Anthropic's new AI Fluency Index tracks how people develop skills using AI tools like Claude. The study analyzed 9,830 anonymized conversations from January 2026, measuring 11 observable behaviors that indicate effective human-AI collaboration. The research reveals that iteration and refinement strongly correlate with AI fluency, while users show less critical evaluation when AI produces polished outputs.

Using the 4D AI Fluency Framework, Anthropic found that conversations involving back-and-forth refinement display more than double the fluency behaviors compared to quick exchanges. Users who iterate on responses are 5.6x more likely to question Claude's reasoning and 4x more likely to identify missing context. However, when Claude creates artifacts like code, documents, or apps, users become more directive but less evaluative, questioning reasoning 3.1 percentage points less and checking facts 3.7 percentage points less.

This pattern suggests that polished-looking AI outputs may create a false sense of completion, reducing critical evaluation. The findings highlight a crucial gap: as AI becomes more capable of producing professional-quality work, users need stronger skills to critically assess outputs. The study establishes a baseline for tracking how AI fluency evolves as models advance, emphasizing that effective AI use requires ongoing refinement rather than simple delegation.