Scientists use Claude most for data and writing, least for new ideas

Scientists use Claude mostly to analyze data and write, and much less to gather data or dream up new hypotheses. That is the main finding of a new Anthropic study of 207,049 Claude sessions on consumer plans, collected from August 2 to September 30, 2026. Anthropic makes Claude, so the study is the company looking at its own product.

What the study found

Use varies by field. Relative to the number of researchers, computer science, mathematics and statistics use Claude more heavily. Health and lab-based physical sciences use it less. A survey of scientists, run alongside the study, matched what the usage data showed.

The researchers also built a measure of how well AI should be able to handle a given task in theory. Claude use was higher in tasks and fields that scored higher. Their summary: scientists appear to use AI where it works best.

How privacy was handled

Anthropic says it did not read or store the details of anyone's research. A language model labeled each session with a field and a task, such as chemistry or data cleaning. Those labels were the only information kept. Results are reported only in aggregate, and only for groups with a minimum number of distinct accounts.

• The labels come from a list called SciNet, with 7,262 tasks across 34 fields.
• No datasets, hypotheses, methods or unpublished papers were extracted.
• The main sample covers Chat, Cowork and other Claude tools.

How much faster could science get

The report lays out scenarios, not predictions. With today's AI, research output in some fields could as much as double, mostly because AI speeds up individual tasks. In a hypothetical case where AI fully automates all non-physical work, highly cognitive fields such as computer science or statistics could see up to a 15-fold increase.

There is a catch the authors point out. The fields with the most to gain include those that build AI. Faster AI research could produce better AI, which could speed research further.

Risks and open questions

The authors name concerns. Less human oversight could lead to dangerous or unethical research. AI could also change researchers' careers, making the work more or less rewarding, and alter the sense of purpose in pursuing knowledge. The scenarios rely on assumptions, including survey answers from scientists about future capabilities.

What it means for everyday work

Even in a field built on discovery, AI is used most for analysis and writing. Those are tasks many businesses also have, such as cleaning up numbers and drafting documents. They are a reasonable place to look first for time savings.