AI chatbots show strong accuracy in tooth whitening advic

Dental Tribune International

Thu. 11 December 2025

save

AMMAN, Jordan: A new comparative study has found that leading artificial intelligence (AI) chatbots provide largely accurate and reliable information on tooth whitening but often at a reading level too advanced for many patients. The research compared three major models and a new model and highlighted a growing need to simplify AI-generated dental guidance for the public.

This detailed analysis of more than 100 patient-derived questions showed that contemporary large language models performed consistently well in responding to common enquiries about tooth whitening. The study assessed ChatGPT-4o, Google Gemini, DeepSeek-R1 and the dentistry-specific model DentalGPT, examining their usefulness, accuracy, reliability and readability across five categories, including safety, techniques and post-treatment care.

Across all four systems, almost seven in ten responses were judged “very useful”, and none were classified as “not useful”. Evaluators using standardised scoring tools reported high marks for overall information quality and reliability, and more than 90% of outputs met the criteria for “very good” content under the CLEAR assessment framework, which rates AI-generated information for completeness, lack of false information, evidence support, appropriateness and relevance. Quality scores too were strong: nearly half of the responses achieved the highest rating.

Notably, performance did not vary significantly between the four models or across the five domains of enquiry. This suggests that, despite differences in training data and model design, large language models are now capable of delivering consistent, broadly accurate explanations on cosmetic dental procedures such as whitening. Even the domain-specific DentalGPT did not outperform the general-purpose systems, indicating that high-quality guidance on this topic is readily available across mainstream AI tools.

The study’s most prominent concern, however, was readability. Responses averaged a Flesch Reading Ease score of 36—categorised as “difficult”—and a SMOG index of 11—equivalent to a late secondary school reading level. The researchers warn that this gap between information quality and accessibility may limit the practical usefulness of AI-generated advice, particularly for patients with lower health literacy. They argue that improved simplification tools and prompt strategies may be essential if chatbots are to support patient understanding effectively.

The authors conclude that while AI systems show real promise as adjuncts to patient education in cosmetic dentistry, they cannot replace personalised professional guidance. They also recommend that models should be refined to ensure that accurate information is delivered in a format that patients can easily understand.

The study, titled “The evaluation of tooth whitening from a perspective of artificial intelligence: A comparative analytical study”, was published online on 24 November 2025 in Frontiers in Digital Health.