Quality, completeness, readability, and response time of AI chatbots in the management of deep caries and pulp...

Artificial intelligence (AI) chatbots are being increasingly used to obtain health-related information; however, the quality, completeness, readability, and consistency of AI-generated responses across different languages remain insufficiently investigated in endodontics, particularly regarding the management of deep caries and pulp exposure.

Thirty open-ended questions were developed on the basis of the 2019 European Society of Endodontology (ESE) position statement and the current literature. The questions were submitted in English and Turkish to ChatGPT-4o, ChatGPT-5, Consensus Pro, and Perplexity Pro. The responses were independently evaluated by two blinded endodontists via the modified DISCERN (mDISCERN), the Global Quality Score (GQS), and a three-point completeness scale. Readability was assessed via the Flesch Reading Ease Score (FRES) for English responses and the Ateşman Readability Index for Turkish responses. Data were analyzed using two-way analysis of variance (ANOVA), intraclass correlation coefficient (ICC) analysis, and post hoc multiple comparisons.

AI chatbots and language significantly affected readability and response time (p < 0.05). Turkish responses demonstrated higher readability scores, particularly for the ChatGPT-4o and ChatGPT-5. Perplexity generated the shortest response times, whereas Consensus had the longest response times. The quality scores (mDISCERN and GQS) were significantly influenced by the AI model but not by language. Consensus achieved the highest quality ratings in both languages.

AI model selection had a greater influence on response quality than language did. Consensus achieved the highest expert-rated quality and completeness scores, whereas Perplexity generated the fastest responses. These findings highlight the importance of carefully selecting AI tools when seeking evidence-based endodontic information across different languages.