Matching clinical claims to evidence for the use of artificial intelligence (AI) in medicine

Technical performance alone does not establish the clinical value of artificial intelligence (AI). Clinical studies can reveal favorable expert ratings, limited clinician engagement, and uncertain effects on patient outcomes within the same intervention. We present an editorial synthesis with six complementary evidence domains: technical performance, external validity, human-AI interaction, workflow integration, clinical utility and outcomes, and lifecycle performance. These domains identify the questions addressed by a study. Each claim should be assessed with respect to study design, risk of bias, performance or effect estimates, uncertainty, and the evaluated setting. Intended use, autonomy, and potential harm determine the evidence required across domains. Journals should distinguish improvements in intermediate measures from demonstrated patient or health-system benefits and require reassessment when changes potentially invalidate earlier findings. The framework supports evaluation of AI-enabled care as delivered in a clinical system.