Objective: This observational, cross-sectional simulation study evaluated ChatGPT-4 as a postoperative information tool for rhinoplasty using standardized questions and blinded ENT specialist ratings. Study Design: This study is an observational, cross-sectional simulation study using blinded expert evaluation. Setting: We used an online Artificial Intelligence (AI) platform accessed under standardized conditions. Methods: Ten typical recovery questions were posed to ChatGPT-4, and the responses were independently rated by ENT specialists for accuracy, clarity, relevance, response time, and patient-centered communication. Responses were also assessed with a structured performance instrument and supported by linguistic and statistical analyses. Results: ChatGPT-4 achieved high scores for accuracy (90%, 95% CI: 84.9–95.1) and clarity (87%, 95% CI: 82.8–91.2), but lower performance in patient-centered communication (77%, 95% CI: 74.0–80.0). Specialist scoring confirmed structured medical reasoning, while machine learning analyses highlighted clarity, diagnostic depth, and empathy as key contributors to higher ratings. Conclusions: ChatGPT-4 demonstrated high clinician-rated accuracy and clarity when answering standardized postoperative rhinoplasty questions, while patient-centered communication remained comparatively lower. These findings suggest that LLM-based tools may complement clinician-delivered postoperative counseling under appropriate oversight, but they are not a substitute for individualized medical advice or surgical follow-up.
16 September 2026 in European Archives of Oto-Rhino-Laryngology
Background: Artificial intelligence (AI) language models are increasingly used in surgical aftercare, yet their performance varies across platforms. The objective of this study is to compare the effectiveness of large language models in providing accurate, clinically relevant guidance for postoperative otoplasty. Methods: Ten commonly encountered postoperative otoplasty questions were presented to both models. The generated answers were independently assessed by ten ENT specialists using structured Likert-based instruments and predefined clinical evaluation. To evaluate reliability and inter-model differences, a range of advanced statistical techniques was applied, including t -tests, effect size calculations, sensitivity and specificity analyses, mixed-effects models, and regression-based modeling. Results: Claude 3.5 Sonnet outperformed ChatGPT-5.0 across all evaluation metrics ( p < 0.001); mixed-effects modeling showed a positive model effect ( β = 0.752), question-level ROC analysis demonstrated complete separation (AUC = 1.00), PCA supported a dominant single factor explaining 70.86% of variance in clinician ratings, and inter-rater agreement was higher for Claude 3.5 Sonnet. Conclusion: Claude 3.5 Sonnet model exhibited higher accuracy and clinical relevance in postoperative otoplasty management, with robust statistical validation supporting its reliability in surgical aftercare.