BMJ Open Study Flags Fabricated Sources in Nearly Half of AI Health Chatbot Responses

A peer-reviewed audit published in BMJ Open has found that nearly 50% of health-related responses from five major AI chatbots contained problematic content, including fabricated sources delivered with high confidence. The study evaluated responses across multiple health queries and measured accuracy against established medical standards.

Researchers tested five unnamed leading AI chatbots on a range of health-related questions. The audit revealed a pattern of responses that appeared credible but contained inaccuracies, invented citations, and unverified medical claims. The confident tone of these outputs made it difficult for users to distinguish reliable information from fabricated content, raising concerns about the readiness of AI systems for health guidance roles.

The findings signal significant risks for individuals relying on AI chatbots as initial health resources. Health authorities and platform developers now face pressure to implement stronger safeguards, transparent sourcing, and clearer limitations before these tools are integrated into medical information ecosystems. The study's peer-reviewed status lends additional weight to calls for oversight and standardized evaluation frameworks.