Accuracy and safety of ChatGPT-4o responses in rhinoplasty postoperative counseling: a panel-based study

Abstract

BackgroundChatGPT and other large language models have emerged as new tools for patient education, yet their clinical safety and reliability remain unclear.ObjectiveTo assess the accuracy and safety of ChatGPT-4o's responses to common postoperative questions following rhinoplasty.MethodsTen consensus-based postoperative questions were identified via a modified Delphi process. ChatGPT-4o responses were generated and evaluated by three independent otolaryngologists using a 10-point Likert scale. Reviewers also assessed the presence of critical errors.ResultsThe average Likert score across responses was 8.87 (95% CI, 8.39 to 9.34), No critical errors were detected. Inter-rater reliability was high (ICC(2,k) = 0.876).ConclusionsChatGPT-4o provided clinically accurate and safe answers to common rhinoplasty postoperative questions.SignificanceThese findings suggest that ChatGPT-4o may serve as a useful adjunct for postoperative patient counseling in structured settings, particularly when physician access is limited.