Securing Safety and Quality in AI-generated Patient Education

Generative artificial intelligence offers personalized patient education, yet clinical inaccuracy and lack of theoretical grounding threaten health care safety. This study validates a “nurse-led” AI protocol for generating safe, theory-guided digital education materials based on Kolcaba’s Comfort Theory. A methodological design established a three-stage iterative prompt engineering process (initial, clinical refinement, and theoretical alignment). Stroke, chronic kidney disease, and COPD served as case models. Quality was assessed via expert panel reviews (n=3) using the Content Validity Index and Ateşman’s Readability Index. The protocol effectively mitigated “AI hallucinations.” Theoretical integration ensured holistic alignment across comfort domains. Readability scores significantly improved from 48.5 to 66.8. High expert consensus (CVI: 0.93-0.95) demonstrated clinical safety, proving nursing expertise is a mandatory safety layer. This framework provides a replicable quality-control protocol for nurses as digital content curators, ensuring AI-generated materials are clinically safe and theoretically sound.