Research

OpenAI Study: Researchers Say ChatGPT Cuts Their Work Time Nearly in Half

OpenAI analysed 580,186 messages from 1,819 researchers and found users estimated the same work would take 1.9 times longer without AI; 98% expect faster discovery in their fields.

OpenAI study finds scientists use ChatGPT to speed research
OpenAI study finds scientists use ChatGPT to speed researchStockSnap / pixabay
By Rebecca Stone6 min read

Why it matters

  • OpenAI analysed 580,186 scientific messages from 1,819 researchers in its ChatGPT for Academic Researchers summer pilot.
  • Participants estimated the same work without AI would take 1.9 times as long.
  • 98% of 250 respondents said AI would speed up discovery in their field.
  • Reading and learning accounted for 28% of scientific messages; writing and communicating 27%.
  • OpenAI is expanding free academic access to more than 200,000 researchers worldwide.

Scientists using ChatGPT estimated that completing the same work to the same standard without AI would have taken 1.9 times as long, according to a study published by OpenAI on November 17, 2026. The finding comes from OpenAI's Economic Research team, which analysed more than 580,000 scientific messages sent during the company's ChatGPT for Academic Researchers summer pilot programme.

The study matters beyond OpenAI's own marketing. It offers one of the larger datasets to date on how AI assistants embed themselves in the daily practice of working scientists, a question with direct stakes for research productivity, academic publishing and the pace of scientific discovery itself. Because the data comes from OpenAI's own user base, the results describe a self-selected group of researchers who already chose to adopt the tool.

How much time did researchers say they saved?

The headline figure is 1.9x. Participants in the survey estimated that replicating the same work to the same standard without AI would have taken nearly twice as long. The study did not test that claim experimentally; it records what 251 survey respondents reported about their own workflows.

The usage data behind that estimate is substantial. OpenAI examined 580,186 scientific messages from 1,819 scientists on the academic plan. The messages spanned the full research cycle: reviewing literature, analysing evidence, writing software and developing research ideas. Smaller shares covered teaching and mentoring, research administration, reviewing research and gathering data.

What do scientists actually use ChatGPT for?

The largest shares of scientific messages clustered around three activities, according to the study's breakdown:

  • Reading and learning: 28% of scientific messages
  • Writing and communicating: 27%
  • Analysing evidence and theories: 17%
  • Developing ideas: 14%
  • Writing research software: 5%

The remaining messages related to teaching and mentoring, research administration, reviewing research and gathering data. The distribution shows AI use concentrated in the text-heavy, interpretive stages of research rather than in data collection — a pattern consistent with the tool's strengths in language tasks.

How do researchers generate ideas with the tool?

The study broke down ideation behaviour in detail. Researchers did not use ChatGPT as a blank-slate idea machine; most brought something of their own to the exchange.

In 35% of ideation-related exchanges, scientists arrived with a specific question and asked ChatGPT to work with it. In 36%, they came with an existing idea and asked the system to refine it. Only 23% of those messages asked ChatGPT to suggest a research question from scratch.

That split suggests researchers treat the model primarily as a collaborator on problems they have already framed, not as an autonomous source of scientific direction.

Did scientists just accept the first answer?

No. OpenAI measured how often researchers followed up to clarify, check or revise responses across different research activities, and the follow-up rates ran high across the board:

  • Writing and communicating: 38%
  • Writing research software: 34%
  • Analysing evidence and theories: 34%
  • Reviewing research: 32%
  • Developing ideas: 31%

Those figures point to iterative use rather than passive delegation. The study explicitly concludes that scientists were not simply handing over work and accepting the first answer, and instead describes a pattern of researchers engaging with the system across several stages of the research process.

For a debate dominated by fears of wholesale outsourcing of scientific labour, the follow-up data offers a counterpoint: the heaviest measured behaviour is correction and refinement, with writing tasks producing the highest follow-up rate of any activity.

Do scientists believe AI will speed up discovery?

Almost uniformly, yes. Among the 250 respondents who answered the discovery-timeline question, 98% said AI would speed up discovery in their field.

The question was concrete. OpenAI asked participants to imagine a discovery that would take 10 years without AI, then estimate how long it would take with AI available. The answers:

  • 22% said such a discovery could happen within three years with AI
  • 50% placed the timeline at four to seven years
  • 19% chose eight to nine years
  • 7% said it would take just under 10 years
  • 2% expected no acceleration at all

Half the respondents, in other words, compressed a 10-year discovery into roughly half the time. The optimism varied by discipline. In engineering, 59% of respondents selected the four-to-seven-year window. In mathematics and statistics, 31% chose within three years — the most aggressive timeline of any field group. Life and health sciences showed a comparatively higher share of respondents selecting the cautious eight-to-nine-year band.

Why do they expect acceleration?

OpenAI asked the scientists who expected faster discovery to explain their reasoning. Of the 367 reasons identified across 224 written responses:

  • 52% said AI would help get research done faster
  • 23% said it could help find more promising research questions
  • 14% said it could support work on a wider range of questions

The dominant rationale is speed of execution, not the replacement of scientific judgement. The second and third most common reasons both concern question selection — where to aim the research effort in the first place.

How did OpenAI protect researchers' privacy?

The method itself is notable. To analyse usage while limiting exposure to sensitive research material, OpenAI used automated classifiers to sort messages by activity type. The research team did not read the underlying conversations or the chatbot's replies. OpenAI also stated that data from the ChatGPT for Academic Researchers programme is not used by default to train its models — a relevant assurance for scientists handling unpublished results.

The study has acknowledged limits. OpenAI noted that the number of answers varied by question, and that some respondents selected multiple scientific fields, meaning the field groups overlapped. The survey base of 251 responses sits atop a much larger behavioural dataset of 1,819 users, and the two evidence types are not interchangeable.

What comes next?

OpenAI said it is expanding ChatGPT for Academic Researchers with free access for more than 200,000 researchers worldwide. The company framed the study's findings through the voice of its users.

"Tools like ChatGPT are helping us accelerate scientific discovery. It's really a playground for a scientist like myself," said César de la Fuente, an antimicrobial researcher quoted in the announcement.

The expansion means the behavioural patterns documented in the study — heavy use in reading and writing, iterative follow-up, and near-universal expectations of accelerated discovery — will soon describe a population an order of magnitude larger than the pilot cohort. Whether the self-reported 1.9x time savings holds at that scale, and across researchers who did not volunteer for a pilot programme, is the open question the next wave of data will have to answer.

Source: itbrief.com.au

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

728 articles

Related articles

  1. Scientists Use AI for Tasks, Not Autonomous Research, OpenAI Finds
  2. OpenAI Says a Quarter of U.S. Workers Now Use ChatGPT on the Job
  3. OpenAI: Enterprise ChatGPT Messages Up 8x as AI Use Deepens
  4. OpenAI Research: ChatGPT Users Are Redrawing Job Boundaries
  5. Largest Study of ChatGPT Use Shows AI Becoming Everyday Tool

« Previous articleNext article »