AI chatbots fail medical misinformation test, returning inaccurate and fabricated advice
AI chatbots fail medical misinformation test, returning inaccurate and fabricated advice
Publish Date: 2026-06-01 08:25:00
Source Domain: www.psypost.org
- Study Findings: An audit of chatbot responses in health and medical fields prone to misinformation found that 49.6% of responses from various chatbots were problematic, with 30% deemed somewhat problematic and 19.6% highly problematic.
- Research Scope: The audit involved assessing five chatbots across five distinct fields prone to misinformation, namely cancer, vaccines, stem cells, nutrition, and athletic performance.
- Chatbot Performance: Chatbot performance varied significantly across the study fields, with better responses in vaccines and cancer compared to nutrition, athletic performance, and stem cells.
- Confidence and Readability: Chatbot outputs were expressed with high confidence and complexity, corresponding to a college level reading difficulty, which may not be accessible to the general public.
- Citation Accuracy: The chatbots demonstrated poor reference quality, often hallucinating – producing incorrect or unsupported statements that sound plausible.
- Study Conclusion: The study concluded that continued deployment of these chatbots without public oversight risks spreading misinformation.
- Experts’ Involvement: The paper was authored by a team of researchers including Nicholas B. Tiller, Alessandro R. Marcon, Marco Zenone, Kristin E. Kidd, Asker E. Jeukendrup, Zubin Master, and Timothy Caulfield.
- Ongoing Development: The research contributes to understanding chatbots’ current states, noting that continual development may produce different findings in future studies.
