Researchers recently investigated chatbot medical information accuracy across five major platforms: Gemini, DeepSeek, Meta AI, ChatGPT, and Grok. Consequently, the findings published in BMJ Open highlight significant risks for patients who rely on these tools for health advice.

Evaluating Chatbot Medical Information Accuracy in Practice

The cross-sectional audit analyzed 250 responses to questions about cancer, vaccines, and nutrition. Furthermore, the study revealed that nearly half of the generated advice was problematic. Specifically, 30% of responses were somewhat problematic, while 19.6% were highly problematic. These outputs often presented a false balance between scientific evidence and non-science-based claims.

However, performance varied significantly by topic. Chatbots demonstrated stronger accuracy regarding vaccines and cancer. Conversely, they performed poorly on topics like stem cells, nutrition, and athletic performance. Moreover, Grok produced the highest number of highly problematic responses compared to its peers.

Reference Quality and Technical Limitations

The researchers also evaluated the quality of citations provided by the AI models. Unfortunately, no chatbot provided a completely accurate reference list. The average completeness score for references was only 40%. Additionally, hallucinations remained a significant issue, as chatbots frequently fabricated citations to appear more authoritative. These tools often responded with high confidence despite providing shaky facts.

Consequently, these models do not weigh evidence or make ethical judgments. Instead, they predict word sequences based on statistical patterns from training data. Therefore, clinicians must warn patients about the limitations of AI-generated health advice and reinforce the need for professional guidance.

Frequently Asked Questions

Q1: Which chatbot generated the most problematic medical information?

The study found that Grok, developed by xAI, produced significantly more highly problematic responses than other models like ChatGPT or Gemini.

Q2: Were the chatbots able to provide accurate medical references?

No, the audit revealed that no chatbot provided a fully accurate reference list, with an average reference completeness score of only 40% due to hallucinations.

Q3: On which topics did the AI chatbots perform the best?

The chatbots showed the strongest performance in answering questions related to vaccines and cancer, though errors still occurred in over 20% of those cases.

Disclaimer: This content is for informational and educational purposes only. It does not constitute medical advice or replace professional judgment. Refer to the latest local and national guidelines for clinical practice.

References

1. Medical information presented by chatbots inaccurate, incomplete: Study - ETHealthworld


2. Generative artificial intelligence-driven chatbots and medical misinformation: an accuracy, referencing and readability audit - BMJ Open (2026)


3. Evaluating Large Language Models on Misinformation-Prone Health Questions - Lundquist Institute for Biomedical Innovation (2026)