Independent newsroom

Friday, November 6, 2026

The File

Every claim traces to a primary source. Uncertainty is labeled. Corrections are public.

AI · Friday, November 6, 2026

Anthropic counts how scientists use Claude

The File desk · Nov. 6, 2026, 6:32 a.m. ET

Partially confirmed1 primary sourcesAdded 6:32 a.m. ET, Nov. 6

Status

Partially confirmed. Anthropic's research page, published November 5, 2026, describes 207,000 research sessions and how the lab says it labeled them. The counts and the scenarios are the company's. This desk did not repeat the count.
  • Confirmed
    The page exists, is dated November 5, 2026, and says what this story quotes.
  • Partially confirmed
    The session count, the task shares, and the doubling scenario are Anthropic's measurements and model.
Anthropic's research page, published November 5, 2026, says the company studied 207,000 Claude and Claude Code sessions from scientists. It says another model labeled each session, and the team kept those labels rather than the chats.1
The page says the point is to measure how scientists use Claude while protecting the details of their work. A classifier, which is a model that sorts text into preset categories, reads each session and applies a label such as the field or the task. The page says those labels are the only information from the conversations that the research team keeps, and that results are reported only in aggregate.1
The sessions it describes are on consumer plans. Relative to the number of researchers in each field, the page says Claude is used more in computer science, mathematics, and statistics, and less in health and in lab-based physical sciences.1
Data analysis, the page says, accounts for about one-third of sessions on Claude. It accounts for nearly half of sessions on Claude Code. Scientists use it less for coming up with ideas and for gathering data, even though a survey on the page says they rank those tasks as the most important for discovery.1
Why it matters (Estimate). Estimate. A reader trying to judge whether these tools are speeding science gets a map of tasks, not a result another group has repeated. The doubling figure is a scenario. The underlying chats are not in the page.
The page's model says that if today's systems were widely adopted, they could double research output in some fields. That figure is the company's scenario, not a count of papers another lab has tallied.1
How we know

Notes

  1. 1.Anthropic, "AI for scientific discovery: How researchers use Claude," published Nov 5, 2026. "Using privacy-preserving methods, we study 207,000 Claude and Claude Code research sessions." "Our goal with this research is to measure the different ways scientists use Claude while protecting the details of their work. To do this, we use privacy-preserving infrastructure to analyze the scientific use of Claude on consumer plans. A language model, used here as a classifier, reads each session transcript and labels it." "The resulting labels ... are the only information from these conversations retained and analyzed by the research team for this report." "Results are reported only in aggregate." "Data analysis alone accounts for about one-third of sessions on Claude and nearly half of sessions on Claude Code." "If existing AI capabilities are widely adopted, they could double research output in some fields." Lead authors named on the page: Johannes Hermle, Ben Jones, and Lukas Althoff. The taxonomy note says SciNet contains 7,262 tasks, spanning 34 fields and 318 subfields. Retrieved 6:18 a.m. ET. https://www.anthropic.com/research/how-researchers-use-claude

What is still unknown or disputed

  • Whether another group can repeat the 207,000 count. The page describes Anthropic's own labels.
  • The survey's sample size. The summary this desk read does not print it.
  • What share of all scientific work, rather than of Claude sessions, the labels represent. The page is about sessions on consumer plans.

Primary sources

Every claim in this story is drawn from the documents below. If a fetch failed, that is recorded on the card.

  1. Source 1

    AI for scientific discovery: How researchers use Claude

    Anthropic · November 5, 2026

    The page says it studied 207,000 Claude and Claude Code research sessions with privacy-preserving labels, published November 5, 2026. Retrieved 6:18 a.m. ET.

    https://www.anthropic.com/research/how-researchers-use-claude