Scientists Say ChatGPT Speeds Up Their Work 1.9x: OpenAI Report
OpenAI looked at how scientists use ChatGPT, analyzing more than 580,000 messages from 1,819 researchers who took part in a summer pilot. Asked to estimate, respondents said the same work of the same quality would take 1.9 times longer without AI. 98% of those who answered expect discoveries in their field to speed up. But these are the participants' own self-assessments, so it is too early to call the effect proven.
AI-processed from OpenAI Blog; edited by Hamidun News
When people say AI will "speed up science," they usually cite bold forecasts. OpenAI decided to look at real data instead: what scientists actually do in ChatGPT every day. The company's economic research team analyzed more than 580,000 scientific messages from 1,819 participants in the summer pilot of ChatGPT for Academic Researchers, and also surveyed 251 people. The result is a rare picture of how professional researchers fold a chatbot into their workflow.
What scientists do with ChatGPT
Participants used the model at every stage of their work: literature review, data analysis, writing code, formulating ideas, and drafting texts about results. To avoid reading personal conversations, OpenAI used automated classifiers. They sorted messages by type of activity, and researchers only saw categories, not the questions and answers themselves. By default, participants' data is not used to train models.
For each scientist, the team calculated the share of messages by task type, then averaged those shares so that an active user did not outweigh the others. The most interesting part concerns ideas: 68% of scientists turned to ChatGPT to develop research concepts. In these conversations, 35% of the time the scientist arrived with a specific ready-made question, 36% of the time they brought an idea and asked for it to be refined, and 23% of the time they asked the model to propose a research question. The remaining 6% of messages could not be assigned to any group.
In other words, in most cases the person remains the author of the direction, while AI works as a conversation partner and editor. But almost a quarter of requests amount to asking where to look in the first place. For product teams, this is a signal: scientific users want more than "find me a paper." They want help framing the problem.
Another metric concerns trust. OpenAI measured how often scientists, after receiving an answer, ask follow-up questions, ask for verification, or ask for the result to be rewritten. The company concludes that researchers stay engaged in the process even after delegating part of the work to the model. This matches developer practice: AI output is not accepted at once but after several iterations.
What scientists expect from AI
Respondents were asked to imagine a discovery that would take 10 years without AI, and to estimate how long it would take with AI. 250 people answered this question, and 98% of them expect discoveries in their field to speed up. Separately, participants estimated that the same work of the same quality would take 1.9 times longer without AI.
Why do they expect acceleration? Across 224 written responses, the team identified 367 reasons. In first place, at 52%, was the simple "doing research faster." Another 23% cited finding more promising questions, 14% cited the ability to cover a wider range of topics, and 11% fell into other categories. For some scientists, then, the value lies not only in saving time but in expanding what they can even try.
The report includes a quote from César de la Fuente, a specialist in antimicrobial research, who calls ChatGPT "a true sandbox for a scientist."
How to read these numbers
These figures should be interpreted carefully. First, the 1.9x estimate is the respondents' subjective guess, not a measurement of time in a controlled experiment. Second, the study was conducted by OpenAI itself, on a pilot of its own product, where participants chose to use it and are probably more favorable toward AI than average. Third, the survey covered 251 people out of thousands of pilot users. The company itself acknowledges that finding out whether these expectations turn into real acceleration of discovery will require further research. Progress also depends on funding, staff, and experimental capacity: a model will not replace a lab.
The practical takeaway for those building AI products for professionals is this. People use AI across the whole chain of work, not in a single function. They expect collaborative iteration, not a one-off answer. And privacy matters to them: OpenAI's approach of anonymized classification may serve as a model for analyzing the use of enterprise and scientific products.
In parallel, the company is expanding ChatGPT for Academic Researchers and promises free access to more than 200,000 researchers worldwide. This is both a step toward real benefit for science and an effort to cement ChatGPT as a standard working tool in universities.
- Nov 21, 2026FDE job postings grew 4.6x in a year: what a forward deployed engineer actually does at the client's site
- Nov 21, 2026A swimming team captain handed ChatGPT three scheduled tasks and cut her routine work by a multiple
- Nov 21, 2026OpenAI GPT-6 Cyber Found About 35 Flaws in iPhone, Mac and Other Apple Devices
- Nov 21, 202622 Claude Agents Spent Six Weeks on Black Hole Problems — and Solved Two
Key Questions Answered
What scientists do with ChatGPT?
Participants used the model at every stage of their work: literature review, data analysis, writing code, formulating ideas, and drafting texts about results. To avoid reading personal conversations, OpenAI used automated classifiers. They sorted messages by type of activity, and researchers only saw categories, not the questions and answers themselves. By default, participants' data is not used to train models.
What scientists expect from AI?
Respondents were asked to imagine a discovery that would take 10 years without AI, and to estimate how long it would take with AI. 250 people answered this question, and 98% of them expect discoveries in their field to speed up. Separately, participants estimated that the same work of the same quality would take 1.9 times longer without AI. Why do they expect acceleration? Across 224 written responses, the team identified 367 reasons.
How to read these numbers?
These figures should be interpreted carefully. First, the 1.9x estimate is the respondents' subjective guess, not a measurement of time in a controlled experiment. Second, the study was conducted by OpenAI itself, on a pilot of its own product, where participants chose to use it and are probably more favorable toward AI than average. Third, the survey covered 251 people out of thousands of pilot users.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.