Moderate evidence
Published recordHealthProblem Space
More
Quality, safety, and readability of large language model responses to patient-derived questions on tooth whitening, resin infiltration, composite bonding, and porcelain veneers
- Truvace record
- TRV-2026-2988raw data (JSON)
- Version
- v1Published
- First published
- Evidence type
- Benchmark or lab evaluation
- Sources
- 1 sourcePeer-reviewed
- sha256
- 78d757ff5134…check it
Patients are increasingly consulting large language model (LLM) platforms for information regarding aesthetic dental procedures prior to clinical appointments. This study evaluated the quality, safety, and readability of responses produced by ChatGPT, Gemini, and Meta AI in relation to patient-derived questions. Forty-two patient-derived questions regarding tooth whitening, resin infiltration, composite bonding, and porcelain veneers were presented to each platform. Two calibrated evaluators independently assess…
The problem
In a study scoring 126 chatbot answers to 42 patient questions on whitening, bonding and veneers, none of ChatGPT, Gemini or Meta AI met the Grade 8 reading target (Flesch-Kincaid 10.2 to 16.5) or all benchmarks for complete, safe answers.
What this doesn’t fix
Initial inter-rater agreement between the two evaluators was only slight (pre-consensus kappa 0.011 to 0.191), so final scores depended on consensus discussion to resolve disagreements.
Sources
1 on file- The Saudi Dental Journaldoi:10.1007/s44445-026-00265-6
Cite this record
v1Truvace Impact Record TRV-2026-2988, v1: “Quality, safety, and readability of large language model responses to patient-derived questions on tooth whitening, resin infiltration, composite bonding, and porcelain veneers.” Truvace, 2026-11-17. /record/TRV-2026-2988 (accessed on the day you copy it). sha256 78d757ff51345dbc…
Change history
1 versionEvery change made to this record since it was first published, and why. Nothing has changed since it went up.
Certified into the record
Check this record hasn’t been altered
How to check it yourself, without trusting this page
Download any version from /api/record/TRV-2026-2988 and hash its canonical field yourself, for example with shasum -a 256. You should get its content_hash. Each version’s text also ends with prev:and the hash of the version before it (version 1 points to 64 zeros). If anyone had changed a single character of any version after it was saved, those hashes wouldn’t line up.
ace