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.
- Medium confidence
- 3 primary sources
- 2023 to 2026
- Last checked
Every study on one scale
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
| 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 ↗ |
| 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 ↗ |
| 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 ↗ |
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.
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-083–5.4%How much time does AI save knowledge workers?
- F-1014–15%How much more do customer support agents get done with AI?
- F-1121–26%How much faster do developers code with AI assistants?
- F-1227–46%Does AI help less-experienced workers more than experienced ones?
- F-1314–25%When does AI make work slower or worse?
The whole chapter · Chapter poster · All of The AI at Work Fact Book