Generative artificial intelligence-driven chatbots and medical misinformation: an accuracy, referencing and...

Area of Science:

• Artificial Intelligence
• Medical Informatics
• Public Health

Background:

• AI chatbots are increasingly used for health information by non-experts.
• This widespread use necessitates an evaluation of their accuracy in health-related queries.
• Concerns exist regarding the potential for AI to generate medical misinformation.

Purpose of the Study:

• To audit the accuracy and safety of responses from popular AI chatbots on health and medical topics.
• To assess the quality of citations and readability of AI-generated health information.
• To identify potential risks associated with AI chatbot use for everyday health queries.

Main Methods:

• Five popular AI chatbots (Gemini, DeepSeek, Meta AI, ChatGPT, Grok) were evaluated.
• 10 questions each from cancer, vaccines, stem cells, nutrition, and athletic performance categories were used.
• Responses were rated by experts for problematic content, citation accuracy, and readability using a predefined matrix.

Main Results:

• 49.6% of responses were problematic (30% somewhat, 19.6% highly problematic).
• Grok chatbot showed a higher rate of highly problematic responses (p=0.038).
• Response quality varied by topic, with weakest performance in stem cells, athletic performance, and nutrition. Citations were incomplete and often fabricated; readability was difficult.

Conclusions:

• AI chatbots demonstrate poor performance in answering health and medical questions prone to misinformation.
• The confident and certain nature of responses, coupled with poor reference quality, poses a significant risk.
• Urgent need for public education and regulatory oversight to mitigate AI-driven health misinformation.