A new OpenAI study highlights growing optimism among scientists about the role of AI in scientific research, with 98% of surveyed researchers expecting ChatGPT to accelerate scientific discovery. The findings suggest that AI tools are increasingly viewed not merely as productivity assistants, but as technologies that could help researchers explore ideas, analyze information, solve complex problems, and speed up the research process.
The survey of more than 1,800 scientists points to a broader shift in how the scientific community views AI. As research becomes increasingly data-intensive and interdisciplinary, AI assistants could help scientists reduce time spent on routine work and devote more resources to experimentation, analysis, and breakthrough discoveries.
How Scientists Are Using ChatGPT in Research
The study analyzed 580,186 scientific messages from 1,819 scientists participating in OpenAI’s ChatGPT for Academic Researchers pilot. It also surveyed 251 participants to understand how researchers use AI and what they believe it could mean for scientific progress.
Researchers used ChatGPT throughout the research process, including reading scientific literature, writing, analyzing evidence, coding, and developing research ideas.
Key findings from the study
| Finding | Result |
|---|---|
| Scientists expecting AI to accelerate discovery | 98% |
| Scientists using ChatGPT to develop research ideas | 68% |
| Research work estimated to take without AI | 1.9× longer |
| Scientific messages analyzed | 580,186 |
| Scientists included in usage analysis | 1,819 |
| Scientists surveyed | 251 |
| Researchers expecting discovery within 3 years instead of 10 | 22% |
The findings indicate that researchers see AI as useful beyond simply writing or summarizing. It is increasingly helping scientists think through hypotheses, identify potential research directions, and evaluate technical problems.
AI Is Becoming a Research Partner for Idea Generation
One of the more notable findings was the use of ChatGPT in developing scientific ideas.
Among scientists’ ideation-related interactions:
- 35% involved researchers bringing a specific question to AI.
- 36% involved researchers sharing an idea and asking AI to refine it.
- 23% involved scientists asking AI to suggest a research question.
- 6% were categorized as other or unclear.
This suggests that AI is being used in both researcher-led and AI-assisted discovery processes. Scientists are not necessarily handing over the entire research process; many are using AI to refine their own thinking, challenge assumptions, and explore possibilities they may not have considered.
Where ChatGPT Fits Into the Scientific Workflow
OpenAI’s analysis found that ChatGPT was used across several major research activities, with reading and writing accounting for a substantial portion of scientists’ interactions.
Scientific research activities
0%8%16%24%32%Reading & learningWriting & communicationEvidence analysis
Reported distribution of major research activities in the study. Other activities include coding, idea development, and additional scientific work.
AI can support researchers in several ways:
- Literature review: Summarizing papers, comparing research findings, and helping navigate large volumes of scientific literature.
- Data analysis: Assisting with interpreting evidence, identifying patterns, and exploring datasets.
- Programming: Writing, debugging, and improving research code.
- Scientific writing: Helping prepare papers, explanations, reports, and communications.
- Research ideation: Exploring hypotheses, generating questions, and refining research directions.
These tasks can be particularly valuable in fields where researchers must work across large datasets, technical documentation, and rapidly expanding scientific literature.
Scientists Estimate Research Could Take Nearly Half the Time
Researchers in the study estimated that completing the same work to the same quality without AI would have required approximately 1.9 times as much time.
This is a self-reported estimate rather than a controlled experiment measuring productivity gains. Nevertheless, it indicates that scientists perceive AI as reducing the time needed to complete a range of research tasks.
The potential impact goes beyond saving time on individual assignments. If scientists spend fewer hours on repetitive or time-consuming activities, they may have more time to explore hypotheses, design experiments, examine unexpected results, or work on complex problems.
The productivity benefit could be especially significant for researchers operating with limited resources, smaller teams, or restricted access to specialized expertise.
Scientists Expect AI to Shorten Discovery Timelines
The study asked scientists to imagine a scientific discovery that would take 10 years without AI and estimate how much sooner it might happen with AI assistance.
0%15%30%45%60%3 years or less4–7 years8–9 yearsLess than 10 yearsNo acceleration
According to the findings, around 22% of respondents believe a discovery that would normally take a decade could be achieved within three years using AI. About half expect it to take between four and seven years.
These figures reflect scientists’ expectations, not verified predictions of future breakthroughs. AI’s actual contribution will depend on the scientific field, the complexity of the problem, the quality of available data, and whether researchers can validate and experimentally test AI-generated insights.
Human Oversight Remains Essential
Despite the high level of optimism, the study does not suggest that scientists are blindly accepting AI-generated answers.
Researchers frequently followed up on AI responses to clarify, check, or revise the output. OpenAI found that scientists followed up after approximately one-third of requests on average, with greater involvement for tasks requiring more scientific expertise.
This is particularly important because AI models can produce incorrect explanations, misinterpret evidence, generate flawed code, or confidently present unsupported conclusions.
In scientific research, an incorrect result can have consequences far beyond a failed chatbot response. It can affect experimental design, research funding, published findings, and subsequent work by other scientists.
AI therefore appears to be functioning as an assistant in the research process, with scientists continuing to exercise judgment over how outputs are interpreted and used.
Research Bottlenecks Could Limit AI’s Impact
Even if AI reduces the time required for analysis, writing, and research planning, scientific discovery still depends on resources that software cannot independently provide.
Researchers identified several constraints:
- Limited staffing: More than half cited insufficient personnel or time as a major research bottleneck.
- Funding limitations: Research funding remains a major obstacle to conducting experiments and pursuing ambitious projects.
- Experimental capacity: Many discoveries require laboratories, equipment, physical testing, clinical trials, or field research.
- Validation: AI-generated hypotheses and conclusions still need to be tested against real-world evidence.
AI may help researchers use existing resources more efficiently, but it cannot automatically eliminate the financial, logistical, and experimental barriers that slow scientific progress.
OpenAI Expands AI Access for Academic Researchers
Alongside publishing the study, OpenAI is expanding its ChatGPT for Academic Researchers initiative, which is intended to provide free access to AI tools for a larger scientific community.
The expansion aims to help researchers use advanced AI models for scientific reasoning, data analysis, coding, literature review, and other research tasks.
The initiative is part of OpenAI’s broader effort to position AI as infrastructure for scientific research, potentially helping universities, laboratories, and research teams gain access to tools that would otherwise require substantial computing or software budgets.
The Bigger Picture
The 98% figure is striking, but its importance lies in what it reveals about scientists’ expectations of AI. Researchers increasingly see AI as a tool that could change the speed and scope of scientific work, from identifying research questions to analyzing results.
However, optimism should not be confused with proof. The study measures usage patterns and researchers’ expectations, rather than demonstrating that AI has already accelerated scientific breakthroughs by a specific amount.
The next major test will be whether AI-assisted research produces reproducible discoveries, better experiments, stronger evidence, and measurable improvements in scientific output. If that happens, AI could become one of the most influential tools in modern research.
Looking Ahead
AI is likely to become more deeply integrated into scientific workflows as models improve in reasoning, coding, data analysis, and the ability to work with specialized research tools. The biggest gains may come when AI helps researchers connect findings across disciplines, explore overlooked hypotheses, and reduce the time between identifying a problem and testing a potential solution.
For OpenAI and the wider AI industry, scientific research is becoming an important proving ground. If AI tools can consistently help researchers produce validated results faster, the technology’s impact could extend far beyond productivity, potentially reshaping how discoveries are made across medicine, mathematics, engineering, biology, and other fields.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.



