A fifteen-minute consultation, and the questions come back once the door has closed. From memory, on a forum, with a consumer AI tool or a confidential AI: the comparison, and three requests to arrive prepared.
"I'm seeing the cardiologist on Thursday, so last night I pasted my results into an AI tool to work out what to ask him." You hear it by the coffee machine, said without a second thought, between two meetings. The idea is a good one: a consultation often lasts a quarter of an hour, and the important questions come back once the door has closed. What is less good is what went along with the results: a name, a date of birth, an insurance number, and everything a set of tests says about someone's health.
Preparing for a doctor's appointment with AI is one of the most useful things you can do with it. You still have to choose where you write, and what you write.
In the consulting room, what you say is protected: your doctor is bound by medical confidentiality, in Switzerland as in most countries, and the Swiss Criminal Code (Art. 321) makes breaching it an offence. That confidentiality stops at the door. What you type yourself into an online service is covered by no secrecy at all: you are the one handing it to a company, on that company's terms.
Health data is among the most sensitive data there is. The GDPR and the Swiss FADP both put it in a category of its own, precisely because it can do harm when it circulates: with an insurer, an employer, a relative. A consumer AI tool often keeps the conversation history, sometimes for a long time, and many are run by companies subject to US law, where the CLOUD Act allows the authorities to demand access to the data they hold.
Nobody is stopping you from trying to understand your health. The only question is: who keeps a record of what you looked for?
There are four ways to prepare for a consultation. They do not produce the same questions, and they do not expose the same things.
| From memory or on paper | Search engine and forums | Consumer AI tool | Confidential AI | |
|---|---|---|---|---|
| What leaves your home | Nothing | Your searches, tied to your browsing | Everything you paste: results, symptoms, sometimes your identity | Nothing to a third party with the confidential models |
| What is kept, and where | Your sheet of paper, in your pocket | Your search history, with the search engine; your posts, on the forum | The conversation history, with the provider | The history in your browser, nothing on our side |
| Who decides | You | The search engine's and the forum's terms | The provider's terms | You, by choosing the model and what you write |
| Quality of the questions | Whatever comes to mind, often not the most useful | Uneven: the worst case surfaces fast, and the anxiety with it | Good, if the request is precise | Good, if the request is precise |
| Effort | Low, but you forget | High: hours of contradictory reading | Low | Low, after two minutes of sorting |
| Main risk | Leaving without asking the question that mattered | Convincing yourself you have an illness you don't have | Leaving your health data with a third party for the long term | A wrong or incomplete answer, as with any AI |
Paper is still your best ally on the day itself. However you prepare, finish with a short written list that you take out at the start of the consultation: a doctor would rather hear three questions up front than a fourth one asked on the way out.
Search engines and forums are useful for general information you can check: what a colonoscopy is, how to prepare for a fasting blood test. Avoid describing your own case there: a forum post is public, and often for a long time.
A consumer AI tool is fine for questions that say nothing about you: "What should I ask before carpal tunnel surgery?" exposes no one. As soon as you paste your results, your medication or your medical history, you are in a different category. Before you do, read what happens to your conversations.
A confidential AI is the right choice as soon as your own situation is part of the request: test results, a timeline of symptoms, a treatment that isn't working. It is also the right choice when you are preparing an elderly parent's or a child's appointment: it is then someone else's data you are handling.
In IA Confidential, your requests are handled by default by confidential models: open models installed on servers in Switzerland and Europe, which keep nothing of your exchanges. For preparing a consultation, they are all you need.
One principle applies to everything that follows: the AI needs neither your name, nor your date of birth, nor your insurance number to help you. "A 45-year-old woman with no particular medical history" is enough for it to reason.
The simplest approach: describe the situation in two sentences and ask for the questions.
I have a first appointment with a gastroenterologist about stomach pain that has lasted two months, mostly after meals. Prepare eight questions for me to ask, in simple sentences, sorted from the most important to the least urgent. Also tell me what I should note down before the appointment so that I can answer the doctor's questions.
The last sentence is the most useful: a doctor always asks the same questions (since when, at what time, what makes it better), and a few days of notes are worth more than vague memories.
The second level: you have received test results, and you want to arrive knowing what the conversation will be about. Copy out the values and the reference ranges, not the laboratory's letterhead.
Here are my blood test results, with the laboratory's reference ranges. Explain to me in simple terms what each value is for and what it means when a value is outside the range, without making a diagnosis. Then suggest five questions to ask my doctor about these results, and tell me which values they will probably want to check again.
If you would rather attach the laboratory's PDF, it is read in your browser. A lab report almost always carries your name, your date of birth and the name of the doctor who ordered the tests: if you have chosen an anonymised model, the assistant flags it as an "Identifying document", and the document is then handled by a confidential model, or only your question is sent. Masking the name does not make a record anonymous anyway, as our guide on anonymisation and pseudonymisation explains.
The third level, for a chronic condition or symptoms that have dragged on for months. Create a space, for example "Health follow-up": all the conversations on the subject are kept there, and you find the preparation for each appointment in one place. In your profile, the Constraints tab lets you state once and for all an allergy, a long-term treatment or a hearing difficulty: this information stays on your device, is passed only to confidential models, and is never sent to an anonymised model.
Here are my notes from the last three months on my migraines, and what the doctor said at the last consultation. Turn them into one page to hand to the neurologist: a dated timeline of the attacks, the treatments tried and their effect, the triggers I noted, then the four questions I absolutely must ask, sorted by priority. Add no facts, do not suggest any cause, and point out anything missing or contradictory in my notes.
The assistant writes this page as a document: you reread it, edit it, print it to PDF or export it to Word, and hand it to the doctor when you arrive. A clear page saves the five minutes usually spent piecing the story back together.
AI can be wrong, and it is wrong with confidence: a misread reference range, a value in the wrong place, an explanation that sounds plausible but is false. It has not examined you, does not know your medical record, and does not know what the doctor saw. Use it to understand and to put things into words, never to decide: no stopping a treatment, no changing a dose, no self-medicating on the strength of an answer.
Ask it explicitly not to make a diagnosis, and be wary of an answer that makes one anyway. If a symptom worries you before the appointment (chest pain, difficulty breathing, sudden weakness on one side of the body), do not wait: call 144 in Switzerland, 999 in the UK or 112 elsewhere in Europe.
Finally, AI prepares the consultation, it does not replace it. The best questions are often the ones the doctor's answers bring up: leave some room on your sheet.
With a confidential model, your symptoms, your results and the answers are processed on servers in Switzerland and Europe, under Swiss and European data protection law, outside the reach of the CLOUD Act, and nothing is kept on our side: the history stays in your browser. We could not hand it over if we were asked to.
Nothing goes to an anonymised model without your decision. If you choose one and your message contains sensitive data, the Confidentiality filter offers to answer with a confidential model, your message untouched, or to send an anonymised version that you review. Attachments only go out as pseudonymised extracts. Sending the message as it is remains your choice; for your health, stay with the confidential models. Your profile, constraints included, is never sent to an anonymised model.
Because the history is in your browser, anyone using the same session on the same device can read it. On a family computer or at the office, use a personal session, delete the conversation once the appointment is over if you don't want to keep it, and sign out when you leave.
On the doctor's side, the same care applies the other way round: our guide on the consultation report shows how a practitioner can use AI without exposing their patients. To prepare your next appointment from a situation you describe without your name: open the chat.
Try it on your own case No account needed · First month free
Open
