Anthropic Releases AI Fluency Index From 9,830 Claude Conversations

What the report measures

On February 23, 2026, Anthropic published its AI Fluency Index, an attempt to measure how effectively people collaborate with AI rather than how often they use it. The study analyzed 9,830 anonymized multi-turn conversations on Claude.ai collected during a seven-day window in January 2026.

The team used the 4D AI Fluency Framework, which defines 24 collaboration behaviors. Anthropic said only 11 of those were directly observable in chat logs, and those 11 formed the basis of this first index.

Main findings

• Iteration and refinement appeared in 85.7% of sampled conversations.
• Conversations with iteration/refinement showed 2.67 additional fluency behaviors on average, versus 1.33 in non-iterative conversations.
• In artifact-producing conversations (code, documents, apps, interactive outputs), users became more directive but less evaluative, with lower rates of missing-context detection (-5.2pp), fact-checking (-3.7pp), and requests for reasoning (-3.1pp).

Why this matters

The report positions AI fluency as a skill-development problem: users who stay in iterative dialogue appear to apply stronger collaboration behaviors, while polished outputs may reduce critical checking. Anthropic frames this release as a baseline for longitudinal tracking, with future work planned on less-observable behaviors and causal interventions.

Primary sources: Anthropic research post and the original X announcement.