Assessing the accuracy of AI-generated responses to periodontal scaling and root planing patient FAQs
Ahmed M. KabliBackground/Objective
Periodontal diseases affect nearly half the global population, making patient education critical for successful outcomes. With increasing public reliance on artificial intelligence (AI) for health information, evaluating the accuracy of AI-generated responses to periodontal questions is essential. To assess the clinical accuracy, reliability, and consistency of ChatGPT and Gemini in answering frequently asked questions (FAQs) regarding scaling and root planing (SRP).
Methods
Fifty binary (yes/no) FAQs about SRP were developed and posed to ChatGPT and Gemini across three independent sessions (morning, afternoon, night), yielding 300 total responses. Two board-certified periodontists evaluated responses against current AAP/ADA guidelines. Accuracy, consistency, domain-specific performance, and inter-model agreement were analyzed using descriptive statistics, an exact two-sided McNemar test, chi-square, Cochran’s
Results
Gemini demonstrated significantly higher accuracy than ChatGPT (99.3% vs. 93.3%; exact two-sided McNemar test,
Conclusions
Both AI models demonstrated high clinical accuracy, with Gemini significantly outperforming ChatGPT. Consistency was good for both models. AI tools may serve as useful supplementary tools for delivering basic information. However, AI platforms must complement, and never replace, expert clinical judgment, professional diagnosis, or individualized patient consultations.