Scholarlink

Purpose <p>This study assessed the accuracy and repeatability of the orthodontics-related information generated by Chat Generative Pre-trained Transformer (ChatGPT, model GPT-4o, 18&#xa0;July 2024, OpenAI, San Francisco, CA, USA) and evaluated its usefulness for patient education by comparing artificial intelligence (AI)-generated responses to questions that patients frequently search for with responses from two orthodontic specialists.</p> Materials and methods <p>In January and February 2025, 30 descriptive questions (15 on basic orthodontics and 15 on clinically advanced orthodontics) on nondecision, informational content in patient education were assessed, including a&#xa0;“briefly summarize within 500 characters” condition. Accuracy was defined as a&#xa0;response match between ChatGPT and the orthodontist, repeatability was the consistency of ChatGPT output over two iterations. Evaluations used a&#xa0;5-point Likert scale for accuracy and a&#xa0;5-point Global Quality Score (GQS) for repeatability. Data were analyzed using R, Wilcoxon signed-rank test and the Mann–Whitney&#xa0;U test.</p> Results <p>Repeated responses from ChatGPT showed high repeatability with consistent overall accuracy. For basic orthodontics questions, accuracy increased slightly from 4.27 ± 1.03 to 4.53 ± 0.83 (<i>p</i> = 0.203). For clinically advanced orthodontics questions, accuracy remained stable (first: 3.67 ± 1.05, second: 3.73 ± 0.46, <i>p</i> = 0.850). In terms of repeatability and quality assessed by GQS, basic orthodontic questions scored moderately (first: 3.13 ± 0.83, second: 3.27 ± 0.70, <i>p</i> = 0.580), whereas clinically advanced orthodontics questions received higher GQS scores (first: 4.20 ± 0.77, second: 3.80 ± 0.56, <i>p</i> = 0.095), indicating potential applicability in patient education contexts.</p> Conclusion <p>The accuracy and repeatability of ChatGPT’s responses varied by question type: basic questions were more accurate, while clinically advanced orthodontic questions resulted in higher repeatability and quality. As these findings are limited to patient education and general information delivery, ChatGPT should not be considered a&#xa0;replacement for professional orthodontic expertise.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

ChatGPT performance in orthodontics

  • 期刊论文

  • 发表日期：

    2026年1月30日

查看全文
  • Heejin Park,
  • SungHyun Lee,
  • KyungMin Clara Lee

摘要

Purpose

This study assessed the accuracy and repeatability of the orthodontics-related information generated by Chat Generative Pre-trained Transformer (ChatGPT, model GPT-4o, 18 July 2024, OpenAI, San Francisco, CA, USA) and evaluated its usefulness for patient education by comparing artificial intelligence (AI)-generated responses to questions that patients frequently search for with responses from two orthodontic specialists.

Materials and methods

In January and February 2025, 30 descriptive questions (15 on basic orthodontics and 15 on clinically advanced orthodontics) on nondecision, informational content in patient education were assessed, including a “briefly summarize within 500 characters” condition. Accuracy was defined as a response match between ChatGPT and the orthodontist, repeatability was the consistency of ChatGPT output over two iterations. Evaluations used a 5-point Likert scale for accuracy and a 5-point Global Quality Score (GQS) for repeatability. Data were analyzed using R, Wilcoxon signed-rank test and the Mann–Whitney U test.

Results

Repeated responses from ChatGPT showed high repeatability with consistent overall accuracy. For basic orthodontics questions, accuracy increased slightly from 4.27 ± 1.03 to 4.53 ± 0.83 (p = 0.203). For clinically advanced orthodontics questions, accuracy remained stable (first: 3.67 ± 1.05, second: 3.73 ± 0.46, p = 0.850). In terms of repeatability and quality assessed by GQS, basic orthodontic questions scored moderately (first: 3.13 ± 0.83, second: 3.27 ± 0.70, p = 0.580), whereas clinically advanced orthodontics questions received higher GQS scores (first: 4.20 ± 0.77, second: 3.80 ± 0.56, p = 0.095), indicating potential applicability in patient education contexts.

Conclusion

The accuracy and repeatability of ChatGPT’s responses varied by question type: basic questions were more accurate, while clinically advanced orthodontic questions resulted in higher repeatability and quality. As these findings are limited to patient education and general information delivery, ChatGPT should not be considered a replacement for professional orthodontic expertise.