Back to news
researchAnthropic2026-11-09

Anthropic published research on November 9 showing scientists mainly use Claude for data analysis and writing, with hypothesis generation comparatively rare

The study covers 207000 research sessions between August 2 and September 30, with data analysis accounting for 33.7 percent of regular research sessions and 48.1 percent of coding sessions, while data collection and research ideation take proportionally less session share than working time

On November 9, Anthropic published a research report showing that scientists mainly use Claude for data analysis and writing, while data collection and hypothesis generation see comparatively little use of AI. The research covered 207000 research sessions between August 2 and September 30, 2026, including 129000 Claude conversation and work sessions plus 78000 Claude Code sessions. The system retains and studies only category labels such as discipline, research subfield and specific task, and does not extract or store research data, hypotheses, experimental methods, preliminary findings or unpublished papers, with all results presented in aggregate.

By discipline, computer science leads at 11.9 percent, with engineering and medicine at 8.4 and 7.7 percent. By task, data analysis accounts for 33.7 percent of regular research sessions and 48.1 percent of coding sessions, against the 27.6 percent of working time researchers typically spend on data analysis. Writing and communication take 32 percent of regular research sessions, well above the 16 percent of time researchers put into such tasks. By contrast data collection accounts for 28.6 percent of researcher working time but only 11.5 percent of research sessions and 11.9 percent of coding sessions.

Research ideation and hypothesis generation are lower still at 4.2 and 2.1 percent, against the 7.1 percent of working time researchers devote to that work. Anthropic also surveyed 528 scientists, mostly from life sciences and health sciences, and the results largely match the session data.

AnthropicClaude科研应用数据分析研究报告