The AI Fluency Index and the Behavioural Limits of Human–AI Systems
Anthropic Research at its best
On 16 February 2026, Anthropic published The AI Fluency Index, authored by Kristen Swanson, Drew Bent, Zoe Ludwig, Rick Dakan and Joe Feller. The report examines how people interact with AI in real conversational settings, drawing on approximately 9,800 multi-turn Claude exchanges. Rather than evaluating model capability, it tracks observable user behaviours such as iteration, questioning reasoning, fact-checking, identifying missing context, and whether users shape the structure of the interaction itself. The purpose is to establish an empirical baseline for human–AI collaboration.
