Anthropic maps how scientists use Claude
An Anthropic study of 207,000 research sessions finds scientists mostly use Claude for data analysis and writing, not idea generation, and that use clusters where AI capability is highest.
Anthropic published a report analyzing 207,000 Claude and Claude Code research sessions, collected with privacy-preserving methods, to measure how AI is used across scientific fields and tasks. The lead authors are Johannes Hermle, Ben Jones and Lukas Althoff.
Relative to the number of researchers in each field, Claude is used most in formal sciences such as computer science, mathematics and statistics, and least in health and lab-based physical sciences. Data analysis makes up about one-third of sessions on Claude and nearly half on Claude Code. Writing is the other main use.
Scientists use Claude less for generating ideas and gathering data, even though respondents to an accompanying survey ranked those tasks as the most important for discovery. On Claude Science, Anthropic's specialized product, idea generation makes up about twice the share of use that it does on Claude.
The authors built a measure of how well AI can, in theory, perform each research task. Tasks in the top 20% on that measure get two to four times as much Claude use as researchers' time on them would predict, while the bottom 20%, mostly physical work, get very little. The authors conclude that scientists "appear to use AI where it works best."
A simple model in the report sketches illustrative scenarios. If current capabilities were widely adopted, research output could double in some fields. If AI could do all non-physical research work, output in computer science and statistics could rise up to 15-fold, compared with three- to fourfold in chemistry and biology, where physical work is the bottleneck. About 40% of surveyed AI-using researchers expect AI to carry out full research projects on its own by 2030.
The report measures usage of Anthropic's own products and was written by Anthropic.