Anthropic Analyzes 207,000 Research Sessions: Scientists Mostly Use Claude for Data Analysis and Writing, and Less for the Idea Generation and Data Gathering They Rate Most Important

Sample and method

Anthropic's Economics team published the report "AI for scientific discovery: How researchers use Claude" on November 5, with Johannes Hermle, Ben Jones and Lukas Althoff as lead authors. The sample is 207,049 research sessions on consumer plans from August 2 to September 30, 2026: 128,914 from Chat and Cowork and 78,135 from Claude Code. A language-model classifier first decides whether a session involves scientific work, then labels it using the SciNet taxonomy (34 fields, 318 subfields, 7,262 tasks); the research team keeps and analyzes only those labels. A separate survey covered 528 scientists, mostly active AI adopters in the life and health sciences.

Where it is used, and what researchers value

Data analysis accounts for 33.7% of Claude sessions and 48.1% of Claude Code sessions, against about 27.6% of researcher time; writing and communication make up 32.0% of Claude sessions, twice their 16.0% share of time. Data gathering takes 28.6% of researcher time but only 11.5% of Claude sessions and 11.9% of Claude Code sessions. In the survey, scientists ranked idea generation and data gathering as the most important steps for discovery, the two areas where use is lowest. Relative to the number of researchers, formal sciences such as computer science, mathematics and statistics use Claude more, while health and lab-based sciences use it less. The report also notes that on Claude Science, the research-focused product, idea generation takes about twice the share it does on regular Claude.

The speedup figures are scenarios

The report scores each task on how well AI could perform it in theory; the top 20% of tasks get two to four times the use that researcher time would predict. From there it models scenarios: with wide adoption of current capabilities, AI could take on 27% to 52% of the researcher time needed for a new study (median 38%), implying speedups from 1.35x in the arts to 1.96x in computer science and 2.02x in statistics. If AI could also do all non-physical work, output in computer science and statistics could reach about 15x, while fields limited by lab steps, such as chemistry and biology, would reach about 3x to 4x.

Keep in mind that the report covers only consumer users of Anthropic's own products, the survey leans toward people who already use AI, and the speedup numbers come from hypothetical scenarios in a simple model, not from measuring paper output or research speed.