Evaluation of ChatGpt-5 and DeepSeek R1 for patient information on talus osteochondral lesions: a cross-sectional comparative content analysis of quality and readability

Background: Talus osteochondral lesions are joint pathologies commonly observed in athletes that can significantly impair ankle function. Artificial intelligence–based models are increasingly being used in orthopedic practice for patient education and clinical decision support. This study aimed to comparatively evaluate the responses of two large language models, ChatGPT-5 and DeepSeek R1, to frequently asked questions about talus osteochondral lesions in terms of content quality and readability. Methods: This cross-sectional comparative content analysis study included 12 of the most frequently asked questions about talus osteochondral lesions, which were presented identically in both models. The content quality of the responses was evaluated by two independent orthopedic specialists using a four-level grading system and a 4-point Likert scale (accuracy, clarity, comprehensiveness, and consistency). Readability was analyzed using the Flesch–Kincaid Grade Level (FKGL) and Flesch Reading Ease Score (FRES) indices. Results: DeepSeek R1 demonstrated a significant advantage in clarity ( P = 0.012), whereas ChatGPT-5 achieved higher comprehensiveness scores ( P = 0.017). In the readability analysis, DeepSeek R1 showed higher FRES scores (53.33 ± 6.47) and lower FKGL values (8.76 ± 1.06), indicating that its responses were easier to read and understand. Discussion: Both artificial intelligence models provided accurate and accessible information on orthopedic topics. ChatGPT-5 generated more comprehensive and academically oriented content, whereas DeepSeek R1 offered simpler and more readable responses. These findings suggest that the two models may complement each other in developing patient education materials, each demonstrating distinct strengths.