Can ChatGPT give medical advice?
People paste lab reports and symptoms into general chatbots and get fluent answers in seconds. Fluency is not the same as clinical judgment, and the sources below draw that line.
Arun Verma, co-founder of Triangle · 2026-10-19
This is not clinical evidence and not medical advice. It does not replace your care team. Discuss anything here with your clinician before you act on it.
Can ChatGPT give medical advice?
No. A general chatbot is not a licensed clinician, and it is not a substitute for care from the people who know your history. The American Medical Association's tip sheet for patients says to use AI to learn, not to diagnose, and not to replace your doctor. Peer-reviewed studies of chatbot health answers find fluent replies that can still be wrong, incomplete, or hard for people to use well. Anything important is still a question for your care team.
What have peer-reviewed studies found about chatbot health answers?
Useful in some settings, and not reliable enough to stand alone. In a 2023 JAMA Internal Medicine cross-sectional study, licensed evaluators preferred ChatGPT responses over physician replies to public online forum questions for rated quality and empathy; the authors said further exploration is warranted in clinical settings, such as using the chatbot to draft replies that physicians could then edit. A February 2026 Nature Medicine randomized study with 1,298 participants found a sharper gap for the public: large language models alone identified relevant conditions in 94.9% of test scenarios on average, yet participants using the same models identified relevant conditions in fewer than 34.5% of cases and the best next step in fewer than 44.2%, both no better than a control group that used ordinary home methods. The authors point to user–model interaction failures, such as incomplete details in the chat, as a central problem. A July 2024 Nature Medicine evaluation of open-access large language models on thousands of real hospital cases found they diagnosed significantly worse than physicians and were not ready for autonomous clinical decision-making. That study could not test ChatGPT because of data-use limits on the hospital records.
Can a chatbot diagnose me or decide my treatment?
No. The AMA tip sheet says not to rely on AI for diagnosis or treatment decisions, and that AI can sound confident even when it is wrong. The 2024 Nature Medicine clinical decision-making study concluded that current open-access models posed a serious risk without extensive clinician supervision when used for autonomous diagnosis and treatment planning. A chatbot reply is not a diagnosis of you. No source cited here says what any one answer means for a given person's care.
How does the FDA talk about generative AI and medical devices?
By intended use and risk, not by brand name. FDA's Clinical Decision Support Software guidance (January 2026) clarifies which software functions for health care professionals can fall outside the device definition, and it states that software functions that support or provide recommendations to patients or caregivers, not clinicians, meet the definition of a device. FDA's page on artificial intelligence-enabled medical devices says CDRH regulates those devices under the FD&C Act with a risk-based approach that considers intended use and technological characteristics. As of August 18, 2026, the FDA published a discussion paper seeking public feedback on how generative AI-enabled medical devices might be risk-assessed, evaluated before market through competency-style testing, and monitored after they reach patients, including questions about foundation models and agentic AI, with comments due October 19, 2026 under docket FDA-2026-N-7874. That paper asks how to oversee GenAI tools that are medical devices; it does not classify any named consumer chatbot.
Why does an assistant without my history answer more generally?
Because it only has what you type into the chat. The AMA tip sheet says AI cannot diagnose or know your full medical history, and that results depend on how you ask. The 2026 Nature Medicine public-user study found people often gave incomplete information in the first messages, which limited what the models could usefully return. A reply based on a pasted snippet or a short symptom list is still a general-language answer, not a read of your full record.
What uses do the sources support?
Learning and visit prep, not deciding care. The AMA tip sheet suggests prompts that explain a medical term, a doctor's diagnosis or instructions in simple words, and prompts that list questions to take to a doctor. It also says not to rely on AI for diagnosis or treatment decisions. A fair question after any answer you care about is what the clinician who ordered the test or knows the history makes of it.
What if someone asks a chatbot during a crisis?
A chatbot is not crisis care. If you are in crisis, call or text 988 in the US. The FCC says that by calling or texting 988 you connect with trained counselors in the 988 Suicide & Crisis Lifeline network, with live chat also available through the Lifeline. For sudden or severe symptoms that need emergency care, the AMA tip sheet points people to immediate care or 911 rather than to AI.
What could I ask my clinician after using a chatbot?
You could ask whether the chatbot's summary matches what is in your record, which parts need a visit or a test, and which questions from the chat are worth keeping on the agenda. A fair question is also which trusted sources your clinician prefers when you want plain-language background between visits.
Sources
- American Medical Association: AI Chatbots for Health: How to Use Safely and Effectively (2026 tip sheet PDF): The American Medical Association tip sheet for patients says to use AI to learn, not to diagnose, and lists among things not to rely on AI for: emergencies, diagnosis or treatment decisions, and replacing your doctor.
- American Medical Association: AI Chatbots for Health: How to Use Safely and Effectively (2026 tip sheet PDF): The AMA tip sheet states that AI cannot diagnose or know your full medical history, may rely on incomplete, outdated or non-vetted information, can sound confident even when it is wrong, and that results depend on how you ask.
- Ayers et al., Comparing Physician and Artificial Intelligence Chatbot Responses to Patient Questions Posted to a Public Social Media Forum, JAMA Internal Medicine 2023: In a 2023 JAMA Internal Medicine cross-sectional study of 195 public forum patient questions, licensed health care professional evaluators preferred ChatGPT responses to physician responses in 78.6% of 585 evaluations, and rated chatbot responses higher for quality and empathy; the authors conclude further exploration is warranted for drafting responses physicians could then edit.
- Bean et al., Reliability of LLMs as medical assistants for the general public: a randomized preregistered study, Nature Medicine (published 9 February 2026): In a February 2026 Nature Medicine randomized preregistered study with 1,298 participants, large language models tested alone correctly identified conditions in 94.9% of cases and disposition in 56.3% on average, yet participants using the same models identified relevant conditions in fewer than 34.5% of cases and disposition in fewer than 44.2%, both no better than the control group.
- Bean et al., Reliability of LLMs as medical assistants for the general public: a randomized preregistered study, Nature Medicine (published 9 February 2026): Bean et al. identify user interactions as a challenge, including cases where participants provided incomplete information and where correct model suggestions were not consistently followed, and conclude that standard medical benchmarks do not predict the failures found with human participants.
- Hager et al., Evaluation and mitigation of the limitations of large language models in clinical decision-making, Nature Medicine (published 4 July 2024): Hager et al. (Nature Medicine, July 2024) found that current state-of-the-art open-access large language models performed significantly worse than clinicians at diagnosis across tested pathologies on real patient cases, did not follow diagnostic or treatment guidelines consistently, and concluded they are currently not ready for autonomous clinical decision-making; ChatGPT could not be tested because of MIMIC data-use limits on external APIs.
- FDA: Clinical Decision Support Software — Guidance for Industry and Food and Drug Administration Staff (January 2026): FDA's Clinical Decision Support Software guidance (January 2026) clarifies types of CDS software functions for health care professionals that are excluded from the device definition by section 520(o)(1)(E) criteria, and states that existing digital health policies continue to apply to software functions that meet the device definition, including those intended for use by patients or caregivers.
- FDA: Clinical Decision Support Software — Guidance for Industry and Food and Drug Administration Staff (PDF, document issued January 29, 2026): FDA's CDS guidance states that software functions that support or provide recommendations to patients or caregivers, not health care professionals, meet the definition of a device.
- FDA: Artificial Intelligence-Enabled Medical Devices (Digital Health Center of Excellence): FDA's Artificial Intelligence-Enabled Medical Devices page states that CDRH regulates medical devices, including AI-enabled medical devices, under the FD&C Act using a risk-based approach that takes into consideration the device's intended use and technological characteristics.
- FDA press announcement: FDA Seeks Public Feedback to Inform Regulatory Approach for Generative AI-Enabled Medical Devices (August 18, 2026): On August 18, 2026, FDA issued a discussion paper on considerations for regulating generative AI-enabled medical devices and sought public feedback on risk assessment, premarket evaluation, postmarket monitoring, and other topics, including a possible competency-assessment approach to premarket evaluation and considerations around foundation models and agentic AI, with comments due October 19, 2026 under docket FDA-2026-N-7874.
- American Medical Association: AI Chatbots for Health: How to Use Safely and Effectively (2026 tip sheet PDF): The AMA tip sheet includes example prompts to explain a medical term, doctor's diagnosis or instructions in simple words, and to ask what questions to raise with a doctor to better understand a situation.
- Federal Communications Commission: 988 Suicide & Crisis Lifeline (updated August 28, 2025): The FCC states that by calling or texting 988, you connect with trained counselors within the 988 Suicide & Crisis Lifeline network, and that live chat services are also available at 988lifeline.org.
- American Medical Association: AI Chatbots for Health: How to Use Safely and Effectively (2026 tip sheet PDF): The AMA tip sheet says that for serious or sudden symptoms, seek immediate care or call 911, and lists emergencies among uses not to rely on AI for.
Start your 7-day free trialhealthmemory is made by Triangle. 7-day free trial, then $9 a month. Cancel anytime.