New to AI3 min readHealth & Care Navigation

AI is not a doctor: and it is not an emergency service

A practical guide to general health-literacy preparation, personal questions for a pharmacist or clinician, and situations that require emergency care.

What you should be able to do

Use general AI only for bounded health-literacy preparation without real health data. Personal diagnosis, treatment and urgency questions go to the appropriate care service.

Saved only in this browser.
In this article

A general-purpose chatbot is not a medical care relationship. It can explain wording or suggest blank headings, but it has not examined you, established a complete history or arranged accountable follow-up. A confident answer cannot decide a diagnosis, medicine change or whether it is safe to wait.

If a medical emergency may be happening, contact local emergency services now and follow their instructions. Do not describe the situation to a chatbot first. If you need urgent help but do not know the right service, use an official clinical triage route. NHS 111 is an England-specific example; it is distinct from emergency response. In the EU, 112 is the emergency number.

Choose the next source, not a more persuasive prompt

Keep real health details out of a general-purpose chatbot, including when asking about a general term. Use public educational text or blank structures; take your personal circumstances to the appropriate care service.

What you needAppropriate role for AI, if anyWhere the real question goes
A blank appointment structureOptional generic headings, with no real health dataFill it privately and bring it to the appointment
The meaning of a general health termA tentative explanation checked against an authoritative sourceAsk a clinician or pharmacist about personal implications
Interpretation of your symptoms or test resultDo not make the decision in general chatClinician or official clinical triage service
A dose, interaction or treatment decisionDo not act on generated adviceAppropriate pharmacist, prescriber or treating service
Possible emergencyNo chatbot stepLocal emergency service immediately

The medical boundary card keeps these routes together. A disclaimer in an answer does not make the rest of it reliable.

Why the boundary matters

WHO’s guidance on large multimodal models identifies false, inaccurate, biased and incomplete health statements as potential sources of harm. It also warns about automation bias, where people overlook errors or delegate difficult decisions inappropriately.

A model may have access to tools or uploaded records. That does not establish that the records are complete, the tool is validated for your situation or the service has taken responsibility for your care. A provider-approved clinical system has a different scope and data arrangement from an ordinary consumer chat; follow the care provider’s instructions for that exact system.

A source-reading example

For general health literacy, start from a public source. The NHS appointment guide recommends preparing important questions and asking what happens next. A generic prompt could ask:

Turn the idea of asking about follow-up into three blank questions. Do not ask for symptoms, records or medicines, and do not supply medical answers.

AI-generated teaching output, checked for scope:

What should happen next? When should it happen? Who should I contact if the expected follow-up does not arrive?

The questions prepare a conversation. They neither assess urgency nor decide care. You can write them yourself without using AI.

Five question types routed to appointment preparation, authoritative sources, clinical triage, a medicine professional or local emergency services.
Use the appropriate care channel for the actual question. A possible emergency goes directly to local emergency services, without a chatbot step. Open the full-size diagram.

Keep real health information in the care channel

Do not paste real symptoms, dates, lab values, medication lists or medical documents into a general-purpose chatbot. Removing a name does not make a distinctive health history anonymous. Use paper, your private records or a channel explicitly approved by your care provider for the exact data and purpose.

OpenAI’s data controls illustrate that opting out of training does not remove chat history; Temporary Chat has its own retention and sharing boundaries. Neither supplies clinical approval.

For a scheduled visit, use the manual appointment brief. Bring current medicine information; the FDA medication-list guidance explains why prescriptions, non-prescription products, vitamins and supplements belong in that record. If records conflict, ask the pharmacist or care team to reconcile them. Do not ask a model to choose which instruction is current.

When an educational question turns into “What does this mean for me?”, take that question to the appropriate care service. If timing is uncertain, use clinical triage rather than deciding on your own that it can wait.

Read next

Continue through the same learning path with the next practical articles.