OpenAI Study Finds Scientists Rarely Request End-to-End AI Research

Unite.AI
Read full story· 4 min
OpenAI studied over 580,000 scientific messages from 1,819 scientists to analyze AI use in research. Most AI requests focused on reading, writing, and analysis tasks, while full end-to-end autonomous research was rare, at 0.4% of messages. The study also found scientists often engaged in iterative collaboration with AI, especially in writing and analysis.

What happened

On November 16, 2026, OpenAI published a study analyzing over 580,000 scientific messages from 1,819 academic scientists, focusing on how they use AI tools like ChatGPT and Codex in their research. The study covered activity from August 20 to September 22, 2026, and included a survey conducted at the end of September and early October. It found that most AI requests were for reading, writing, and analysis tasks, while only a small fraction involved asking AI to handle entire research projects autonomously. Alongside the study, OpenAI announced an expansion of its free ChatGPT access program for academic researchers, opening 50,000 new grants.

Key details

  • End-to-end autonomous research requests made up just 0.4% of messages.
  • Reading and knowledge acquisition accounted for 27.7% of messages; writing and communication, 26.9%.
  • 89.7% of scientists asked AI to explain scientific concepts at least three times.
  • Scientists estimated AI saved them about 47% of their time on research tasks.
  • OpenAI's new grant round will provide free ChatGPT Plus access to up to 150,000 researchers.

Why it matters

Academic researchers are mainly using AI for specific tasks like reading, writing, and analysis, not for fully automated research projects. This means scientists still need to review, clarify, and correct AI outputs, which limits the extent to which research can be automated. The expanded access to ChatGPT Plus will allow more researchers to use these tools, but the main benefit is likely to be increased efficiency in individual tasks rather than complete automation of research workflows.


More in AI Research