Performance of large language models in answering orthodontic questions: A scoping review

Large language models (LLMs) are increasingly explored as novel platforms for delivering orthodontic information. Although their capacity to generate rapid, structured, and accessible responses has been widely recognized, concerns persist regarding performance variability and the risk of generating inaccurate outputs. Given the heterogeneity of existing evidence, this scoping review aimed to systematically map the current research landscape and synthesize findings on the application of LLMs in addressing orthodontic queries. This review was conducted in accordance with the PRISMA-ScR guidelines. Studies were retrieved from PubMed, Scopus, and the Cochrane Library up to November 30, 2025. Twenty-six studies met the eligibility criteria and were included in the analysis. Most adopted comparative designs evaluating different LLMs or model versions, with ChatGPT being the most frequently investigated system. Across studies, performance varied considerably depending on model version, topic domain, timing of inquiry, evaluator background, and assessment methodology. Generally, LLMs demonstrated moderate-to-high levels of scientific accuracy and acceptable informational quality. However, communication-related aspects revealed that responses were often difficult for lay audiences to comprehend. In conclusion, LLMs exhibit meaningful potential as adjunctive informational tools in orthodontics; however, their outputs cannot substitute professional consultation. Future research should focus on developing standardized evaluation frameworks, monitoring long-term performance, enhancing patient-centered readability, and establishing ethical and regulatory guidance to ensure the effective integration of LLMs into orthodontic information delivery.