Skip to main content
Back to timeline
OpenAI NewsSource publication:

OpenAI tallies how 1,819 scientists used ChatGPT: self-reported work would take 1.9 times as long without AI, and 98% expect faster discovery

Research institutions:OpenAI

Synopsis

OpenAI's economic research team analyzed 580,186 scientific messages from 1,819 scientists in the ChatGPT for Academic Researchers summer pilot and surveyed 251 of them. Respondents estimated that the same work at the same quality would have taken about 1.9 times as long without AI, and 98% of the 250 who answered a discovery-timeline question expected AI to accelerate discovery in their field. These figures come from usage records and self-reports and do not establish causal effects.

Source-provided article image: Science at the Frontier: How scientists use ChatGPT
Figure 1 · OpenAI · Page 5

Interpretation

ChatGPT has entered several stages of scientists' daily research workflow, with recorded use from literature review and evidence analysis to code writing and idea formation. Previously there was little stage-by-stage accounting, drawn from real usage records, of where in research AI is actually used. The tally covers 580,186 scientific messages from 1,819 scientists on the ChatGPT for Academic Researchers plan between August 20 and September 22, 2026, averaging each scientist's share of messages per activity with equal weight per scientist.

Idea generation is not wholly led by AI; in most cases scientists bring their own question or idea into the conversation. Among messages about research ideas, 35% had the scientist bring a specific question, 36% had the scientist bring an idea for ChatGPT to refine, 23% had ChatGPT suggest a research question, and 6% could not be categorized. These shares come from averaging each scientist's shares of ideation messages, using a threshold of at least three ideation messages to count as ideation use.

Scientists keep following up to clarify, check or revise ChatGPT's answers after delegating tasks, measured separately by activity. The study defines a scientist involvement score as the share of requests followed by clarification, checking or revision in the next scientific message in the same conversation within seven days. Only initial requests with a complete observation window and an assessable outcome are included, averaged within each activity for each scientist.

Perspective

The analysis covers 1,819 scientists and their 580,186 scientific messages on the ChatGPT for Academic Researchers plan from August 20 to September 22, 2026; the survey ran September 30 to October 7 with 251 respondents and varying counts per question, and the research team saw only outputs from privacy-preserving classifiers, not the underlying messages or ChatGPT responses.

Productivity is scientists' self-reported estimate and discovery timing is expectation rather than observed outcome, and the study explicitly states these findings do not establish causal effects; scientists also reported constraints on staffing, funding and experimental capacity, and whether expectations translate into faster discovery will take more research.