Anthropic finds scientists use Claude mostly for data analysis, not ideation, across 207k sessions
A company research post covering August 2–September 30 consumer sessions says data analysis is 33.7% of Claude and 48.1% of Claude Code research use, while ideation is only about 4% — below how researchers spend their own time.
Why it matters
Anthropic's study is one of the largest public looks at how researchers actually use frontier models day to day. It finds Claude is used most where models already look strong — data analysis and writing — and least for ideation and data gathering, the steps scientists say matter most for important results. That gap frames debates about whether AI will accelerate discovery or mostly speed up paper polishing. Watch for field-level follow-ups, lab policies on Claude Code, and whether ideation use rises as agents improve.
Key facts
- 207,049 consumer Claude/Claude Code research sessions, 2 Aug–30 Sep 2026
- Data analysis: 33.7% of Claude, 48.1% of Claude Code research sessions
- Writing/communication: 32.0% of Claude research sessions
- Ideation/hypothesis generation: ~4.2% Claude, ~2.1% Claude Code
- Scientists rank ideation first for important results, but use AI less there
- Company research post, Anthropic, 5 November 2026