Anthropic has released a new research report examining how scientists use its Claude AI models in their daily research activities. The findings reveal that data analysis and scientific writing are the two most common applications, while data collection, research planning, and hypothesis generation account for a much smaller share of AI-assisted work.
The study provides an interesting look at how artificial intelligence is becoming part of scientific workflows, particularly in fields that rely heavily on programming, quantitative analysis, and large amounts of research data.
Anthropic’s latest developments in AI also include next-generation Claude models, highlighting the company’s continued efforts to improve AI reasoning and performance.
Anthropic Analyzes More Than 207,000 Scientific AI Sessions
According to Anthropic’s report, the research examined 207,000 scientific work sessions recorded between August 2 and September 30, 2026.
The dataset included three types of interactions:
- 129,000 Claude conversations and Cowork sessions.
- 78,000 Claude Code sessions.
To protect researchers’ privacy, Anthropic focused on classification information, including academic disciplines, specialized research areas, and task categories. The company did not extract or retain specific research materials, experimental methods, scientific hypotheses, preliminary findings, or unpublished papers.
All findings were presented in aggregated form rather than exposing individual research conversations.
This approach allowed Anthropic to identify broader patterns in scientific AI adoption without collecting sensitive details from ongoing research projects.
Computer Science and Mathematics Researchers Use Claude More Frequently
The report found significant differences in Claude adoption across academic disciplines.
Computer science researchers were the largest group in the dataset, accounting for 11.9% of scientific Claude sessions. Engineering represented 8.4%, while medicine contributed 7.7%.
Computer science, mathematics, and statistics also showed relatively strong adoption when Anthropic compared AI usage with the size of each discipline’s overall research workforce.
Claude Code displayed an even stronger concentration in quantitative and computational research.
Although computer science researchers account for only 4.3% of the global research workforce, they generated 32.7% of the scientific sessions involving Claude Code.
Artificial intelligence and machine learning researchers also demonstrated particularly high adoption rates. They represent approximately 1.2% of researchers worldwide but contributed 5% of Claude research sessions and 15.2% of Claude Code sessions.
By comparison, researchers working in arts, humanities, and some social science disciplines used Claude less frequently.
Anthropic suggests that these differences are not simply determined by the number of researchers in each field. The suitability of individual research tasks for AI assistance also plays an important role.
The growing competition between AI developers is also driving improvements in AI models and automated workflows, making advanced AI tools increasingly useful for professional tasks.
Data Analysis Is the Most Common Scientific Application of Claude
One of the report’s most notable findings concerns the types of work scientists delegate to AI.
Data analysis ranked first among Claude-assisted research activities. It accounted for 33.7% of general Claude scientific sessions and an even higher 48.1% of Claude Code sessions.
For comparison, researchers reported spending approximately 27.6% of their overall working time on data analysis.
Writing and communication were another major application. These activities represented 32% of Claude scientific sessions, roughly twice the 16% of working time researchers typically devote to these tasks.
The figures suggest that scientists are particularly interested in using AI to process information, interpret research data, prepare documents, and communicate findings.
Claude Code’s higher share of data-analysis activity also highlights its relevance to research workflows involving programming, statistical processing, and computational tools.
Why Scientists Use AI Less for Data Collection and Hypothesis Generation
Interestingly, tasks that require gathering original research information or developing new scientific ideas were less frequently handled with Claude.
Anthropic’s report identified the following differences between researchers’ working time and their AI usage:
| Research activity | Researchers’ working time | Claude sessions | Claude Code sessions |
|---|---|---|---|
| Data analysis | 27.6% | 33.7% | 48.1% |
| Writing and communication | 16% | 32% | Not specified |
| Data collection | 28.6% | 11.5% | 11.9% |
| Research ideas and hypothesis generation | 7.1% | 4.2% | 2.1% |
Data collection represents a substantial part of scientific work, accounting for 28.6% of researchers’ time. However, it appeared in only 11.5% of Claude sessions and 11.9% of Claude Code sessions.
Research planning and hypothesis generation showed an even smaller presence. These activities accounted for just 4.2% of Claude sessions and 2.1% of Claude Code sessions, compared with 7.1% of researchers’ overall working time.
This pattern suggests that scientists currently rely more on AI for supporting and accelerating established research activities than for independently generating new research directions.
Survey of 528 Scientists Supports the Findings
Anthropic also conducted a separate survey involving 528 scientists, most of whom worked in life sciences and health sciences.
The participants were active AI users, making the survey particularly useful for understanding how researchers incorporate these tools into their existing workflows.
The survey results largely matched the findings from the conversation data. Respondents identified data analysis and writing as their most frequent AI-assisted activities.
Meanwhile, data collection, research planning, and hypothesis generation remained less common applications.
The consistency between the survey responses and the analysis of Claude sessions strengthens the report’s overall picture of how scientists currently approach generative AI.
What Anthropic’s Research Says About AI in Science
Anthropic’s findings highlight an important distinction between using AI as a research assistant and relying on it to conduct scientific discovery independently.
For now, Claude appears to provide the greatest value in tasks involving information processing, quantitative analysis, programming, and written communication. These are areas where AI can help researchers reduce repetitive work and spend more time interpreting results.
However, the relatively low adoption of AI for data collection and hypothesis generation suggests that human researchers continue to play a central role in defining research questions, gathering evidence, and developing scientific ideas.
As AI models and research tools become more capable, their role in scientific workflows may expand beyond analysis and writing. Whether they can become more effective partners in experimental design and original discovery will be an important development to watch.