Can Parents and Patients Understand Myopia Using Large Language Model-Based Chatbots?

Abstract Purpose This study aimed to compare the reliability of myopia-related information from AI chatbots using a set of commonly asked questions by parents and patients on myopia, which is an emerging disease of the 21 st -century. Design Prospective comparative reliability study Methods The study used ChatGPT(OpenAI(2025)GPT-5), Gemini(Gemini 2.0,Google,2025) and DeepSeek (DeepSeek-R1). Twenty myopia-related questions were framed from the perspective of parents and patients, covering general questions, prevention and control, and complications of myopia. Based on their experience in the field of myopia, two senior clinicians, one junior clinician and one researcher(all≥3 years of experience in myopia) rated the responses generated by AI chatbots on a 5-point Likert scale(1:very poor, 2:poor, 3: acceptable, 4:good and 5:very good). Results Overall, combined rating for tested chatbots had median score of 4(“good”). Gemini received significantly lower ratings than other two chatbots (p≤0.001), with a median rating of 3(“acceptable”). ChatGPT and DeepSeek had median score of 4(“good”) and there was no significant difference in ratings (p=0.48). Both ChatGPT(66.0%) and DeepSeek(67.5%) had high proportions of “good” and “very good” ratings, compared to Gemini(40.0%). Combined “poor” and “very poor” ratings were highest for Gemini(7.5%), followed by ChatGPT(5.0%) and DeepSeek(4.0%). For general questions on myopia, ChatGPT and DeepSeek were rated “good”; for complications of myopia, ChatGPT was rated as “good”, while others were rated “acceptable”. Conclusions ChatGPT and DeepSeek demonstrated consistently high-quality responses, while ratings for Gemini were slightly lower but remained adequate. These findings suggest AI chatbots can support patients or parents in understanding myopia.