Accepted for/Published in: JMIR Formative Research
Date Submitted: Oct 31, 2024
Date Accepted: Sep 10, 2025
Fact-checking large language model responses to a healthcare prompt: A comparison study
ABSTRACT
Background:
Generative artificial intelligence refers to algorithms that can generate text, images or other data types. Tools such as ChatGPT offer a new opportunity for patients and clinicians to access health information.
Objective:
Evaluate the accuracy and efficiency of automated fact-checking by a large language model.
Methods:
Design: Parallel comparison of a large language model to an expert human using a clinical scenario. Setting: A 23-year-old female questioning the safety of retinoid treatment for acne by sending prompts to a large language model (ChatGPT). Interventions: We asked ChatGPT to suggest improvements to a patient’s initial prompt and compared a clinical expert’s evaluation of the large language model's responses to a fact-check undertaken by the large language model. Outcome measures: Accuracy and consistency of health-related claims and the time to complete fact-checking.
Results:
There was 86% agreement between the clinical expert and the large language model for fact-checking. The expert review took 18 minutes, and the model 42 seconds. The model responses had some inconsistency but had zero fabrication and no obvious omission.
Conclusions:
Large language models can improve prompts and conduct efficient fact-checking. Human experts need to perform additional checks.
Citation
Request queued. Please wait while the file is being generated. It may take some time.
Copyright
© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.