Teaching Hallucination Literacy in Medical Education: A Claim-Level Verification Framework and Toolkit

Abstract Generative artificial intelligence can produce fluent, persuasive responses that intermix accurate, unsupported, outdated, and fabricated information. Therefore, recommendations to simply “verify the output” are insufficient as a teachable competency. We operationalize hallucination literacy as five observable competencies: deconstructing AI responses into discrete claims, identifying required evidence, calibrating verification burden to the clinical context and consequences of error, evaluating evidence or obtaining accountable expert review, and documenting a final verification decision. The framework includes a model-agnostic verification workflow, claim verification matrix, worked example, instructional sequence, and formative assessment rubric for journal clubs, evidence-based medicine curricula, scholarly writing, and supervised clinical reasoning.