Quality, transparency, and readability of artificial intelligence chatbot responses to fibrous dysplasia-related searches: a cross-sectional comparative study

Samenvatting

Introduction Fibrous dysplasia (FD) ranges from incidental monostotic lesions to complex craniofacial, polyostotic, and syndromic disease. Short online searches may therefore represent different clinical questions. We compared the quality, transparency, and readability of FD-related responses from four consumer artificial intelligence (AI) chatbots. Methods In this cross-sectional online study, 10 unmodified English-language queries from a worldwide Google Trends dataset were submitted once to ChatGPT 5.5, Gemini 3.5 Flash, Microsoft Copilot, and Perplexity on August 11, 2026, yielding 40 responses. Responses were evaluated with DISCERN, a prespecified 20-item binary EQIP framework, GQS, and the JAMA benchmark criteria. Six readability indices were calculated. Between-model comparisons used Kruskal–Wallis tests with epsilon-squared effect sizes, followed where appropriate by Dunn post-hoc comparisons with Bonferroni adjustment. Results DISCERN, EQIP, JAMA, and GQS differed among chatbots ( p < 0.001, < 0.001, < 0.001, and 0.037, respectively). Perplexity had the highest median DISCERN [43.00 (41.00, 45.00)], EQIP [70.00 (66.25, 73.75)], and JAMA [1.00 (1.00, 1.00)] scores. All six readability indices differed among models (all p ≤ 0.037), and the formula-based scores indicated difficult text. Discussion The chatbots showed different profiles across the evaluated quality, transparency, and readability measures. Perplexity achieved higher scores on several evaluated instruments, while visible transparency cues remained limited across models. The findings describe a single-time-point, single-generation assessment of this selected query set.