Anthropic's Economics team published a technical report on November 5, 2026 that measures how researchers use Claude. The report, "AI for scientific discovery: How researchers use Claude," analyzes 207,049 consumer-plan sessions from August 2 to September 30, 2026: 128,914 on Claude (Chat and Cowork) and 78,135 on Claude Code. It also draws on a survey of 528 scientists.
Sessions were labeled by field, subfield and task using the SciNet taxonomy, which covers 7,262 tasks across 34 fields and 318 subfields. Anthropic says this was done with privacy-preserving infrastructure that kept only the labels, and results are reported in aggregate. The survey respondents were recruited from applicants to Anthropic's AI for Science grant programs in the US and Canada. The authors note these respondents are likely heavier AI users than typical scientists.
Usage is skewed toward computational work. Computer science accounted for 11.9% of Claude sessions and 32.7% of Claude Code sessions, though only 4.3% of the world's researchers work in the field. Data analysis made up 33.7% of Claude sessions and 48.1% of Claude Code sessions, compared with an estimated 27.6% of researcher time. Ideation (4.2%) and data gathering (11.5%) were well below their share of researcher time. Survey respondents, meanwhile, ranked those two activities as the most important for producing important results. The report finds that tasks in the top fifth by estimated AI capability got about twice the expected use on Claude and about four times on Claude Code.
The authors also model how much AI could speed up research. With current capabilities, they estimate AI could take on 27% to 52% of researcher time, implying a speedup of 1.35x to 2.02x. If every non-physical research step were automated, fields with few physical tasks could reach about 15x to 16x. Biology, chemistry and similar fields would reach only 3x to 4x, because physical steps remain about 40% of their time. The authors describe these as illustrative scenarios, not forecasts.
The survey also asked about expectations. About 40% of respondents expect AI to carry out full research projects by 2030, about 64% by 2035, and 16% say it will never happen. For teams building scientific AI tools, the report points to the gaps: today's use clusters around coding and analysis, while hypothesis generation, data gathering and physical experimentation remain largely untouched.