Multi-model comparison of artificial intelligence responses to caregiver-oriented pediatric dental scenarios

Background/aim This study aimed to comparatively evaluate the quality, reliability, accuracy, and readability of responses generated by ChatGPT 5, Gemini 3, and Microsoft Copilot (free and premium versions) to caregiver-oriented pediatric dental scenarios. Materials and methods Fifteen caregiver-oriented pediatric dental case scenarios were developed using anonymized clinical photographs, radiographic images, and relevant clinical information selected from real pediatric dental cases. The scenarios were presented to ChatGPT 5, Gemini 3, and Microsoft Copilot (free and premium versions), and the AI-generated responses were independently evaluated by an experienced pediatric dentistry specialist and a senior PhD resident in pediatric dentistry using validated assessment tools, including a five-point Likert scale (accuracy), the Global Quality Scale (quality), the modified DISCERN instrument (reliability), and the Flesch Reading Ease Score (readability). Both inter- and intra-observer agreement were found to be excellent (ICC > 0.88). Statistical analyses were performed using the Kruskal–Wallis test, the Mann–Whitney U test with Bonferroni correction, one-way ANOVA, and Games–Howell post hoc tests, with the level of statistical significance set at p < 0.05. Results Significant overall differences were observed for the Global Quality Scale ( p = 0.013) and modified DISCERN scores ( p = 0.015). Although ChatGPT-5 Free achieved accuracy comparable to its premium counterpart, a significant difference was identified between the free versions of ChatGPT-5 and Microsoft Copilot ( p = 0.002). Readability differed significantly across the evaluated models ( p < 0.001), with Microsoft Copilot Premium generating the most caregiver-readable responses, whereas Microsoft Copilot Free produced the most linguistically complex outputs. Conclusion Large language models show potential to support caregiver-oriented communication in pediatric dentistry; however, they should currently be regarded as adjunctive tools rather than substitutes for professional clinical assessment. Differences in performance among platforms highlight the need for careful interpretation of AI-generated information and ongoing professional supervision.