Can an AI assistant manage your health?
Yes - a general AI assistant gets you a surprising distance. If you are willing to become the systems integrator, the librarian, the prompt engineer and the project manager of your own health, and stay all four of those things for the next twenty years. Here is exactly where it breaks, in the order you will hit it.
We are not going to tell you the models are bad. They are remarkable, they improve every few months, and for explaining what a marker measures or what a term means, a good general model is genuinely excellent. Hundreds of millions of people take health questions to these assistants every week, and they are not wrong to.
The question worth asking is different: what happens when you try to run your actual health through one, week after week, for years? That is where it stops being about the model at all.
1It cannot see you, because your data is not anywhere
A model is only ever as good as what you paste into it. And what you would need to paste in does not exist in one place.
To genuinely set this up yourself, you would have to:
- Log into every lab portal you have ever used and export the results
- Dig prescriptions out of WhatsApp threads and email attachments
- Request records from clinics that consider those records theirs, not yours
- Reconstruct the years where a doctor handed you a prescription and no report at all
- Export your wearable history separately, per device, per cloud, per subscription, and do it before you cancel any of them
- Normalise all of it into something readable
- Then do the whole thing again every time a new result arrives, forever
Most people have four or more pools of health data and no way to join them. The chatbot does not have a data problem. You do - and it hands the problem straight back to you.
We wrote about why this happens, and why no lab, clinic or device maker is structurally set up to fix it, in your health data has no home.
2The setup is a specialist skill
Even with the data in hand, using a general model well for health means knowing how to structure a project, write instructions that persist, keep files current, and hardest of all - ask the right question.
That last one is the real filter. The quality of the answer depends entirely on the quality of the prompt, which means you have to already know what matters in order to find out what matters. That is circular, and it is exactly what someone without medical training cannot do.
The person who most needs the help - busy, non-technical, mildly worried - is precisely the person who will never get through the setup.
3It forgets, and it stops
Even working well, the conversation is fragile.
- You hit a limit mid-thought. You are part-way through working something out and the conversation simply ends.
- A new session means re-establishing everything. Re-upload, re-explain, re-anchor. Remember my triglycerides. Remember I am pre-diabetic. Remember what we agreed last week.
- Context gets compressed or dropped silently. You do not know what it still holds. It may be answering off half your data, confidently, and you would never know.
- Behaviour changes mid-plan. One person we spoke to had fed a general model his data daily for over a week while working a plan, and it simply declined to continue. No explanation.
4It is a cage
This is the one people underestimate until it happens to them. You upload your life into one platform and you are stuck there - not by contract, by inertia. Every report, every conversation, every correction lives inside one vendor’s walls.
A better model ships next month. Moving means reconstructing everything by hand, so you stay on the worse one. Models get retired and the replacement behaves differently. You cannot even evaluate which model is better for you without running your health in parallel across two of them.
Every AI platform wants to put you in its cage and have you live there. The alternative is to put the person at the centre and let the models come and go.
That is a design choice, and it is ours: we swap the engine underneath whenever something better ships. You keep the relationship, the history and the context, and never have to care where the intelligence came from.
5It waits
A chatbot is reactive by nature. It answers; it does not act.
It never opens the conversation. It does not notice your steps dropped, your sleep slid, or that you went quiet for eleven days. It has no idea what you did after its last answer, so it never learns whether the advice worked - ask again next month and it starts from zero. It does not flag that a re-test is due, or brief the professional you are about to see.
A manager works the other way round: your resting heart rate has been climbing this week - how is sleep? Or, when you mention what you ate: good, that is high fibre, which helps your triglycerides. You never asked. It already knew.
6Nobody is accountable
A general model will analyse your report confidently and sometimes be wrong, in fluent, reassuring prose, and you have no way to tell the difference. In a 2023 assessment published in Clinical Chemistry, a leading general assistant answered only about a quarter of laboratory medicine questions correctly, with a further third partially correct.
It is also built to agree with you - ask a leading question and you will get a validating answer, which is dangerous in health. On consumer tiers your health conversations may be used to improve the model. And it has no sense of your boundaries: it will push for detail you did not want to give.
What a health manager does instead
| General AI chatbot | A health manager | |
|---|---|---|
| Proactive | Waits for you to ask | Watches your data and comes to you |
| Trusted | Confident, unvalidated, may be working off half your data | Reviewed by qualified experts before it reaches you |
| Private | Your health data may train the model | Never sold, never used to train, never used to sell you anything |
| Effortless | You integrate, organise, prompt and maintain | A photo, a voice note, a report - it is handled |
They are all extraordinary models. But a model is not a manager. A general assistant answers your health questions. A manager manages your health.
Questions people ask
Can an AI assistant interpret my blood test results?
It can explain what a marker measures and what a value outside the reference range generally means, which is genuinely useful. What it cannot do is read your result against your own history, your medications, your sleep and your previous panels, because that information is scattered across labs, clinics and devices and does not exist in one place for you to paste in. A 2023 assessment published in Clinical Chemistry found a leading general assistant answered only about a quarter of laboratory medicine questions correctly, so treat any interpretation as a starting point for a conversation with a clinician, not a conclusion.
Is it safe to upload my medical reports to an AI assistant?
Consider what you are agreeing to. On consumer tiers, conversations may be used to improve the model unless you turn that off, and the terms are not the same as those a health service operates under. If you do use a general model, remove identifying details first. Several people we interviewed started anonymising their own health data before pasting it in, which tells you something about how the experience feels.
What can a health manager do that a chatbot cannot?
Four things. It comes to you rather than waiting to be asked, because it watches your data continuously. It works from a picture you did not have to assemble. Its output is validated by qualified people before it reaches you. And it acts in the world - flagging when a re-test is due, briefing the professional you are about to see, and following up on whether what you tried actually worked.
Will AI models eventually be good enough to do this alone?
The models are already extraordinary and will keep improving, and we adopt the best ones as they ship rather than betting against them. But model quality was never the limitation. The limitation is that your health history has no home, that nobody is accountable for what an unvalidated answer tells you, and that a model cannot book a test or brief a doctor. Those are structural problems, not capability problems.
Fyxlife helps you understand and manage your everyday health. It is not a medical device or a healthcare provider, and does not diagnose, treat or replace your doctor. Always seek the advice of a qualified healthcare provider with any questions about a medical condition.