Should you trust health advice from an AI chatbot?

AI chatbots are increasingly being used for health advice, but a new BBC investigation reveals they’re dangerously unreliable when used by the public—despite performing well in controlled tests with expert input. A University of Oxford study found AI correctly diagnosed medical scenarios 95% of the time when fed precise clinical data, but accuracy plummeted to just 35% when real people described their symptoms. The problem lies in how humans phrase vague, emotional, or incomplete symptoms—leading AI to reinforce biases, miss red flags, and sometimes even recommend dangerous alternative therapies. Experts warn that AI’s overconfident tone and training on unreliable internet sources make it a poor substitute for medical professionals. While future specialized health AI tools may improve, none are currently subject to clinical trials, and doctors remain skeptical. The takeaway? Use AI as a starting point, not a diagnosis—always consult a real healthcare provider. The episode underscores a growing public risk: people are trusting AI with their health, often because it’s convenient, private, or less intimidating than a doctor. But the consequences of misdiagnosis—delayed treatment, false reassurance, or harmful advice—can be life-threatening. Listeners share mixed experiences: some found AI helpful for basic symptom checks, others received dangerous or inaccurate recommendations. The core message is clear: AI is not a doctor, and its confidence is not a guarantee of correctness.