Surveyed scientists estimate work would take 1.9 times as long without AI

OpenAI says its team analysed 580,186 messages from 1,819 scientists in a free pilot and surveyed 251. The company says the productivity figure is self-reported and the findings do not show cause and effect.

Share
Surveyed scientists estimate work would take 1.9 times as long without AI
Illustration: AI-generated

Scientists who used ChatGPT in a pilot programme estimate that the same work, at the same quality, would have taken 1.9 times as long without AI. The figure comes from a study by OpenAI's Economic Research team, and it rests on the scientists' own estimates. OpenAI says the findings do not establish causal effects.

The team studied 580,186 messages that 1,819 scientists sent to ChatGPT in a pilot of ChatGPT for Academic Researchers, from 20 August to 22 September 2026. It also surveyed 251 of them from 30 September to 7 October 2026.

Unite.AI, which summarised the full paper, reports that OpenAI cautions that participants applied for expanded AI access and chose to answer an AI survey, so the results may not apply to all scientists.

The 1.9 figure was built from eight research activities. For each, scientists estimated how long 1 hour of work with AI would have taken without it. When a scientist used no AI for an activity, the time was counted as unchanged.

To protect privacy, OpenAI used automated classifiers, which are software tools that sort messages by type of research activity, such as reviewing literature or writing code. The research team saw only those categories, not the messages or ChatGPT's replies.

The survey also asked scientists to imagine a discovery that would take 10 years without AI. Of those who answered, 98% expected AI to speed up discovery in their field. That is what scientists expect, not a measured result.

What did scientists mostly ask ChatGPT to do? Unite.AI's summary says reading and learning was the biggest use, at about 28% of the average scientist's scientific messages, with writing and communicating next at about 27%. Requests for AI to carry out a whole research project made up only 0.4%. It adds that, on average, 31.8% of a scientist's first requests were followed by a message asking ChatGPT to clarify, check or revise.

Unite.AI's summary also reports that the self-reported time saving ranged from 1.36 times for administration to 2.85 times for research software and code. It says checking or correcting outputs was the difficulty surveyed scientists chose most often, at 45.4%.

OpenAI says it will take more research to know whether these expectations lead to faster discovery. It adds that progress also depends on funding, staff and experimental capacity. Unite.AI reports that limited time or staffing was chosen by 51.8% of respondents as a constraint, and limited funding by 43.8%.

Alongside the study, OpenAI is expanding the programme to give up to 200,000 researchers worldwide free access to ChatGPT. The company says 20,000 researchers in 33 countries have joined since the pilot began in July. Applications for the next round open on 23 November 2026.

Sources