Anthropic · 2026-11-05 · notable
How researchers use Claude — Anthropic maps AI across 207,049 science sessions
In short
Anthropic's Economics team studied 207,049 science sessions on Claude and Claude Code and surveyed 528 scientists. Data analysis led use, and the team estimates current AI could take on 27–52% of researcher time.
Anthropic · 2026-11-05 · 3 sources checked
Anthropic measures what scientists actually ask Claude to do, and how much research time AI could take on.
What is it?
'AI for scientific discovery: How researchers use Claude' is a November 5, 2026 report from Anthropic's Economics team. It looks at 207,049 research sessions from August 2 to September 30, 2026, split into 128,914 on Claude and 78,135 on Claude Code. A survey of 528 scientists adds their own view of where AI fits.
How does it work?
The team sorted each session by field and by type of research task, then compared that mix with how researchers worldwide spend their time. Data analysis was 33.7% of Claude sessions and 48.1% of Claude Code sessions, against 27.6% of researcher time. Writing was 32.0% of Claude sessions against 16.0% of researcher time. Tasks in the top 20% for capability got about 2.1x the expected use on Claude and about 3.9x on Claude Code.
Why does it matter?
The report gives a measured, field-by-field picture instead of a single headline. At today's capability it estimates AI could take on 27–52% of researcher time, a 1.35x–2.02x speed-up. In stronger automation scenarios, some non-physical fields reach about 15–16x, while bench sciences like biology and chemistry stay near 3–4x because lab work is the bottleneck. About 40% of surveyed researchers expect AI to run full research projects by 2030.
Who is it for?
research leaders, science funders and AI-for-science teams
Try it
https://www-cdn.anthropic.com/6e7c6446c0d8405dba73b91dabd36aac069dea2e/AI-for-Scientific-Discovery.pdf