Anthropic Insights

Anthropic Insights is an internal analytics tool that lets Anthropic understand how millions of Claude conversations are being used without anyone actually reading them. It works a bit like a survey company that never publishes any individual's answers, only aggregate tables showing 'X percent of respondents said this.' When a researcher asks a question like 'What kind of help is this person requesting?', an AI tags each conversation with an answer, and only the resulting category-by-category percentages are shown. No one ever opens the original conversations — only the final aggregate numbers remain.

The tool was originally called Clio, and has recently been renamed Anthropic Insights. For a long time it was used only inside Anthropic, but recently the company opened access to it to three outside groups — research labs at Stanford and Oxford, and the frontier AI evaluation nonprofit METR — drawing attention as an attempt to chip away at the problem of real-world usage data being concentrated in the hands of just a few companies.

Articles tend to introduce it as "an internal Anthropic tool, originally called Clio." Because the name sounds like a product feature, it's easy to mistake it for a chatbot function, but it's actually not something end users interact with directly — it's an internal research system Anthropic and outside researchers use to analyze conversation data through anonymized aggregation.