Connect
Written by: Mayo Clinic Staff
AI can answer health questions in seconds. But should you trust it with your symptoms? Here's what to know before you rely on it.
Imagine you've been feeling tired for weeks. Your usual strategies, like rest and extra coffee, aren't helping. Before deciding whether to schedule an appointment with a healthcare professional, you open an artificial intelligence (AI)-powered chat tool and type, "What health conditions cause fatigue?"
Within seconds, a list of answers appears. It includes stress, anemia, thyroid issues, depression, chronic illness and cancer. The information feels organized and sounds accurate — and a little scary.
But is this answer trustworthy? And what should you do with this information?
AI-created health information is widely used
Nearly 8 in 10 adults in the U.S. turn to the internet for answers to health questions. Instead of scrolling through websites, many people find answers in the AI-generated summary that appears at the top of their search results. (1)
But it doesn't stop there. More than 1 in 5 adults worldwide are turning directly to AI chatbots, like ChatGPT and Gemini, to ask health questions. (2) It's fast, convenient and free. But Mayo Clinic experts warn that AI-generated information isn't always reliable or accurate.
Why you can't always trust AI
When it comes to using AI for health information, there are a few key limitations to keep in mind:
1. Diagnosing and treating illness is too complex for a machine
AI tools don't have access to your full medical record — and you shouldn’t upload or share it with them. These tools can't examine you or run tests the way a healthcare professional can. They don't have the ability to reason like a human or explain how they came to a conclusion. (3) These qualities are necessary for making safe and accurate medical decisions.
2. AI can be wrong, even when it sounds confident
AI chatbots give answers based on patterns in data. They don't "know" facts in the way a health professional does. Sometimes AI information sounds true but is completely incorrect. This is known as hallucination. (4, 5) For example, when asked how to get more minerals from food, AI has been known to recommend eating rocks. (6)
3. AI-created information may be biased
AI systems are trained on large amounts of data that may contain bias or gaps. That means it may not reflect everyone's experience fairly. (7, 8) For example, an AI system that learned from information about people in the United States and parts of Europe might miss signs of depression. That's because it doesn't know that in some cultures, people show sadness through physical symptoms like headaches or tiredness, rather than talking about their feelings.
4. AI spreads misinformation
AI doesn't know what's true and what's not. (6) It may pull answers from flawed or misleading sources it finds online. When people see alarming health stories online, they often forward or repost them — even if they know the information may not be true. As false information spreads, there’s more of it online. AI may then repeat that false information. (9)
5. People and AI don't communicate well together
One study found that people seeking health information don't tend to give AI chatbots enough specific information for clear, accurate answers. (4) And small differences in how symptoms are described can completely change the answer from AI, making it less accurate. (4,5) For example, when two people asked about the same symptoms but used different words, AI told only one of them the correct answer, which was to get emergency care. (4)
How to use AI more safely and effectively
If you still want to use AI to learn about a health topic, here are practical steps to reduce risk:
- Use AI for general education, not diagnosis. AI is best suited for explaining medical terms or giving general wellness advice. For example, you might ask, "What does hypertension mean in plain language?" or "How can I add more movement into my day?"
- Cross-check everything. Verify information with trusted sources, like websites for Mayo Clinic or the American Medical Association. Most importantly, review what you learn with your healthcare team. (6)
- Ask clear, specific questions. Instead of asking, "Is coughing a bad sign?" try, "What are common causes of chronic dry cough in adults?" Clear, focused questions tend to produce more useful and balanced answers. (6) Remember: Healthcare professionals are trained to ask the right follow-up questions. AI isn't. (4)
- Protect your privacy. Don't share personal information, like your full name, date of birth, address, medical records or insurance details. Even health details should be shared cautiously, especially on public or free platforms. (6)
The bottom line
Think of AI as a research assistant, not as your healthcare professional. It can be a helpful tool for summarizing ideas or getting big-picture information. AI can be extremely useful in preparing questions to ask your care team. But when it comes to your health, the safest and most effective decisions are still made with a trusted healthcare professional.
Related links:
References
- Many in U.S. consider AI-generated health information useful and reliable. Annenberg Public Policy Center. https://www.annenbergpublicpolicycenter.org/many-in-u-s-consider-ai-generated-health-information-useful-and-reliable/. Accessed Feb. 10, 2026.
- Yun HS, et al. Online health information-seeking in the era of large language models: Cross-sectional web-based survey study. Journal of Medical Internet Research. 2025; doi:10.2196/68560.
- Ullah E, et al. Challenges and barriers of using large language models (LLM) such as ChatGPT for diagnostic medicine with a focus on digital pathology — A recent scoping review. Diagnostic Pathology. 2024; doi:10.1186/s13000-024-01464-7.
- Bean AM, et al. Reliability of LLMs as medical assistants for the general public: A randomized preregistered study. Nature Medicine. 2026; doi:10.1038/s41591-025-04074-y.
- Giorgi S, et al. Evaluating generative AI responses to real-world drug-related questions. Psychiatry Research. 2024; doi:10.1016/j.psychres.2024.116058.
- What doctors wish patients knew about using AI for health tips. American Medical Association. ama-assn.org/practice-management/digital-health/what-doctors-wish-patients-knew-about-using-ai-health-tips. Accessed Feb. 17, 2026.
- Yoon SC, et al. Digital psychiatry with chatbot: Recent advances and limitations. Clinical Psychopharmacology and Neuroscience. 2025; doi:10.9758/cpn.25.1346.
- Thakkar A, et al. Artificial intelligence in positive mental health: A narrative review. Frontiers in Digital Health. 2024; doi:10.3389/fdgth.2024.1280235.
- Saeidnia HR, et al. Generative AI and health misinformation: Production, propagation, and mitigation — A systematic review. BMC Public Health. 2026; doi:10.1186/s12889-025-26148-9.
Like
Helpful
Hug
Connect
@sueinmn "... but we have people here on Connect openly saying that they get their medical advice from ChatGPT and similar sources ..."
I think it would be helpful for Mayo Clinic to have a page that states the priority that people should apply when evaluating medical treatments. Here is my version; feel free to adapt it:
1. Your doctor, provided you used common sense in picking him/her (eg, not just because he's a golfing buddy or a relative). For serious issues, it is best to pick a doctor in a research institution. Why is your doctor the first source? Because he/she knows your medical history the best, AND what the reliable studies & treatments are. Of course, you should cross-check your doctor's advice with the information from point #2 below. If there are consistent discrepancies, change your doctor & find a competent & educated one you trust. If you can't find such a doctor, you probably are unconsciously applying some bias in your selection.
2. Use established, independent medical studies that are VERIFIED with other independent studies. Remember 1989, when two chemists claimed they found a way to do "cold fusion" using palladium? The price of palladium not only went through the roof, but it also became "unobtanium" (yes, that is a word), as researchers around the world tried to reproduce the experiments. As a Caltech graduate, I didn't know much about fusion, but I knew enough to be highly skeptical. Good scientists & researchers are trained to be skeptical. Users making medical decisions should be skeptical as well. Even great discoveries in science initially face fierce skepticism: (1) The Big Bang Theory (initially a term of ridicule) in 1929 faced such resistance, with Einstein saying to the researcher, "Your math is OK, but your physics is terrible." It was 35 years before some evidence popped up, & another 10 years before the evidence was confirmed & the Nobel Prize was finally awarded in 1978. (2) Louis Pasteur faced ridicule in his development of an anthrax vaccine. (3) The woman who was recently awarded the Nobel Prize for her work in developing an mRNA vaccine for COVID, spent years of resistance. (4) There are many other examples.
My point is, good scientific discoveries usually face legitimate skepticism until VERIFIED. By contrast, "junk science" are quick attempts to bypass the scientific method of independent verification, & instead, use the media to popularize their "discoveries". Be a skeptic!
3. Use forums like this one from Mayo Clinic, but understand that unless forums like this are moderated by individuals who support the principles in point #2 above, the forum may be useless or worse. We recently had someone on this forum quoting all sorts of statistics, who claimed to be an impartial researcher, but was poorly informed & biased as well. Experiences have some value, but medicine is not an exact science. Developing statistics or reaching definitive conclusions from random (often biased) comments is worse than foolish. This is a good place to find out about treatment options, but not to make decisions without taking into account points #1 & #2 above. Again, be a skeptic!
4. Always "drill down" in AI results to the original source of the information. Ignore references to YouTube videos; these are generally even less authoritative than forums, because there is no meaningful feedback. Be REALLY skeptical of AI & YouTube!
5. Advertising is the worst! Claims of "clinically studied" or "clinically proven" mean NOTHING. Pictures of happy people (or happy dogs) are trying to manipulate you! Just IGNORE advertising claims!
6. Discard your biases. Be skeptical of yourself!
When I first came to this forum about a year ago, it was a good place to find out about options & share experiences. Around the beginning of the year, I have seen a large increase in the number of people searching for or advocating "alternate medicine". We even had someone wanting to blame prostate cancer on the COVID vaccine.
Let the reader beware!
-
Like -
Helpful -
Hug
6 Reactions@nycmusic I agree that it can dangerous. Your right. I always ask for center of excellence. But you still need to screen the data.
-
Like -
Helpful -
Hug
2 Reactions@readandlearn Thanks, I'll ask our moderators to consider a post like this. An even bigger trick will be getting members to read it.
-
Like -
Helpful -
Hug
10 ReactionsExcellent post and "the bottom line" so true! I hope everyone will read this.
-
Like -
Helpful -
Hug
7 Reactions@nycmusic I agree completely. I recently took a photograph of a younger weight lifters body and picture of me and asked chatgpt to put my face on the weight lifters body. It took me about 10 tries to get it right by asking the question different each time. But I got it. The scary thing is I could not tell the difference.
-
Like -
Helpful -
Hug
4 Reactions@readandlearn
Arthur Schopenhauer said
-
Like -
Helpful -
Hug
2 Reactions@royanthony there is no replacement for serious research…A I can help, just don’t depend on it ! I have spotted very conflicting info sometimes. Beware, it’s here to stay, but needs common sense regulation.
-
Like -
Helpful -
Hug
4 ReactionsI use A.I. all the time. I use Microsoft Co-Pilot. What I like about it is that it is the Amazon or research site. Just like Amazon it searches all information on a subject then puts into a readable format. What Co-pilot does is list where A.I. got the information and source it is referring to.
At that point you can go on and read the full report from the source versus a A.I. written summary. I do this all the time if the information is important. On Co-pilot it says "A.I. can make mistakes." And it does as is interpreting what you are putting in not the interpretation of a human being. What is scary to me is how much information is stored on you. If you type something in it remembers every thing you have asked or information provided to you and relates that input to how it answers you.
I still go on WEBMD, Mayo, Cleveland Clinic, but to asked A.I. for local business, phone numbers addresses can't be beat. I also did some old coin research and submitted a picture of a coin that I thought was irrelevant. Co-pilot came back with (sent picture) what it was and the value. It came back worth $500.00 and they were correct. I took it to a precious metal company and that is exactly what they paid me for it.
But back to topic. Getting information on selling things is vastly different that relying on A.I. for medical guidance. That I would not recommend. You can always look up symptoms but got to use medical profession for diagnosis and treatments. On Co-pilot if you asked such questions it comes back not a medical site and cannot give medical guidance or diagnosis. It will then list web sites with information you are asking about.
-
Like -
Helpful -
Hug
3 ReactionsI have not tried Copilot. I have tried Google's AI in web searches, & I find it very unreliable. It makes definitive statements that are not true, & the links it provides do not lead to the conclusions displayed. It consistently makes incorrect assumptions about information from similar products.
-
Like -
Helpful -
Hug
1 Reaction@readandlearn
Co-pilot comes from Microsoft. I had it included when I subscribed to Microsoft 365. I find it very useful and informative. What I like the most is when it cites information at the bottom of the post it lists where the source came from.
What I found one day Co-pilot mentioned Mayo Community Connect (MCC) as a source from feedback of posts on our stie. It made reference to those on MCC citing personal experiences with a subject.
-
Like -
Helpful -
Hug
2 Reactions