What can an AI assistant safely tell you about your health?

Health questions are among the most common things people ask AI assistants, and for good reason: the answers come instantly, in plain words, at two in the morning. The same qualities make it easy to forget what an assistant is and isn't. A general chatbot is a language model that predicts helpful text. It has not examined you, it may not know your history, and nobody has certified it as a medical tool.

The useful question is not "should I ever ask?" but "what kind of help is safe to take from it?" A simple rule works well: use AI to understand and prepare, and leave deciding and treating to a professional.

Where an assistant helps

• Translating jargon. "What does 'benign essential tremor' mean?" or "What is an HbA1c test measuring?" are questions a good model answers well.
• Making sense of a document you already have. It can explain the sections of a lab report or a discharge summary in general terms, and tell you what each value usually measures.
• Preparing for an appointment. It can turn a jumble of symptoms into a timeline and suggest questions to ask.
• Understanding instructions. It can restate a care plan in simpler words or make a medication schedule into a checklist, as long as you check it against the original.
• Healthy habits. Sleep, exercise, stress and diet routines are general wellness topics where a chatbot's suggestions carry low risk.

Where to stop

• Diagnosis. A list of possible causes is not a diagnosis. Models can sound certain and still be wrong, and they cannot examine you.
• Medication changes. Never start, stop or change a dose based on a chatbot. Ask your pharmacist, who can see your full medication list.
• Urgent symptoms. Chest pain, trouble breathing, signs of stroke, thoughts of self-harm: call emergency services or a health line, not an app.
• Private details. Before you paste a lab report, decide whether you want it stored. Remove your name, health card number and address unless you are using a service designed for health records.

A worked example: preparing for a doctor's visit

Here is a prompt that keeps the assistant in the "understand and prepare" lane:

I have an appointment with my family doctor next week. For three weeks I have had
tiredness, headaches in the afternoon and more thirst than usual. I am 52 and take
medication for blood pressure.
Do not diagnose me. Instead:
1. Put my symptoms in a short timeline I can read aloud.
2. List five questions I could ask my doctor.
3. Tell me which of these symptoms, if they got worse, would mean I should seek care sooner.
Keep it under 200 words.

The answer gives you a better conversation with your doctor, not a substitute for it. If the reply lists warning signs that apply to you now, act on them rather than waiting for the appointment.

How are health AI tools regulated in Canada and the US?

In both countries, the deciding factor is intended use: what the maker says the software is for. Software meant for a medical purpose, such as diagnosing or treating a disease, is regulated as a medical device. Software for general wellness or information is not.

In the United States, the Food and Drug Administration (FDA) updated its general wellness guidance on January 6, 2026. Software meant only to support a healthy lifestyle, such as fitness, sleep or stress apps, is not a device, and the FDA does not intend to review other low-risk wellness products. Wearables that estimate things like blood pressure or oxygen levels can fall into this category only if their readings are not meant for diagnosis or treatment. The guidance adds a caution worth remembering: being treated as a wellness product does not mean a product has been shown to be safe or effective. FDA guidance documents are recommendations rather than binding rules.

In Canada, Health Canada treats software as a medical device when it is intended for a medical purpose. Software that only supports a decision by a professional, patient or caregiver can be excluded, but only if it meets all four of the department's criteria. For AI tools that are devices, Health Canada published updated pre-market guidance on April 1, 2026, for machine learning-enabled medical devices. Makers applying for a licence for Class II, III or IV devices must show good practice in data, training, testing, clinical validation, transparency and monitoring after release. The guidance also introduces a predetermined change control plan, a way to approve in advance how a model may be updated without a new review each time.

What this means for you: a general-purpose chatbot is not marketed for a medical purpose, so it has not passed the review a licensed diagnostic tool must pass. A health app that claims to detect or diagnose a condition should be able to tell you whether it holds a Health Canada licence or FDA authorization. If it can't, treat its output as information to discuss with a professional, never as a verdict.