion-docs/motions-ai-notetaker-overview) read 24 Oct 2026 - [Taskade pricing](https://www.taskade.com/pricing) - [Linear pricing](https://linear.app/pricing) - [EVY pricing](https://evy.so/pricing/) and [brand facts](https://evy.so/about/#facts) More comparisons ## Keep reading [**EVY vs Notion AI**Price for 12 people, agents, planning and team voice notes](https://evy.so/compare/evy-vs-notion/) [**Motion alternatives**Planning tools compared](https://evy.so/compare/motion-alternatives/) [**Best AI for business work**The models behind these tools, ranked](https://evy.so/compare/best-ai-for-business/) [**The AI chief of staff guide**How to set one up](https://evy.so/guides/ai-chief-of-staff/) ## See what EVY does with your Claude or ChatGPT plan Free for you and your whole team. Read how it works first, then try it when you are ready. [How EVY works](https://evy.so/?utm_source=best-ai-workspace) [The AI to-do list →](https://evy.so/ai-to-do-list/) -------------------------------------------------------------------------------- title: "Best LLMs for Writing 2026: Creative, Copywriting & Content" description: "The best LLMs for writing in 2026, ranked for creative writing, copywriting, and content. Scored on EQ-Bench, LMArena, and Artificial Analysis. Updated weekly." last_updated: "2026-10-24" source: "https://evy.so/compare/best-llms-for-writing/" -------------------------------------------------------------------------------- ## The short answer Updated October 2026 The best LLM for writing right now is **GPT-6 Astra** (OpenAI), which tops the EQ-Bench Creative Writing leaderboard ahead of Claude Fable 5.1 and Claude Opus 5. For the best quality per dollar, **GLM-5.3** (Zhipu AI) delivers near-frontier writing at a fraction of the cost. Full ranking across 52 models below. 1. 1 GPT-6 Astra OpenAI EQ Creative 2173 2. 2 Claude Fable 5.1 Anthropic EQ Creative 2162 3. 3 Claude Opus 5 Anthropic EQ Creative 2133 ## How we rank models for writing This ranking combines three independent data sources to give the most complete picture of writing quality across frontier LLMs. No single benchmark captures the full picture — so we aggregate: Show benchmark details Hide benchmark details - EQ Creative [EQ-Bench Creative Writing](https://eqbench.com/) — specialist benchmark using trained raters to assess narrative quality, emotional depth, prose style, and character voice. Elo scale ~1300–2175. The most relevant signal for marketing copy, long-form content, and creative work. - Arena Text [LMArena Text](https://arena.ai/leaderboard/text) — crowd-sourced human preference leaderboard. Broad signal across all text tasks: a model that consistently wins votes is generally pleasant, clear, and useful to read. Elo scale ~1346–1507. - EQ General [EQ-Bench General](https://eqbench.com/) — measures emotional intelligence in roleplay scenarios. A proxy for character voice quality and tonal control — useful for brand voice work. Note: high EQ-General does not automatically mean strong creative writing; interpret alongside EQ Creative. - Speed [Artificial Analysis](https://artificialanalysis.ai/models) — median output tokens per second across providers. Matters for iterative draft workflows where waiting costs time. ~75 tokens ≈ 55 words. Prices are per 1M tokens (input / output) and reflect standard API pricing. A dash (—) means the model has not yet appeared on that leaderboard — never an estimated or interpolated value. [Related **The model is half the answer.** See how the AI workspaces that run these models compare. →](https://evy.so/compare/best-ai-workspace/) -------------------------------------------------------------------------------- title: "EU AI Data Residency 2026: Which Providers Keep Your Data in Europe" description: "Which AI providers actually process and store data inside the EU, checked line by line against vendor documentation. Residency, retention, training use, and the exceptions each vendor writes down." last_updated: "2026-10-24" source: "https://evy.so/compare/eu-ai-data-residency/" -------------------------------------------------------------------------------- ## The short answer Sources checked 12 October 2026 Three answers exist, and vendors rarely separate them for you. 1. 1 **EU inference with no geography caveat.** Scaleway, OVHcloud, IONOS, STACKIT and Aleph Alpha, all EU owned. The OpenAI API with a project set to Europe, Vertex AI on a regional endpoint, and Bedrock in an EU region also qualify, under US ownership. 2. 2 **Conditional.** Microsoft Copilot and Azure sit inside the EU Data Boundary, and Microsoft documents the circumstances in which data still leaves it. 3. 3 **No EU option.** The first-party Claude API offers "us" and "global" for inference and US only for storage. Claude with EU residency exists, through Bedrock or Vertex, not through Anthropic. Hosting location is one input among several. A processing agreement, a sub-processor list, a retention setting and a training exclusion decide as much as the map does. ## The comparison Every cell comes from the vendor's own documentation, linked in the last column with the date it was read. Where a vendor does not publish an answer, the row says so. | Provider | Owner HQ | EU inference | EU storage | Training on your content | Retention default | Source | | --- | --- | --- | --- | --- | --- | --- | | Microsoft Copilot (Microsoft 365) Microsoft | United States US owned | Conditional | Conditional | No. Prompts, responses and data accessed through Microsoft Graph are not used to train the foundation models. | Prompts and responses are stored as Copilot activity history and governed by your Microsoft 365 retention policies. | [Microsoft Learn, Data, Privacy and Security for Microsoft Copilot](https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy) Checked 2026-09-06 | | Azure (EU Data Boundary services) Microsoft | United States US owned | Conditional | Conditional | Microsoft states customer data is not used to train its foundation models. | Set by the service you deploy and your own configuration. | [Microsoft Learn, What is the EU Data Boundary?](https://learn.microsoft.com/en-us/privacy/eudb/eu-data-boundary-learn) Checked 2026-09-06 | | OpenAI API (Platform) OpenAI | United States US owned | Yes | Yes | API inputs and outputs are not used to train models by default. | Abuse monitoring logs retained up to 30 days unless you are approved for ZDR. | [OpenAI API docs, Your data](https://developers.openai.com/api/docs/guides/your-data) Checked 2026-10-12 | | Claude API (first party) Anthropic | United States US owned | No | No | Commercial API inputs and outputs are not used to train models by default. | Governed by the workspace and by Zero Data Retention where it applies. | [Claude Platform docs, Data residency](https://platform.claude.com/docs/en/manage-claude/data-residency) Checked 2026-10-12 | | Gemini Enterprise Agent Platform (formerly Vertex AI) Google | United States US owned | Yes | Yes | Google states customer data is not used to train its models without permission. | Set by the service configuration and your project settings. | [Google Cloud docs, Data residency (Gemini Enterprise Agent Platform, formerly Vertex AI)](https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/data-residency) Checked 2026-09-21 | | Amazon Bedrock AWS | United States US owned | Yes | Yes | Model providers have no access to customer prompts and completions. | Governed by your account configuration and the abuse detection documentation. | [AWS docs, Data protection in Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/data-protection.html) Checked 2026-10-12 | | La Plateforme Mistral AI | France EU owned | Conditional | Conditional | Opt out available. Check the terms for the tier you are on. | 30 rolling days by default, zero with ZDR on supported endpoints. | [Mistral Help Center and Privacy Policy](https://help.mistral.ai/en/collections/789667-data-governance) Checked 2026-10-12 | | Generative APIs Scaleway (Iliad) | France EU owned | Yes | Yes | Not used for training, retraining or improving the base models, and not accessible to the model creators. | Prompt content is not stored unless the service requires it, as with batch processing. | [Scaleway docs, Generative APIs privacy policy](https://www.scaleway.com/en/docs/generative-apis/reference-content/data-privacy/) Checked 2026-10-12 | | AI Endpoints OVHcloud | France EU owned | Yes | Yes | Your data is never used to train our models. | Zero data retention. Only data required for billing is kept. | [OVHcloud, AI Endpoints](https://www.ovhcloud.com/en/public-cloud/ai-endpoints/) Checked 2026-10-12 | | AI Model Hub IONOS | Germany EU owned | Yes | Yes | Data is not used for training, fine-tuning, or otherwise improving any AI models. | Data sent via the API is processed only to provide the service. | [IONOS docs, AI Model Hub governance and compliance](https://docs.ionos.com/cloud/ai/ai-model-hub/governance-and-compliance/eu-ai-act) Checked 2026-10-05 | | AI Model Serving STACKIT (Schwarz Digits) | Germany EU owned | Yes | Yes | Vendor states customer data stays under customer control. | Set by the service configuration. | [STACKIT, AI Model Serving](https://www.stackit.de/en/product/stackit-ai-model-serving/) Checked 2026-09-06 | | PhariaAI Aleph Alpha | Germany EU owned | Yes | Yes | Runs on infrastructure you select, so the answer is yours to set. | Yours to configure. | [Aleph Alpha docs, PhariaAI deployment options](https://docs.aleph-alpha.com/phariaai-install-config-guide/latest/installation/before-you-start/deployment-options.html) Checked 2026-10-12 | | Open weight models on your own server No vendor | Wherever you put it | You decide | You decide | Nobody trains on your data, because nobody else sees it. | Whatever you configure. | No vendor documentation applies Checked 2026-09-06 | Conditional means the vendor offers EU processing and also documents circumstances in which data moves outside it. Read the exceptions below before treating conditional as a yes. ## What each vendor writes down and buyers miss The residency claim on a pricing page is usually true. The detail that decides whether it holds for your firm sits three clicks deeper. Microsoft · Microsoft Copilot (Microsoft 365) Conditional An EU Data Boundary service for tenants with an EU or EFTA sign-up location. Microsoft states that EU traffic stays inside the boundary. The catch Anthropic models offered inside Copilot sit outside the EU Data Boundary. Tenants that bought Multi-Geo are out of scope even with an EU address. LLM calls route to the nearest data centre and can go to other regions when capacity is tight. Microsoft · Azure (EU Data Boundary services) Conditional Regional Azure services you deploy into an EU Data Boundary region are in scope for the boundary. Non-regional services have to be configured separately. The catch The commitment covers customer data and pseudonymised personal data, and Microsoft documents limited circumstances in which data still moves outside the boundary. Read that transfer list before you rely on it. OpenAI · OpenAI API (Platform) Yes Data residency is configured per project. Europe (EEA and Switzerland) supports regional storage and regional processing. The catch Abuse monitoring logs are generated for all API usage and kept up to 30 days by default. You need approval for Zero Data Retention or Modified Abuse Monitoring to change that. Anthropic · Claude API (first party) No Two settings govern geography. Inference geo controls where the model runs, workspace geo controls where data sits at rest. The catch Neither setting offers Europe. EU residency for Claude models exists only through AWS Bedrock or Google Gemini Enterprise Agent Platform (formerly Vertex) EU endpoints, so the answer depends on which door you come through. Google · Gemini Enterprise Agent Platform (formerly Vertex AI) Yes Regional endpoints such as europe-west4 keep processing in that region. The EU multi-region endpoint keeps processing inside EU locations. The catch The global endpoint does not support data residency. If the region is left out of the call, the guarantee goes with it. ML processing commitments now reach well beyond the US and EU, into Brazil, Canada, the UK, Australia, India, Japan, Singapore and South Korea, so check the model's own location table rather than assuming it stops at Europe. AWS · Amazon Bedrock Yes Bedrock runs in EU regions, and each region holds its own model deployment accounts operated by the Bedrock service team. The catch Cross-region inference profiles move traffic between regions inside a geography. Check the profile you are actually calling is an EU one, because the model choice does not decide this for you. Mistral AI · La Plateforme Conditional Mistral states data is hosted in the European Union by default. Depending on the feature, it can be temporarily transferred to sub-processors outside the EU, under Article 46 GDPR safeguards. The catch Inputs and outputs are kept for 30 rolling days for abuse monitoring unless Zero Data Retention is switched on, and ZDR is not available for the stateful products such as chat, agents, libraries and batch files. Scaleway (Iliad) · Generative APIs Yes Serverless models are hosted in a data centre in Paris operated by OPCORE. Personal data may be stored in the Paris region. The catch Aggregated and anonymised usage data is kept up to 6 months, and request content can be stored temporarily when traffic causes errors or security issues. OVHcloud · AI Endpoints Yes OVHcloud states the service is hosted and run in the European Union. The catch The model catalogue is open weight models. If you need a frontier proprietary model, this route does not carry one. IONOS · AI Model Hub Yes Inference endpoints and managed vector databases run in ISO 27001 certified IONOS data centres in Germany. The catch Open weight models only, and IONOS positions itself as the AI system provider while model documentation obligations stay with the model developer. STACKIT (Schwarz Digits) · AI Model Serving Yes Data is stored and processed exclusively in Europe. STACKIT belongs to the Schwarz Group and states it is independent of external investors. The catch A smaller model catalogue than the hyperscalers, so check the specific models you need are on it before you plan around this. Aleph Alpha · PhariaAI Yes Installs into an environment you choose using Helm and Kubernetes: on premises, or on a cloud provider you pick, including StackIT, AWS, Google Cloud, or Azure. The catch Aleph Alpha does not offer PhariaAI as a SaaS product themselves. This is a stack to deploy rather than an endpoint to call, so the real cost is the operations around it. No vendor · Open weight models on your own server You decide Llama, Mistral, Qwen and the other open weight families run on a European VPS or on hardware in your office. Nothing leaves the machine you rent or own. The catch There is no vendor to hold to a commitment and no support line. You own the updates, the evaluation and the uptime, and quality sits below the frontier models. ## In the vendors' own words Quoted verbatim, with the source and the date it was read. These are the sentences worth putting in front of whoever signs the contract. > Models provided by Anthropic as a subprocessor are currently excluded from the EU Data Boundary. [Microsoft Learn, Data, Privacy and Security for Microsoft Copilot](https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy) Read 2026-09-06 > These commitments are subject to limited circumstances where Customer Data, personal data, and Professional Services Data will continue to be transferred outside the EU Data Boundary. [Microsoft Learn, What is the EU Data Boundary?](https://learn.microsoft.com/en-us/privacy/eudb/eu-data-boundary-learn) Read 2026-09-06 > your customer content (as defined in your services agreement) for that project will be stored at rest in the selected region to the extent the endpoint requires data persistence [OpenAI API docs, Your data](https://developers.openai.com/api/docs/guides/your-data) Read 2026-10-12 > Workspace geo: Only "us" is currently available. [Claude Platform docs, Data residency](https://platform.claude.com/docs/en/manage-claude/data-residency) Read 2026-10-12 > Because the model providers don't have access to those accounts, they don't have access to Amazon Bedrock logs or to customer prompts and completions. [AWS docs, Data protection in Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/data-protection.html) Read 2026-10-12 > Zero data retention: We keep only the data required for billing purposes. [OVHcloud, AI Endpoints](https://www.ovhcloud.com/en/public-cloud/ai-endpoints/) Read 2026-10-12 > data is not used for training, fine-tuning, or otherwise improving any AI models [IONOS docs, AI Model Hub governance and compliance](https://docs.ionos.com/cloud/ai/ai-model-hub/governance-and-compliance/eu-ai-act) Read 2026-10-05 ## Six questions to put to any AI vendor Ask for the answers in writing and note the date you got them. Every one of these terms has moved at least once in the past year. 1. **Where does inference physically run?** Ask for the region, not the continent. 2. **Where does data sit at rest?** This is a separate setting from inference at three of the providers above. 3. **What happens when your capacity is tight?** Capacity routing is where documented residency quietly becomes conditional. 4. **Which sub-processors touch our content?** Ask for the current list and how you are told when it changes. 5. **How long are prompts and outputs retained, and can that be set to zero?** Abuse-monitoring retention is the default almost everywhere. 6. **Is our content excluded from training by contract?** A policy statement and a contractual commitment are different instruments. This page is documentation research, not legal advice. Terms change, and the answer that matters is the one in your own contract on the day you sign it. ## Who maintains this EVY is a free AI workspace for small teams, built in Tallinn by Helix Studio OÜ. Teams ask where their AI processes data before they move company knowledge anywhere, and the honest answer needed a table nobody had built. What EVY itself does with your data is on the [trust page](https://evy.so/trust/). So we built it, and we keep it current whether or not it leads anywhere near us. Where a provider's own documentation contradicts our positioning, the documentation wins and the row says so. ## Frequently asked Which AI providers keep your data in the EU? As of 12 October 2026, EU inference is available without a geography caveat from Scaleway Generative APIs (Paris), OVHcloud AI Endpoints, IONOS AI Model Hub (Germany), STACKIT AI Model Serving (Germany), Aleph Alpha PhariaAI, the OpenAI API with a project set to the Europe region, Google's Gemini Enterprise Agent Platform (the renamed Vertex AI) on a regional or EU multi-region endpoint, and Amazon Bedrock in an EU region. Microsoft Copilot and Azure are conditional, because the EU Data Boundary carries documented exceptions. The first-party Claude API has no EU option at all. Does Microsoft 365 Copilot keep data in the EU? For tenants with an EU or EFTA sign-up location, Copilot is an EU Data Boundary service and Microsoft states that EU traffic stays within the boundary. Three documented exceptions matter. Anthropic models offered inside Copilot are currently excluded from the EU Data Boundary. Tenants that purchased Multi-Geo Capabilities are out of scope even with an EU address. Microsoft also documents that Copilot calls to the model route to the closest data centres and can call into other regions where capacity is available during high utilisation periods. Does Claude have EU data residency? Not through Anthropic directly. Anthropic's data residency documentation states that the inference\_geo parameter accepts only "us" and "global", and that workspace geo, which governs where data sits at rest, is currently US only. EU residency for Claude models is available by running them through AWS Bedrock or Google's Gemini Enterprise Agent Platform (formerly Vertex AI) on EU endpoints, where the region is set by the platform rather than by Anthropic. Is an AI tool GDPR compliant just because it is hosted in the EU? No. Hosting location is one input. A data processing agreement, a current list of sub-processors, a retention policy you can point at, a statement that your content is not used for training, and a record of processing activities all sit alongside it. EU hosting with no processing agreement leaves you in the same place you started, and a US provider with a signed agreement and documented safeguards may be the more defensible choice depending on what you process. What is the CLOUD Act concern with US AI providers? The US CLOUD Act allows US authorities to compel US-headquartered providers to produce data they control, including data stored outside the United States. This is why the ownership column on this page is separate from the hosting column. It is a factor to weigh and document rather than an automatic disqualification, and how much weight it carries depends on what your firm processes and what your professional body expects of you. What should a firm ask an AI vendor before signing? Six questions cover most of it. Where does inference physically run, and can you name the region? Where does data sit at rest? What happens when your capacity is tight, does traffic move? Which sub-processors touch our content, and are any of them outside the EU? How long are prompts and outputs retained, and can that be set to zero? Is our content excluded from training by contract rather than by policy? Ask for the answers in writing, and note the date, because these terms change. How often is this comparison updated? Every source page is re-fetched weekly by an automated check that compares the residency, retention, sub-processor and training language against the version we last read. Anything that moved gets read by hand and the row is updated with a new date. Rows nobody has touched in 45 days are re-read anyway, so a vendor that quietly starts blocking us cannot leave a stale answer standing. Where a fact cannot be verified from a primary source, the row says so rather than guessing. ## How this is maintained - Every claim comes from the vendor's own documentation. Secondary write-ups are not used where a docs page exists. - Quotes are verbatim. Where an exact sentence could not be copied from the source, no quote is shown. - An automated job re-fetches every source each week and diffs the residency, retention, sub-processor and training language against the last version we read. Only what moved gets re-read by hand. - Each row carries the date it was last read against its source. Any row untouched for 45 days is re-read regardless. - Facts that could not be verified are marked unclear rather than filled in. -------------------------------------------------------------------------------- title: "EVY vs Notion AI: price, agents and team voice compared | EVY" description: "EVY vs Notion AI compared for small teams: price for 12 people, whose AI does the work, agents, planning, team voice notes and EU data. Every Notion fact links to Notion." last_updated: "2026-10-24" source: "https://evy.so/compare/evy-vs-notion/" -------------------------------------------------------------------------------- ## The short answer 30-second read Pick Notion AI if your team lives in shared docs and databases, needs Windows or mobile, or wants AI that writes and searches inside those pages. Pick EVY if you want AI that does the work and a team that talks: EVY plans the day, hands tasks to agents on your Mac, and carries voice notes between teammates. Free, on the AI plan you already have. Price for 12 people Notion Business: **€234 a month** billed yearly, €282 monthly, before agent credits.[1](#s1) EVY: **€0**, plus the Claude or ChatGPT plans your team already pays for. At a glance ## EVY and Notion AI, side by side Ten things small teams ask about. Green marks the stronger option for that row, and Notion wins some of them. EVY vs Notion AI, checked against notion.com on 12 October 2026 | Question | eEVY | NNotion AI | | --- | --- | --- | | What is it? | **AI workspace that runs the work**Plans your day, hands tasks to agents or people, keeps company knowledge in one place. | **Docs and databases with AI**Pages, wikis and databases, with Notion Agent to write, search and edit them. | | Priceper member | **€0**Free. Setup in the desktop app. Stronger | **€19.50 a month for full AI**Billed yearly, €23.50 monthly. Free and Plus get a limited AI trial.[1](#s1) | | Whose AI does the work? | **Yours**Claude Code or Codex, on the Claude or ChatGPT plan you already pay for. Stronger | **Notion's**Models are picked and billed by Notion. Custom Agents cost $10 per 1,000 credits.[2](#s2) | | Agents | **On your Mac, every step shown**Agents get the plan and company knowledge, ask when stuck and save files to your folders. Stronger | **Inside the workspace**Notion Agent edits pages and databases. Custom Agents run on a schedule or trigger, paid in credits.[2](#s2) | | Planning | **Built in**Goals and initiatives per space, and a daily To-do EVY sets with you at a morning check-in. Stronger | **Build your own**Project and task databases from templates. Flexible, and you maintain them. | | Team communication | **Voice notes to teammates**Say who it is for and talk. They play it, read the gist and answer by voice. Stronger | **Comments and @mentions**Text on pages. AI Meeting Notes record calls on Business.[1](#s1) | | Docs and databases | **Notes and knowledge**Private and Shared notes, company knowledge any agent can read. No custom databases. | **The category leader**Nested pages, relational databases, wikis and thousands of templates. Stronger | | Talking to the AI | **Hold ⌥⌘ and speak**Talk to EVY from anywhere on the Mac. Dictation and meeting notes built in. Stronger | **Type in Notion**Ask Notion Agent from any page or the sidebar. | | Platforms | **macOS only for now**One desktop app. | **Everywhere**Web, Mac, Windows, iOS and Android. Stronger | | EU data residency | **Set by your AI provider**Agents run on your own Claude or ChatGPT account, under that provider's terms. | **Enterprise only**Through Notion's account team. Mail, Calendar and beta features are excluded.[3](#s3) | Notion prices as listed for the EU on notion.com/pricing on 12 October 2026. Where each one wins ## Different jobs, some overlap ### NNotion AI is stronger for - **Wikis and databases.** Relational databases, nested pages and templates that EVY does not try to match. - **Mixed devices.** Teams on Windows, phones and the browser. - **Big-company controls.** Enterprise adds EU residency, SAML, audit logs and zero retention with AI providers.[3](#s3) - **Writing inside pages.** Drafts, summaries and database autofill where the doc already lives. ### eEVY is stronger for - **Getting work done by agents.** Hand a task to Claude Code or Codex with the plan and context loaded, and get the file back on your Mac. - **Talking to your team.** Voice notes to one person or several, with urgent ones kept apart. - **No second AI bill.** Agents run on the Claude or ChatGPT plan you already have. - **Running several things at once.** One space per business or client, with one daily plan across all of them. Team communication ## Talk to your team instead of typing Notion keeps a team in sync through comments on pages. EVY adds voice: quick notes between people, with the AI listening in only when you ask it to. 1. **Say who it is for**"To Jeff: the Olea deck moves to Thursday, can you take the budget slide?" One person or several. 2. **It lands on their rail**Urgent notes sit apart from the normal pile, so nothing important waits behind small talk. 3. **Play, read or answer**Jeff plays it at any speed, reads the gist, answers by voice, or asks EVY to turn it into a task. Same task, both tools ## One real request, start to finish A design studio has a hotel pitch on Thursday. Here is what each tool does with the same ask. The ask"Pull three hotel case studies from our past projects for the Olea pitch." N**Notion AI**Business plan 1. Open Notion Agent in the sidebar and type the request. 2. It searches pages you can access, plus connected apps through Enterprise Search. 3. It writes a new Notion page with the three case studies. 4. You turn the page into the format the pitch needs. Result: **a Notion page** e**EVY**Claude Code on your Mac 1. Say it with ⌥⌘, or it is already on today's To-do from the morning check-in. 2. The task goes to Claude Code, running on your own Claude plan. 3. Claude reads 41 past projects and your "How we work" note. Each step shows in the task. 4. The PDF lands in the Olea pitch folder. You review it, and send Jeff a voice note that it is ready. PDFResult: **Hotel case studies.pdf**, in your folder Price, worked out ## What a 12-person team pays each month Notion Free €0 AI as a limited trial only. Notion Plus €114 /mo 12 × €9.50, yearly €138 billed monthly. AI still a limited trial. Notion Business €234 /mo 12 × €19.50, yearly €282 billed monthly. Full Notion AI, Custom Agents billed on top in credits. EVY €0 12 × €0 Agents run on the Claude or ChatGPT plans you already have. If someone on the team has no AI plan yet, a personal Claude or ChatGPT plan costs about €20 a month. Notion prices from notion.com/pricing, EU, 12 October 2026.[1](#s1) Using both ## Can you keep Notion and add EVY? Yes. Notion can stay the wiki and the home of your databases. EVY plans the week, hands work to agents on your Mac and carries the voice notes your team sends each other. If your team mainly uses Notion as a task list, EVY can replace that part: its To-do and Direction views cover goals, initiatives and daily tasks without a database to maintain. How we compared ## Method and limits ### What we checked - Notion's pricing page and help centre, read in October 2026. - EVY as it ships today on macOS. - Every Notion claim links to a Notion page below. ### What we did not test - Notion Enterprise, Notion Mail and Notion Calendar. - Speed or quality of answers. Both depend on the model behind them. - Spot something out of date? Write to [support](https://evy.so/support/) and we fix it. Questions ## EVY vs Notion AI, answered ### Is EVY a Notion alternative? For planning, agent work and team communication, yes. EVY plans your day, hands tasks to Claude Code or Codex on your Mac and lets teammates send each other voice notes. For large wikis and custom databases Notion is the stronger tool, and you can run both. ### Can Notion AI use my own ChatGPT or Claude subscription? Notion's pricing page does not offer it. Notion AI is paid for through Notion plans, and Custom Agents through Notion credits at $10 per 1,000. EVY hands tasks to Claude Code or Codex on the Claude or ChatGPT plan you already have. ### How much does EVY cost compared with Notion AI? EVY is free. Notion's full AI set needs the Business plan at €19.50 per member a month billed yearly, or €23.50 billed monthly. For 12 people that is €234 or €282 a month, before any agent credits. ### Can I send voice messages to teammates in Notion? Notion works through text comments and @mentions on pages, and AI Meeting Notes record calls on the Business plan. In EVY you say who a voice note is for and talk. It lands on their rail, marked urgent or normal, and they can play it, read the gist, answer by voice or ask EVY about it. ### Does Notion keep data in the EU? Only on the Enterprise plan, through Notion's account team. Notion Mail, Notion Calendar and beta features are not covered, and zero data retention with AI providers is also Enterprise only. ### Does EVY work on Windows? Not yet. EVY is a Mac app. Notion runs on the web, Mac, Windows, iOS and Android. Sources 1. [Notion pricing](https://www.notion.com/pricing) notion.com, read 12 Oct 2026 2. [Custom Agent pricing](https://www.notion.com/help/custom-agent-pricing) Notion Help Centre, read 9 Oct 2026 3. [Data residency](https://www.notion.com/help/data-residency) Notion Help Centre, read 9 Oct 2026 More comparisons and guides [**EU AI data residency**Where each major AI tool processes and stores your data](https://evy.so/compare/eu-ai-data-residency/) [**Best AI workspace for small teams**9 tools ranked on price, agents and EU data](https://evy.so/compare/best-ai-workspace/) [**EVY pricing**Free, on the AI plan you already pay for](https://evy.so/pricing/) [**All comparisons**Every tool and model we have compared](https://evy.so/compare/) ## Try EVY next to Notion Free. Bring the Claude or ChatGPT plan you already have, and EVY sets up your first space with you. Keep your Notion workspace, nothing to migrate to start. - €0 - macOS - Claude Code · Codex [Get EVY free for Mac](https://evy.so/download/?utm_source=compare-notion)