A third of American adults asked an AI chatbot a health question in the past year, according to KFF polling from March, and 41 percent of those users uploaded personal medical information. A practicing gastroenterologist warns of two pitfalls: privacy and guidance. Free versions of ChatGPT, Gemini, and Claude use conversations to train future models by default, making medical details training data. Privacy researchers demonstrated that 87 percent of Americans can be identified from ZIP code, sex, and date of birth. A chatbot sounds confident even when wrong, and its answers can be incorrect or incomplete, as it lacks knowledge of medications, family history, or omitted details. In a published case report, a man asking ChatGPT about replacing chloride substituted sodium bromide, a toxic chemical cousin of table salt, leading to three weeks in the hospital with bromide poisoning. Gallup found that one in seven recent AI health users skipped a doctor’s visit based on chatbot advice, and 11 percent reported advice they later learned was unsafe. The article offers pointers: flip the privacy switch under Data Controls in ChatGPT, with equivalents in Gemini and Claude, or use temporary mode; and strip identifiers like name, date of birth, address, phone number, and record numbers before pasting lab reports or visit summaries.