Anthropic’s study of research use found Claude sessions clustered around data analysis, even as surveyed scientists placed ideation and data gathering higher among activities important to research results.
Anthropic released the findings Nov. 5 after examining 207,049 sessions tied to consumer plans from Aug. 2 to Sept. 30, 2026. Data analysis represented 33.7% of sessions on Claude, which includes Chat and Cowork, and 48.1% on Claude Code. The company estimated researchers spent 27.6% of their time on data analysis.
Ideation and hypothesis generation appeared in 4.2% of Claude sessions and 2.1% of Claude Code sessions. Data gathering accounted for 11.5% and 11.9%, respectively. Anthropic’s estimates put researchers’ time at 7.1% for ideation and hypothesis generation and 28.6% for data gathering.
In a survey of 528 scientists, ideation and hypothesis generation ranked first for importance to research results, followed by data gathering. Respondents were mainly active AI users in life and health sciences, Anthropic said.
The company pointed to practical and measurement factors behind the gap. Gathering data often involves physical work, which limits the role AI can play, while idea generation may be folded into sessions labeled as other tasks. Anthropic also suggested that some scientists prefer to develop ideas themselves or see current AI as limited in that area.
Anthropic compared task use with estimates of how researchers allocate their time. Tasks in the top fifth of its AI capability ratings appeared at roughly twice that benchmark on Claude and four times on Claude Code.
A separate sample of Claude Science sessions put ideation and hypothesis generation at 8%. Anthropic’s product announcement says Claude Science entered beta June 30, before the study period. The study also flagged possible concerns about reduced human oversight, ethical issues and effects on researchers’ careers and sense of purpose.
