Using an AI chatbot for a health question? How to check the answer before you act on it

The short answer

Treat a chatbot's health answer as a first draft, not a decision. Before you act on it:

1. Skip the chatbot in an emergency, a crisis or a possible poisoning. Use the numbers in the box below.
2. Open the sources yourself. Ask where the answer comes from, then click each link and make sure it exists and says what the chatbot claims.
3. Compare the key fact with a trusted reference, such as MedlinePlus, the CDC, the NIH or the medicine's own label.
4. Check the date and the fit. Health guidance changes, and a general answer may not fit your age, a pregnancy, your other conditions or your other medicines.
5. Ask a pharmacist or clinician before you start, stop or change any medicine or dose, and keep your name, birth date and record numbers out of the chat.

Why a confident answer can still be wrong

AI chatbots write by predicting likely words. That makes them fluent, and fluency can feel like expertise. Three problems come up again and again.

• False content stated with confidence. The National Institute of Standards and Technology (NIST), a federal science agency, calls this "confabulation." It notes that answers can even include made-up reasoning or citations that seem to back them up.
• Old information. Guidance changes. In April 2024, for example, the U.S. Preventive Services Task Force began recommending a mammogram every two years for women ages 40 to 74; its 2016 version had left screening in the 40s as an individual decision. An answer built on older material can miss a change like that.
• No view of you. A chatbot cannot examine you, read your chart or know your other medicines unless you type them in, and typing them in has privacy costs (see below).

The World Health Organization's January 2024 guidance on the technology behind many chatbots warns of "false, inaccurate, biased, or incomplete statements" and of "automation bias," trusting a machine's output too readily. The Federal Trade Commission tells consumers not to rely only on a chatbot for medical advice.

None of this makes chatbots useless. In a KFF poll fielded in February and March 2026, about one in three U.S. adults said they had used an AI chatbot for health information in the past year. Explaining a medical word or preparing questions for a visit are reasonable uses. The answer is a starting place.

Step 1: Sort the question before you ask it

Step 2: Ask in a way that makes checking easier

• Ask it to name its sources and give working links, ideally to government health agencies or professional medical organizations.
• Ask it to say how certain it is, and what a clinician would need to know about you to be sure.
• Ask for questions to bring to your doctor or pharmacist.
• Describe yourself in general terms, such as "an adult in their 60s who takes a blood pressure medicine," not with your name, birth date or a copy of your test report.

One more habit: ask the same question again in different words. If the two answers disagree on something important, treat both as unchecked until you look it up.

Step 3: Open every source it gives you

This is the step that catches invented citations.

• Does the link open? A dead link or an error page is a warning sign.
• Is the page really from the organization the chatbot named, on its own website?
• Does the page actually say what the chatbot said, for the same group of people (adults or children, pregnant or not)?
• For a research study, can you find it by its title on PubMed, the National Library of Medicine's free index of medical research? If you cannot find it anywhere, assume the citation may be invented.
• Does the page show when it was written, reviewed or updated?

If a chatbot cannot give sources, or its sources do not hold up, its answer has not been checked, however detailed it sounds. MedlinePlus, the National Library of Medicine's consumer health site, suggests a simple test for any health information: see whether other reliable sources say the same thing, and talk it over with your health care provider.

Step 4: Compare it with a trusted reference

Free public sources cover most everyday questions.

When the chatbot and the reference disagree, go with the reference, and take the disagreement to a pharmacist or clinician.

Step 5: Check the date and the fit

• Look for the date. MedlinePlus, CDC and FDA pages show when they were last updated or reviewed. Older is not always wrong, but on a fast-changing topic a recent date matters.
• Check who the advice is for. General answers often assume a healthy adult. Ask whether it changes for children, older adults, pregnancy, or kidney or liver disease.
• Check your whole medicine list. If you take several medicines, ask a pharmacist to check for interactions.

Warning signs in a chatbot's health answer

• Exact doses, percentages or statistics with no source.
• Words like "always," "never," "guaranteed" or "completely safe."
• Advice that conflicts with your prescription label or with what your clinician told you.
• No questions or caveats about age, pregnancy, allergies or other medicines when those clearly matter.
• A push toward a particular product, supplement or clinic.
• A tool that calls itself, or implies it is, a doctor, nurse or therapist. In California, a 2025 law (AB 489), in effect since January 1, 2026, bars companies that build or run AI systems from using titles or wording, in their ads or in the tool itself, that suggest the advice comes from a licensed health professional.

Protect your privacy before you type

HIPAA, the federal health privacy law, applies to health plans, most health care providers and the businesses that work for them. According to the U.S. Department of Health and Human Services, HIPAA generally does not protect health information you put into apps on your own phone or tablet, unless the app was provided by your health care provider, health plan or one of their contractors. The same logic applies to a chatbot you choose to use on your own.

Before you share anything:

• Leave out your name, birth date, address, phone number, and insurance or Medicare numbers.
• Do not upload a whole test report or visit note with identifying details. To understand a result, type just the test name, the number and the units.
• Check the tool's settings and privacy policy. Are chats saved? Can people at the company read them? Are they used to train the system? Can you delete them?
• Take extra care with sensitive topics: sexual and reproductive health, mental health, substance use, and anything about a child.

In the same 2026 KFF poll, 41% of people who used AI for health said they had uploaded personal medical information such as test results or doctors' notes. If you are thinking of connecting your medical records to an app, see Sharing your health records with an app: what to check first.

In California: A law signed on September 30, 2026 (AB 1979) will treat a business that offers consumers a "health care chatbot," an AI chat system marketed to support their health care, as a health care provider under the state's Confidentiality of Medical Information Act, so the confidentiality duties that bind doctors' offices would bind it too. The law states no start date of its own, so under California's standard rule it is expected to take effect January 1, 2027. It may not reach general-purpose chatbots that are not marketed for health. For the policy detail, see California's Health-AI Experiment Is Now Law.

When your own clinic uses AI to write to you (California)

Since January 1, 2025, California hospitals and other licensed health facilities, clinics, physician offices and group practices that use generative AI to create written or spoken messages to patients about their clinical information must say so in the message. They must also tell you how to reach a human health care provider or staff member. The notice goes at the top of letters and emails, stays visible throughout chats and video, and is spoken at the start and end of audio messages. The rule does not apply when a licensed or certified health care provider has read and reviewed the message, and it does not cover routine business messages such as scheduling and billing. The requirement comes from AB 3030, now section 1339.75 of the Health and Safety Code.

If a message from your clinic carries that notice, use the same checks, and use the contact instructions to ask a person about anything that matters to your care. The Healthcare AI governance dossier follows California's newer health-AI laws as they take effect.

Take it to a person: what to bring and what to ask

Bring:

• A screenshot or printout of the chatbot's answer, including the sources it gave.
• Your current medicine list, including over-the-counter products and supplements.

Ask:

• "A chatbot told me this. Is it right for me?"
• "Is this safe with the other medicines I take?"
• "Which source would you trust on this?"
• "What symptoms would mean I should call you, or go to urgent care or the emergency room?"

At many pharmacies, you can ask a pharmacist about a medicine without an appointment. For more on judging health information in general, see How to Evaluate Online Medical Information and How to Spot and Avoid Health Misinformation. For nonprescription medicines, see Safe Use of Over-the-Counter Medicines.

Sources

1. Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI 600-1) — National Institute of Standards and Technology, July 2024.
2. WHO releases AI ethics and governance guidance for large multi-modal models — World Health Organization, January 18, 2024.
3. Operation AI Comply: Detecting AI-infused frauds and deceptions — Federal Trade Commission consumer alert, September 2024.
4. Poll: 1 in 3 adults are turning to AI chatbots for health information — KFF, March 25, 2026 (fielded February 24 to March 2, 2026).
5. Breast Cancer: Screening — U.S. Preventive Services Task Force, final recommendation April 30, 2024.
6. Recognizing medical emergencies — MedlinePlus, National Library of Medicine, reviewed January 8, 2025.
7. 988 Suicide & Crisis Lifeline — 988 Lifeline, accessed October 20, 2026.
8. Poison Help — Health Resources and Services Administration, accessed October 20, 2026.
9. Evaluating Health Information — MedlinePlus, National Library of Medicine, updated February 26, 2024.
10. Pseudoephedrine — MedlinePlus, National Library of Medicine, revised June 15, 2025.
11. Drugs, Herbs and Supplements — MedlinePlus, National Library of Medicine, accessed October 20, 2026.
12. About DailyMed — National Library of Medicine, accessed October 20, 2026.
13. Drug Safety Communications — U.S. Food and Drug Administration, content current as of September 15, 2026.
14. What Vaccines are Recommended for You — Centers for Disease Control and Prevention, reviewed June 12, 2024.
15. Dietary Supplement Fact Sheets — NIH Office of Dietary Supplements, accessed October 20, 2026.
16. About PubMed — National Center for Biotechnology Information, National Library of Medicine, accessed October 20, 2026.
17. Protecting the Privacy and Security of Your Health Information When Using Your Personal Cell Phone or Tablet — U.S. Department of Health and Human Services, Office for Civil Rights, reviewed June 29, 2022.
18. Your Rights Under HIPAA: guidance materials for consumers — U.S. Department of Health and Human Services, reviewed May 30, 2025.
19. AB 1979, Health care services: artificial intelligence, chaptered text (Chapter 854, Statutes of 2026) — California Legislature, via LegiScan, chaptered September 30, 2026.
20. California Constitution, Article IV, Section 8 — reproduced by Justia, accessed October 20, 2026.
21. AB 489, Health care professions: deceptive terms or letters: artificial intelligence, chaptered text (Chapter 615, Statutes of 2025) — California Legislature, via LegiScan, chaptered October 11, 2025.
22. AB 489 advisory — California Board of Psychology, accessed October 20, 2026.
23. All Facilities Letter 25-07: Assembly Bill (AB) 3030 - Health Care Services: Artificial Intelligence — California Department of Public Health, February 24, 2025.
24. AB 3030 (2024), Health care services: artificial intelligence — CalMatters Digital Democracy bill record (Chapter 848, Statutes of 2024), accessed October 20, 2026.