Guideline Concordance of ChatGPT-Generated Responses on Antibiotic Prophylaxis in Endodontics.

Abstract

Antibiotic prophylaxis decision-making in endodontics represents a core competency in dental education, requiring the integration of systemic risk assessment, guideline interpretation, and antimicrobial stewardship principles. Although artificial intelligence (AI) tools such as ChatGPT are increasingly accessed by students and clinicians for clinical guidance, their educational reliability in teaching evidence-based antibiotic prophylaxis remains unclear. This descriptive study evaluated the accuracy and guideline concordance of ChatGPT-5 responses to predefined clinical scenarios in endodontics. Seventeen clinically relevant questions grounded in current guidelines from the American Heart Association (AHA), American Association of Endodontists (AAE), and American Dental Association (ADA) were submitted to the model. Responses were independently assessed by an experienced endodontist using a five-point Likert scale addressing clinical correctness, guideline concordance, and potential educational risk. Quantitative agreement was summarized descriptively, and a directed qualitative content analysis examined patterns of correct indication, expanded indication, depth of clinical reasoning, and risk of misinterpretation from an educational perspective. ChatGPT demonstrated full agreement with expert, guideline-based recommendations in 10 of 17 scenarios (58.8%), particularly in classical teaching domains such as infective endocarditis prophylaxis, antibiotic selection, dosage, and timing. Partial agreement was observed in four scenarios (23.5%), primarily due to unnecessary expansion of prophylactic indications. Disagreement occurred in three scenarios (17.6%), mainly involving immunocompromised conditions, including HIV infection with low CD4 counts and chemotherapy-related neutropenia. Although no response was deemed clinically unsafe, expanded recommendations in medically complex cases represented a potential educational risk for inappropriate antibiotic prescribing. The study supports cautious, guideline-based interpretation of AI-generated clinical information rather than conclusions about educational effectiveness or curricular implementation.