Using AI for Health Questions Safely: What to Check

It is late at night, and you have a question about a new prescription or a lab result. You type it into an AI chat tool and get a fast, clear answer. That can be very helpful. But how do you know whether the answer is right, current, and safe for you? This guide gives you a simple checklist for using AI for health questions.

What AI tools can and cannot do

AI chat tools are built on large language models. They are trained on huge amounts of text and learn to produce answers that sound natural. [1]

They can be good at:

• Explaining medical words and test names in plain language.
• Summarizing what a condition is and how it is usually treated.
• Helping you prepare questions for your clinician.
• Pointing you to guidelines and studies, when they cite sources.

They have real limits: [1]

• They can state wrong facts with full confidence. This is sometimes called a hallucination.
• They can invent sources, or cite a real source for a claim it does not make.
• Their knowledge may be out of date.
• They do not know your full history, exam findings, or test results, unless you share them, and even then they cannot examine you.
• Their answers can change depending on how you ask the question.

The World Health Organization has warned that these tools can give inaccurate, incomplete, or biased health information, and it calls for clear oversight and transparency. [1] [2]

What doctors look at

When a clinician judges health information, they ask the same questions you can ask of an AI answer. [3]

Who is behind it?

• Is the source a guideline body, a government health agency, a medical journal, or a hospital?
• For an AI tool, who made it, and what is it designed to do?

What is the evidence?

• Does the answer cite specific guidelines or studies?
• Do the sources exist? Open them and check.
• Does the source actually say what the AI claims?
• Is the evidence from good studies, such as randomized trials or systematic reviews?

Is it current?

• When was the guideline or study published?
• Has the advice changed since then?

Does it fit the person?

• Does the advice apply to your age, sex, pregnancy status, and health conditions?
• Does it match your country? Screening ages, drug approvals, and doses can differ between Canada and the US.

Is it a regulated medical device?

Some AI software is regulated as a medical device, for example tools that help read scans. The FDA keeps a public list of AI-enabled medical devices it has authorized. [4] Most general chat tools are not regulated in this way. They are information tools, not diagnostic devices.

A safety checklist

Before you act on any AI health answer, check:

• Sources: are there citations, and do they check out?
• Dates: is the information recent?
• Country: does it match Canadian or US guidance, as relevant to you?
• Fit: does it apply to someone with your health history?
• Certainty: does the answer say when evidence is limited or experts disagree?
• Next step: does it suggest you confirm with a clinician or pharmacist?

If an answer tells you to start, stop, or change a medicine, do not act on it alone. Talk to your clinician or pharmacist first.

How to ask better questions

The way you ask can change the quality of the answer.

• Ask for sources: "Which guideline or study supports this?"
• Name your country: "What do Canadian guidelines say?" or "What does the FDA label say?"
• Ask about uncertainty: "Where do experts disagree?"
• Ask for numbers: "How many people out of 100 benefit?"
• Ask general questions, not "What do I have?" Use the answer to prepare for your visit, not to replace it.

If two tools, or two answers from the same tool, disagree, treat that as a sign to check with a trusted source or your clinician.

Protect your privacy

• Read the privacy policy before you share health details.
• In the US, the main federal health privacy law covers doctors, hospitals, and health plans. Many consumer apps are not covered by it.
• In Canada, privacy laws vary by province and by sector. Check how the tool stores and uses your data.
• Share as little identifying information as you can, such as your full name, date of birth, or health card number.

Questions to ask your clinician

• I read this answer from an AI tool. Is it correct for me?
• Does this guideline or study apply to someone like me?
• Is this advice current in Canada or the US?
• Are there trusted websites you suggest for my condition?
• Is there anything in this answer that could be unsafe for me?
• Can we go over the sources it cited?

When to get care quickly

Do not use an AI tool in an emergency. Call 911 or go to the emergency department for:

• Chest pain or pressure, or trouble breathing.
• Signs of stroke, such as face drooping, arm weakness, or trouble speaking.
• Severe bleeding, a severe allergic reaction, or fainting.
• Thoughts of harming yourself. In Canada and the US, you can also call or text 988.

Bottom line

AI tools can make health information easier to understand, but they can be wrong, out of date, or not right for you. Check the sources, the dates, and the country. Protect your privacy. Never use AI in an emergency, and talk with your clinician or pharmacist before making any change to your care.

Primary AI can show you the guideline and study sources behind this topic in plain language, so you can bring better questions to your appointment.