ONLINEAGENT_OPS 2026.Q3 HOME HUBSMODELSBUILDINGCOMPANIESFILMMAKINGSECURITYLIVING WITH ITSKILLSALL HUBS, ONE TERMINAL → ARTICLES CRAFT RECORD BLOG TOOLS SKILLS SEARCH /
HOME›ARTICLES›ASKING AI ABOUT YOUR HEALTH: HOW TO DO IT SAFELY
ARTICLES · GUIDEBEGINNER

Asking AI About Your Health: How to Do It Safely

How to ask AI chatbots health questions safely: what they help with, what studies found they get wrong, privacy, and when to see a professional instead.

READ12 min
WORDS2,384
SECTIONS7
TYPEGUIDE
CHECKED12 OCT 26

You can ask an AI chatbot about your health, as long as you use it to understand and to prepare, not to decide. It is good at explaining medical words, making sense of a letter and helping you write questions for your doctor.

It is not reliable at working out what is wrong with you or how urgent it is, and what you type is often not protected the way your medical records are. If something might be an emergency, skip the chatbot and call for help.

◈ EMERGENCIES FIRST

If you think someone’s life could be at risk, do not ask a chatbot. Call your local emergency number. In the US that is 911: “In an emergency, dial 911 from your phone immediately.” 1 In the UK it is 999, and NHS 111 helps if you are not sure. 2 Across the EU, 112 works everywhere. 3 Elsewhere, use your country’s number.

TL;DR — THE SHORT VERSION
  • Good for: explaining terms, summarising a letter in plain words, and drafting questions for your appointment.
  • Not good for: diagnosing you or judging urgency. In one large study, people using chatbots did no better than people using their usual sources.
  • Privacy: in the US, health privacy law usually does not cover what you type into a chatbot you signed up for yourself.
  • Check it: compare the answer with a trusted health website and with a professional before you act on it.
  • Emergencies: call your local emergency number, not a chatbot.

What can AI help with on health questions?

The plain point: a chatbot is a patient explainer. Use it to understand what you have been told and to get ready for a conversation with a professional.

The comparison: think of it as a well-read friend who helps you prepare for an appointment, not the doctor in the room.

A chatbot is a program that answers in everyday words. The World Health Organization lists “patient-guided use”, such as looking into symptoms and treatment, as one of the ways these tools are already used in health. 4 Companies now build features for this. OpenAI describes its ChatGPT Health space as a way to “understand recent test results, prepare for appointments with your doctor”, and says it “is not intended for diagnosis or treatment.” 5

Uses that fit that line:

  • Explain a word. “What does ‘benign’ mean on a scan report?”
  • Translate a letter into plain English, with names and ID numbers removed first.
  • Prepare questions for a GP, a specialist or a pharmacist.
  • Understand a condition you have already been diagnosed with, then check what it says against a trusted source.

This matches the site’s rule in when not to use AI: prepare with it, do not decide with it.

UNDERSTAND A LETTER — ADAPT THE BRACKETS
Explain this medical letter in plain English for someone with no medical training: [paste the letter with my name, address, dates of birth and ID numbers removed]. List any terms you explain, and end with five questions I could ask my doctor about it. Do not guess at a diagnosis or tell me what treatment to choose.
PREPARE FOR AN APPOINTMENT — ADAPT THE BRACKETS
I have a [10]-minute appointment about [ongoing knee pain]. Help me write a short summary of what I have noticed, using my notes: [when it started, what makes it better or worse]. Then suggest the most useful questions to ask. Keep it to one page.
TAKEAWAY

Ask it to explain and to help you prepare. Bring what it gives you to a professional, rather than acting on it alone.

How accurate is AI about health?

The plain point: chatbots know a lot of medicine, but that knowledge often gets lost in a real conversation with a worried person.

The comparison: it is like a student who aces the written exam but has never met a patient. Knowing the textbook is not the same as asking the right questions.

One large test with real people comes from the University of Oxford. Nearly 1,300 people were given doctor-written scenarios and asked to work out the likely condition and what to do next. On their own, the chatbots named the right condition in 94.9% of cases. When people used the same chatbots, they named a relevant condition in fewer than 34.5% of cases and picked the right next step in fewer than 44.2%, no better than people using their usual sources. 6

The researchers put this down to the conversation itself. People did not know what details to give, and the answers mixed good and poor advice. Small changes in wording also gave very different answers. 7 The GP who was lead medical practitioner on the study warned that asking a chatbot about symptoms can mean “failing to recognise when urgent help is needed.” 7

A wrong detail in your question can spread. Researchers at Mount Sinai wrote 300 medical cases, each with one made-up detail, such as a test or a condition that does not exist. Across six chatbots, the AI went along with the invented detail in 50% to 82% of answers. A warning in the prompt helped but did not fix it. 8 In practice: if you misremember a medicine’s name, the chatbot may explain the wrong medicine confidently.

Patient-safety bodies say similar things. ECRI, a US patient-safety organisation, put the misuse of AI chatbots first on its list of health technology hazards for 2026, noting they “are not designed or regulated for healthcare purposes.” 9 The WHO points to “documented risks of producing false, inaccurate, biased, or incomplete statements”. 4

◈ IN PLAIN TERMS

A chatbot predicts likely-sounding words. It does not examine you, and it can state a wrong answer as calmly as a right one. That is called a hallucination, and why AI makes things up explains where it comes from. In health, a calm wrong answer is the dangerous kind.

TAKEAWAY

A chatbot can know the medicine and still get your case wrong. Treat any answer about what is wrong, or how urgent it is, as a guess to check.

Is what I tell a chatbot about my health private?

The plain point: often not in the way your medical records are. The rules depend on where you live and who runs the tool.

The comparison: telling your doctor is like a sealed letter with legal rules around it. Typing into a general chatbot is more like telling a company’s customer service team.

In the US, the health privacy law HIPAA covers health plans, most health care providers and the companies working for them. The US health department says it generally does not cover information you share online or enter into apps for your own use, unless a covered provider gave you the app. 10 The same page warns that data collected by apps “may also be sold to a data broker”. 11

In the UK, data protection law treats health information as a “special category” that needs extra care, because misuse “could create significant risks” to people’s rights. 12 That is a duty on the company. It does not stop the company storing what you type, so the habits below still apply. Reasoning, 5 Oct 2026 — a legal duty to handle data carefully is not the same as a promise not to keep it; check each app’s own privacy policy. Other countries have their own rules; this page did not check them.

Company promises vary. Some tools now keep health chats in a separate space. OpenAI says “Conversations in Health are not used to train our foundation models.” 5 That is the company’s own policy, which it can change, not a health privacy law.

  • Leave out what identifies you: name, address, date of birth, patient or insurance numbers.
  • Check the training setting, and whether the app remembers you between chats. AI memory explains why a new chat is not always a blank page.
  • Ask before sharing someone else’s health details, such as a parent’s or a child’s.

What happens to what you type covers storage, training and who can read it, one question at a time.

TAKEAWAY

Assume a general chatbot is not covered by health privacy law. Share the medical question, not the details that say who you are.

How do I check a health answer from AI?

The plain point: check the answer against a trusted health source, then talk it through with a professional before acting.

The comparison: treat the chatbot’s answer like a tip from a stranger on a bus. It might be right, but you would confirm it before changing your medicine.

MedlinePlus, run by the US National Library of Medicine, gives two steps for any health information: question it, then “Discuss the information you find with your health care provider before you rely on it.” 13 Applied to a chatbot:

  1. Ask where it came from. Request sources, then open each one. A chatbot can invent references that look real. How to check an AI citation takes about a minute.
  2. Cross-check with a health body. Look the topic up on your national health service’s website or a government health site, not a forum.
  3. Ask the same question a second way. The Oxford team found small changes in wording gave very different answers, so a change in the answer is a warning sign.
  4. Watch for “too good to be true”. MedlinePlus warns to “beware of dramatic writing, promises of cures, and claims that sound too good to be true.” 14
  5. Bring it to a professional. A pharmacist, nurse or doctor can tell you whether a general answer fits your case.

Check your own question too. Use the exact name from the box or the letter, as a misspelled name can be answered as if it were a different thing.

TAKEAWAY

Open the sources, check a health body’s page, and ask a professional before you change anything you do.

When should I skip the chatbot and see a professional?

The plain point: whenever the question is “what should I do now?” rather than “what does this mean?”

The comparison: a map helps you understand the area, but if the car is smoking you call for help rather than reading the map.

  • Possible emergencies. Call your local emergency number (see the box at the top).
  • Not sure how urgent it is. Use your country’s non-emergency health line, such as NHS 111 in England, or call your doctor.
  • Starting, stopping or changing a medicine. Ask a pharmacist or the prescriber.
  • New, worsening or worrying symptoms. The Oxford study is the warning here: chatbots helped people no more than their usual sources did.
  • Anything about your mental health that feels heavy. Read AI chatbots and mental health, which lists crisis lines.

The WHO also warns about “automation bias”: trusting the machine so much that errors slip past or hard choices get handed over. 4 A health decision should stay with you and the people treating you.

TAKEAWAY

“What does this mean?” is a fine question for a chatbot. “What should I do now?” belongs to a professional.

Questions people ask

Can I trust ChatGPT for medical advice?

Not to diagnose you or tell you how urgent something is. In an Oxford study of nearly 1,300 people, those using chatbots identified a relevant condition in fewer than 34.5% of cases, no better than people using their usual sources; use it to understand and prepare instead. 15

Is a chatbot covered by HIPAA?

Usually not, if you signed up for it yourself. The US health department says HIPAA generally does not protect data you enter into apps for personal use unless a covered provider, such as your doctor or health plan, gave you the app. 11

Can AI explain my test results?

It can explain what the terms and ranges usually mean, and some tools are built for this. OpenAI still says its health feature “is not intended for diagnosis or treatment”, so ask your doctor what the results mean for you. 5

What should I do if the AI says it could be serious?

Do not wait for the chatbot to confirm it. If it might be life-threatening, call your local emergency number; if you are unsure, call a non-emergency health line or your doctor, as NHS 111 does in England. 16

How do I ask AI about health safely?

Remove names and ID numbers, ask it to explain rather than diagnose, ask for sources and open them, and take the answer to a professional. MedlinePlus advises discussing any health information with your provider before relying on it. 14

Where to go next

QUICK CHECK · 3 QUESTIONS

Pick an answer to see if you got it. Nothing is recorded.

1. Which kind of health question does the page say a chatbot is fine for?

  • What does this mean? Such as explaining a word on a letter
  • What should I do now? Such as whether to change a medicine
  • How urgent is this? Such as whether to go to hospital
Show the answer

Answer: What does this mean? Such as explaining a word on a letter. The page says use a chatbot to understand and prepare; questions about what to do now belong to a professional.

2. What should you do if you think someone’s life could be at risk?

  • Describe the symptoms to a chatbot first
  • Wait to see whether a chatbot says it is serious
  • Call your local emergency number, not a chatbot
Show the answer

Answer: Call your local emergency number, not a chatbot. The page puts emergencies first: if a life could be at risk, skip the chatbot and call for help.

3. Is what you type into a chatbot about your health protected like your medical records?

  • Yes, health privacy law always covers it
  • Often not, so leave out your name, date of birth and ID numbers
  • Yes, as long as you delete the chat afterwards
Show the answer

Answer: Often not, so leave out your name, date of birth and ID numbers. The page says health privacy law usually does not cover a chatbot you signed up for yourself, so share the question, not who you are.

SOURCES
  1. 1911.gov, Calling 911, last updated 24 Jan 2023, read at source 5 Oct 2026“In an emergency, dial 911 from your phone immediately.” ↑
  2. 2NHS, When to call 999, page last reviewed 3 Feb 2023, read at source 5 Oct 2026“999 is for life-threatening emergencies like serious road traffic accidents, strokes and heart attacks.” and “NHS 111 can check your symptoms and tell you what to do.” ↑
  3. 3European Union, Single emergency number, last updated 22 Jun 2026, read at source 5 Oct 2026“112 is the European emergency number you can dial free of charge from fixed and mobile phones everywhere in the EU.” ↑
  4. 4World Health Organization, WHO releases AI ethics and governance guidance for large multi-modal models, 18 Jan 2024, read at source 5 Oct 2026“Patient-guided use, such as for investigating symptoms and treatment;” ↑a“there are also documented risks of producing false, inaccurate, biased, or incomplete statements, which could harm people using such information in making health decisions.” ↑b“LMMS can also encourage ‘automation bias’ by health care professionals and patients, whereby errors are overlooked that would otherwise have been identified or difficult choices are improperly delegated to a LMM.” ↑c
  5. 5OpenAI, Introducing ChatGPT Health, 7 Jan 2026, read at source 5 Oct 2026“so ChatGPT can help you understand recent test results, prepare for appointments with your doctor” and “Health is designed to support, not replace, medical care. It is not intended for diagnosis or treatment.” ↑a“Conversations in Health are not used to train our foundation models.” ↑b“It is not intended for diagnosis or treatment.” ↑c
  6. 6Bean et al., Nature Medicine, Reliability of LLMs as medical assistants for the general public: a randomized preregistered study, published 9 Feb 2026, read at source 5 Oct 2026“in a controlled study with 1,298 participants” and “Tested alone, LLMs complete the scenarios accurately, correctly identifying conditions in 94.9% of cases and disposition in 56.3% on average. However, participants using the same LLMs identified relevant conditions in fewer than 34.5% of cases and disposition in fewer than 44.2%, both no better than the control group.” ↑
  7. 7Oxford Internet Institute, New study warns of risks in AI chatbots giving medical advice, 9 Feb 2026, read at source 5 Oct 2026“Participants often didn’t know what information the LLMs needed to offer accurate advice, and the responses they received frequently combined good and poor recommendations” and “LLMs provided very different answers based on slight variations in the questions asked”. ↑a“Patients need to be aware that asking a large language model about their symptoms can be dangerous, giving wrong diagnoses and failing to recognise when urgent help is needed.” ↑b
  8. 8Omar et al., Communications Medicine, Multi-model assurance analysis showing large language models are highly vulnerable to adversarial hallucination attacks during clinical decision support, published 2 Aug 2025, read at source 5 Oct 2026“We created 300 physician-validated simulated vignettes, each containing one fabricated detail (a laboratory test, a physical or radiological sign, or a medical condition).” and “Hallucination rates range from 50 % to 82 % across models and prompting methods.” and “While prompt engineering reduces errors, it does not eliminate them.” ↑
  9. 9ECRI, Examining the Greatest Health Technology Threats of 2026, 26 Jan 2026, read at source 5 Oct 2026“AI chatbots and other large language models (LLMs) are not designed or regulated for healthcare purposes.” ↑
  10. 10US Department of Health and Human Services, Protecting the Privacy and Security of Your Health Information When Using Your Personal Cell Phone or Tablet, content last reviewed 29 Jun 2022, read at source 5 Oct 2026“the HIPAA Rules do not protect the privacy of your Internet search history, information you voluntarily share online, or your geographic location information. In most cases, unless the app is provided to you by a covered entity or its business associate, the HIPAA Rules also do not protect the privacy of data you’ve downloaded or entered into mobile apps for your personal use, regardless of where the information came from.” ↑
  11. 11US Department of Health and Human Services, Protecting the Privacy and Security of Your Health Information When Using Your Personal Cell Phone or Tablet, read at source 5 Oct 2026“It may also be sold to a data broker, someone who obtains and shares consumer information without their knowledge, often selling it for marketing or other purposes.” ↑a“In most cases, unless the app is provided to you by a covered entity or its business associate, the HIPAA Rules also do not protect the privacy of data you’ve downloaded or entered into mobile apps for your personal use”. ↑b
  12. 12Information Commissioner’s Office, What is special category data?, read at source 5 Oct 2026“This is because use of this data could create significant risks to the individual’s fundamental rights and freedoms.” ↑
  13. 13MedlinePlus, US National Library of Medicine, Evaluating Health Information, read at source 5 Oct 2026“Discuss the information you find with your health care provider before you rely on it.” and “Where does the information come from? The content pages should have links or references to the sources of the information.” ↑
  14. 14MedlinePlus, Evaluating Health Information, read at source 5 Oct 2026“So, beware of dramatic writing, promises of cures, and claims that sound too good to be true.” ↑a“Discuss the information you find with your health care provider before you rely on it.” ↑b
  15. 15Bean et al., Nature Medicine, Reliability of LLMs as medical assistants for the general public, 9 Feb 2026, read at source 5 Oct 2026“participants using the same LLMs identified relevant conditions in fewer than 34.5% of cases and disposition in fewer than 44.2%, both no better than the control group.” ↑
  16. 16NHS, When to call 999, read at source 5 Oct 2026“NHS 111 can check your symptoms and tell you what to do.” ↑

911.gov, Calling 911, last updated 24 Jan 2023, read at source 5 Oct 2026.

Bean et al., Nature Medicine, Reliability of LLMs as medical assistants for the general public: a randomized preregistered study, 9 Feb 2026, read at source 5 Oct 2026. A publisher correction fixed a chart axis label only.

Omar et al., Communications Medicine, Multi-model assurance analysis showing large language models are highly vulnerable to adversarial hallucination attacks during clinical decision support, 2 Aug 2025, read at source 5 Oct 2026.

This page explains how to use AI tools around health. It is not medical advice. The comparisons on this page (the well-read friend, the student who aces the exam, the sealed letter, the tip from a stranger, the map and the smoking car) are this site’s own ways of explaining, not claims from these sources.

Free printable AI cheat sheets, and AI news every week.

Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.

SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK
aipromptgeneer://$ about this siteAI PROMPT GENEERAI explained in plain English. Every claim links to its source, with the date it was checked.NEWSLETTERWhat changed in AI each week, in plain English.SUBSCRIBE FREE ↗AI PROMPT GENEER · MACHINE ASSISTANCE DISCLOSEDEVERY ARTICLE CARRIES ITS OWN CHECKED DATE