In a recent TIME magazine article, Sudheesha Perera, a resident physician at Yale New Haven Health, and Dr. Murali Doraiswamy, a professor of psychiatry and medicine at Duke University, walk readers through how to get more out of AI chatbots when asking about health. They note that tools like ChatGPT, Claude, and Gemini are getting better at synthesizing medical information and that more people are using them for advice. Two recent moves show how fast things are shifting: Amazon Health AI now offers Prime members personalized guidance that can interpret labs, and a Utah startup called Doctronic has been cleared to autonomously renew prescriptions through an AI bot.
The authors argue that these tools fill a real gap left by physician shortages and long waits for specialists, which makes it worth knowing both how to pick a chatbot and how to use it well. Perera and Doraiswamy remind users a privacy issue while using chatbots, such as, typing symptoms into a chatbot hands health data to a company not bound by HIPAA, so faster, more personalized advice comes with some privacy risk. They suggest leaving out full names, birthdates, and street addresses, nudging the age slightly, and using an adjacent ZIP code for clinical trial searches. The major chatbots are also not interchangeable. A 2025 Duke study graded ChatGPT, Claude, Gemini, Copilot, and Perplexity on advice for a common back-pain condition and found meaningful gaps, and purpose-built tools like OpenEvidence, ChatGPT Health, and Amazon Health AI may offer more evidence-based answers than a free general chatbot.
Once a tool is chosen, here are the five tips Perera and Doraiswamy recommend.
- Stress test the answer: Push back the way a good doctor pushes a colleague. Language models tend to be sycophantic and agree with whatever framing they get, so a worried patient gets reassured and a self-diagnosing one gets validated. If the bot says symptoms are probably nothing, reply with something like “I am still worried, give me two alternative explanations and explain why each is plausible” to force a real differential diagnosis.
- Ask for two high-impact articles or consensus guidelines: Request references from a clinical guideline, expert consensus statement, or systematic review in a journal such as the New England Journal of Medicine, JAMA, or the Lancet, then paste each title into a search engine to confirm it is real and says what the bot claims. LLMs sometimes fabricate plausible citations, a failure mode called hallucination, so an answer with no credible sources is not one to act on.
- Ask the same question three different ways: These models are probabilistic, so answers shift with the wording. An Oxford study showed two people describing the same condition got different advice when one called it “the worst headache ever” (told to go to the ER) and the other called it a “terrible headache” (told to take aspirin and stay home). Vary how the symptom and time course are described, and trust answers that hold steady across framings.
- Be complete and honest, just as you would with your doctor: The bot’s quality of advice depends on the detail provided. Sharing medical history, lab results, medications, and lifestyle habits leads to a more personalized answer, while minimizing or embellishing risks a confident reply that is quietly off the mark.
- Use it to prepare for your doctor: Close the session with “What are the three most important questions I should ask my doctor at my next visit?” That turns the chatbot from a stand-in for care into a tool that supports the patient-physician relationship. Bringing a printed summary along helps make better use of a 15-minute appointment.
Used with care, the authors say, AI chatbots offer real benefits: round-the-clock access to information, more agency over one’s own care, less stigma around sensitive questions, and better appointment quality. Even so, they close with a steady reminder that AI is not yet a proven substitute for an experienced doctor, and that a chatbot does not worry about a patient’s wellbeing the way a clinician does. The best healers pair technical know-how with judgment and empathy, qualities the technology has not learned to provide.
Acknowledgement: This summary was generated using AI writing tools as part of my AI Explorations process. For more information, please refer to the original article listed below.