Anthropic is preparing to introduce an AI Fluency scorecard inside Claude that evaluates each user’s interaction skills across 11 behavioral indicators, Testing Catalog reported on May 26. The feature treats prompting and conversational design as a measurable competence rather than an undifferentiated user attribute.
The behavioral indicators are not fully disclosed in Testing Catalog’s reporting, but the framing implies a structured taxonomy: precision in question formulation, willingness to iterate, willingness to provide context, structured task decomposition, appropriate skepticism toward outputs, and similar dimensions that distinguish effective Claude users from frustrated ones.
The product implications cut in two directions. For Claude, formalizing user fluency provides a calibration signal for which users get more capable model defaults, longer context windows, or earlier access to new features. For users, the scorecard creates a feedback loop that incentivizes learning specific prompting patterns Claude’s own evaluator considers effective. The risk is that users optimize for the score rather than for their actual outcomes, the standard problem with any visible metric.
Teams onboarding non-technical users to Claude Cowork or Claude.ai for the first time should watch for whether Anthropic exposes the scorecard externally; if so, it becomes a useful training reference but also a metric users may game.
Reported by Testing Catalog on 2026-05-26.
