Article · AI translation
Thanks to IT Home reader 咩咩洋 for the tip!
IT Home reported on November 9 that Anthropic recently published a research report showing that scientists mainly use Claude for data analysis and writing, while using AI less often to collect data or generate research hypotheses.
IT Home learned that Anthropic’s study used 207,000 scientific research sessions from August 2 to September 30, 2026, including 129,000 Claude conversations and Cowork sessions and 78,000 Claude Code sessions. The system retained and studied only classification labels such as the discipline, research subfield and specific task associated with each session; it did not extract or store specific content such as research data, hypotheses, experimental methods, preliminary findings or unpublished papers. All results were presented in aggregate to protect privacy.
By discipline, researchers in fields such as computer science, mathematics and statistics used Claude more frequently, while use was relatively lower in healthcare and in physical sciences that require laboratory work.
Computer science (CS, IT Home notes) accounted for the largest share of Claude research sessions, at 11.9%; engineering and medicine accounted for 8.4% and 7.7%, respectively. Even after accounting for the number of researchers in each discipline, the share of Claude use in computer science, mathematics and statistics remained markedly higher than those fields’ share of researchers.
Claude Code use was even more concentrated in quantitative research fields. CS researchers accounted for just 4.3% of researchers worldwide but contributed 32.7% of Claude Code research sessions. AI and ML also stood out: researchers in that field accounted for just 1.2% of the global total, yet contributed 5% of Claude research sessions and 15.2% of Claude Code research sessions.
By contrast, Claude use was relatively low in the arts, humanities and some social sciences. Anthropic believes that differences in Claude use across disciplines are closely related not only to the size of their researcher populations but also to how well their research tasks suit AI.
For specific research tasks, data analysis and writing were the work scientists most often used Claude to help complete.
The report showed that data analysis accounted for 33.7% of Claude research sessions and 48.1% of Claude Code research sessions. By comparison, researchers typically spend 27.6% of their working time on data analysis. Writing and communication tasks accounted for 32% of Claude research sessions, also markedly higher than the share of time researchers devote to such tasks (16%).
By contrast, data collection accounted for 28.6% of researchers’ working time but only 11.5% of Claude research sessions and 11.9% of Claude Code sessions. Research ideation and hypothesis generation accounted for even smaller shares, at 4.2% and 2.1%, respectively, while researchers spent 7.1% of their time on this type of work.
Anthropic also surveyed 528 scientists. Respondents came mainly from the life sciences and health sciences, and all were researchers who actively used AI. The survey results broadly matched the actual session data: respondents most often used AI for data analysis and writing, while use for data collection, research ideation and hypothesis generation was relatively low.