AI · health information · safer prompts
AI and health information: how to use AI more safely
Patients increasingly use AI before seeing a doctor. A simple ban is not useful. A safer approach is to show how health-related questions can be asked more carefully.
Available appointments and booking are handled in Mehiläinen’s appointment service.
Start safely
Try a better question first
The purpose of this page is not to make AI a doctor. It is to offer a safer way to use tools that many people already use. Start with one better prompt rather than with settings. This is general information, not personal medical advice.
Do not enter identifying details. Use AI for understanding and preparation – not as a substitute for diagnosis, medication changes, urgent decisions or mental health crisis support. Ask for warning signs first, but remember: their absence from an AI-generated list does not prove the situation is safe.
Emergency 112 · Medical Helpline 116117 · MIELI Crisis Helpline 09 2525 0111 / English 09 2525 0116
Copy this after your own question.
Ask the necessary clarifying questions first. Tell me the warning signs and when to seek medical care or urgent care, and when to call 112 in a life-threatening emergency. Use reliable sources if you make medical claims. Present possible causes as probabilities, not as a certain diagnosis. End by saying what cannot be concluded from this information.
The key idea
You can guide the answer
AI answers can sound too certain, too broad or poorly sourced. You can ask the tool to show uncertainty, use sources, ask clarifying questions first and separate general health information from personal treatment decisions.
Do this
Use AI to structure health information
AI is most useful when it helps organise information before a medical appointment. Its role is not to decide treatment, but to help you ask better questions and understand what needs closer attention.
- prepare a symptom summary for a doctor
- make a list of questions for the appointment
- ask for medical terms to be explained in plain language
- summarise blood pressure or blood glucose measurements for an appointment
- ask for warning signs and sources for medical claims
AI can misread, copy or rearrange measurements. If you use it to summarise a log, check the numbers yourself and bring the original log to the appointment.
Do not do this
Do not use AI to replace urgent care or clinical examination
The worst use is asking AI to reassure you when the situation needs human assessment. If symptoms seem severe or rapidly worsening, the first step is medical help. AI is also not the right support in an acute mental health crisis.
- Diagnose me.
- Can I stop this medication?
- Is this definitely harmless?
- Reassure me that I do not need a doctor.
Privacy
Do not paste everything into AI tools
AI tools are not confidential healthcare systems. Do not paste personal identity codes, full names, exact addresses, other people’s information, prescriptions, medical records, laboratory reports or images as they are.
Remove identifiers first and describe the situation at a general level: age, sex if relevant, main symptom, duration, underlying conditions, medication and possible alarming symptoms. For measurements, prefer typing the values as text instead of uploading a photo: it protects privacy better and reduces reading errors.
Most AI services have settings that can limit whether conversations are used for model training. Some also offer temporary chats. Check the current settings in the service you use, because they change over time.
Frequent users
Save a permanent instruction if you use AI often
A saved instruction can make answers more cautious, better sourced and clearer about uncertainty without rewriting the same request every time.
The same principle works in other services, such as Gemini, Claude and Copilot. Look for custom instructions or a similar persistent instruction setting.
- Open Settings.
- Select Personalization.
- Make sure customization is enabled.
- Paste the instruction into the Custom Instructions field.
- Open Settings.
- Select Customize ChatGPT.
- Make sure customization is enabled.
- Paste the instruction into the Custom Instructions field.
Checked 7/2026. Menu names may vary depending on app language and version. Do not save personal identity codes, detailed medical records or other identifying information in custom instructions.
In health questions, act as a cautious, critical and honest assistant. Ask clarifying questions if the information is insufficient. Show uncertainty, warning signs, sources and what cannot be concluded from the information given. Do not treat this AI tool as a medical record system.
Answer health questions cautiously and with sources when medical claims are made. Ask clarifying questions if the information is insufficient. Always mention warning signs and when to seek medical care or urgent care, and when to call 112 in a life-threatening emergency. Separate general health information from personal treatment decisions. Present possible causes as probabilities, not as a certain diagnosis. Do not advise changing medication without a doctor’s assessment. Do not invent sources, studies or links. State clearly when you are uncertain and what cannot be concluded from the information given.
Emergency
In an emergency, do not ask AI
If symptoms may be life-threatening, call 112. In Finland, for urgent but non-life-threatening medical advice, contact Medical Helpline 116117. In a mental health crisis, MIELI Crisis Helpline is 09 2525 0111 in Finnish and 09 2525 0116 in English. Physical warning signs include severe chest pain, severe shortness of breath, stroke symptoms, confusion, impaired consciousness, heavy bleeding, a severe allergic reaction or rapidly worsening general condition.
Questions
FAQ
Yes, they can make answers more cautious, more transparent about uncertainty and more likely to cite sources. They do not make answers error-free.
No. AI can still be wrong. The purpose is to make unsupported certainty easier to notice.
Prefer typing the values as text. A photo can contain identifying details you did not notice, and AI can misread handwriting or numbers. Bring the original log to the appointment.
Sources and background
These sources do not replace individual medical assessment. They describe the starting point of this page: AI can help organise information, but safe use requires caution, privacy protection and human assessment when symptoms or context require it.
- WHO: Ethics and governance of artificial intelligence for health: large multi-modal models
- WHO: Ethics and governance of artificial intelligence for health
- Finnish Data Protection Ombudsman: AI systems and data protection
- Finnish National Cyber Security Centre: Be cautious with AI
- OpenAI: ChatGPT Custom Instructions
- OpenAI: Data Controls FAQ
- MIELI Mental Health Finland: Contact information
Related pages