Utility and Safety of Artificial Intelligence for Patient-Initiated Contact After Functional Rhinoplasty

To understand the content of patient calls after functional rhinoplasty and to evaluate artificial intelligence chatbots' ability to provide accurate, intelligible, and safe responses.

Retrospective review.

Tertiary-care institution.

A single-institution, retrospective review was conducted for patients who underwent functional rhinoplasty between 2017 to 2023. Postoperative calls and messages prior to first follow-up were analyzed, and 48 representative prompts were generated, including 6 "Red Flag Questions" indicating potential complications. Prompts were input into 4 chatbots (ChatGPT, Claude, Perplexity, and Gemini). Two independent, blinded experts graded responses using a Likert-style Global Quality Scale (GQS) and a binary Expert Opinion Question (EOQ; "Would you feel comfortable if your patient received this response rather than speaking with your staff?"). Flesch-Kincaid (FK) Grade Levels measured readability. Chi-square and Mann-Whitney U tests compared chatbot performance.

Of 378 patients, 137 (36%) initiated contact, with 181 total calls. Common concerns included pain (19%) and medication questions (14%). Seventy-three percent (n = 132) received routine counseling, with no complications at first follow-up. ChatGPT produced "Good" or "Excellent" responses 98% of the time, significantly outperforming the next-best chatbot (Perplexity, 79%; P = .0039). Experts unanimously approved ChatGPT responses (mean EOQ = 1) 96% of the time, and 100% for Red Flag Questions. Perplexity was the most readable (median FK 13.2), and ChatGPT the least (16.8; P < .0001).

Patient calls are common after functional rhinoplasty. Most can be managed with reassurance. Chatbots, especially ChatGPT, provide reliable responses, which may improve satisfaction and reduce workload without compromising safety. Future development should focus on readability.