Create a Personal Health AI

AI is going to be a big deal for healthcare. The thing is, you can start using it right now to make your healthcare better. A lot better.

You don’t need to be good at AI. No fancy plug-ins or MCPs or connectors. You don’t need to understand healthcare either. You don’t need to trust AI’s medical advice or think AI can diagnose.

All you need to do is start talking to the AI of your choice about your health. That’s it.

Share your worries, your experiences, your history. Share how it makes you feel, what you are doing, what you struggle with. When a health thing happens, like a visit, a vaccine, you need a med, just write about it. If you don’t know what to write, ask it to interview you.

Get Started

Start a new conversation thread. Make sure you have an account so your info is saved each time you come back, but that’s all you need. You can do more in a bit, but start here.

You can write whatever you want, but something like:

Hi, I’d like your help keeping track of my health. This is the start of my health journal. I’m going to write in here and, unless I ask a specific question, just respond with a brief 1 - 2 sentence summary showing me you understand. [Brain dump]

OR

Hi, I’d like your help keeping track of my health. This is the start of my health journal. I’m going to write in here and, unless I ask a specific question, just respond with a brief 1 - 2 sentence summary showing me you understand. First, though, I'd like you to interview me as if i was a new primary care patient in a wellness visit to start us off.

The key here is not to try to ‘document’. This isn’t a health record. You’re not trying to organize your meds or have a perfect health history. The goal is to manage your health from your perspective, not the health systems’. Every time any health concern or random health thing comes up, write about it in the health conversation.

That’s it. You’re doing it. You’re using AI as a personal health record.

Ask for help

You’ve given AI some context. Great. Now you can tell it what’s going on. The key, as always, is to write too much when talking with AI. For example:

I have a headache today

This is not that helpful. It’s good to log the datapoint, but it’s not going to be all that useful. Like I said, share as much as you can.

I have a headache today. It's behind my eyes and in the back of my head. I've had it all morning. Probably dehydrated but I keep getting these in the morning. A part of me is worried I have brain cancer even though I know that's silly. I don't think I'm hung over, I'm only having 3 drinks before bed. Maybe it's staring at my phone for an hour each morning. Maybe I'm just stressed about ...

You get it.

What do I do with this?

The simplest and best way to use a personal health record is to think of it like a clinical friend who can help with any time you touch the healthcare system. Let’s say you’ve got a sinus infection, you need to go in for a visit to get antibiotics. You can use the AI a couple ways:

First, if you haven’t talked to the AI yet, ask it to help you figure that out: “I think I’ve got a sinus infection and need antibiotics, can you help me figure out if that’s the case? I don’t want to pay for an office visit if they’ll send me home”

Second, if you have been talking to it, say updating it on your cold every other day with symptoms and stuff (again, not perfect, just your experience), you can say, “I feel like I’ve had this cold forever, is it a sinus infection?” The AI can look back at the convo and say ‘no, it’s only been two weeks, and it seems like you’re starting to improve, so probably not.’ Or ‘Probably. It’s been a month and your congestion is worse, now you’ve got headaches.’

Third, if you do need to go in, you can ask for help. “OK, I have an appointment. I never know what to say. I’m worried I’ll say the wrong thing! Help me prepare.” Here the AI can really shine, by helping you describe your symptoms or prepare for questions the provider might ask. After the visit, you can update and also share how the visit made you feel.

Context is why this works

When you’re getting care, there are two things you actually need your clinician to be good at. The first is medical knowledge. This is what we mostly think of when we think of ‘medicine’. It’s knowledge / facts (what’s Keflex?) and methods / process (how to do diff diagnosis). It’s basically the ‘science’ of medicine.

The science of medicine is also what ‘can AI do healthcare’ tests are benchmarking. Can AI figure out some obscure issue with the thinnest of data points and a bunch of misdirections? The answer is yes, but it won’t be a good clinician unless it has the thing every clinician needs: context.

The second part of medicine is patient context. This is the provider’s ability to gather the maximum amount of context about you from the sources available. “Any important surgeries or family medical history I should know? Any new concerns?” Some of this is bedside manner, some of it is knowing what to ask, some of it is ineffable. This is the art of medicine.

The best way to think of these things is the kind of care you get from a phenomenal clinician you get to meet for 10 minutes vs the kind of care you get from your clinician friend you’ve known for 10 years. Until now, you had to choose. Either see the best clinician for a given need or someone with deep nuanced context.

All good clinicians exist to combine the science of medicine (knowledge, methods, process) with the art of medicine (relationships, context). The two skill sets are mutually reinforcing. A good PCP uses their limited time with you to build up a picture of you. The stronger the relationship, the easier it is to use that context to wrap around a given issue. Your best clinician friend has an incredible amount of context to wrap around an issue.

AI has all the time in the world. It does not bore from hearing you kvetch about your knee pain, doesn’t get exhausted by your hypochondria, and will answer every detailed little question. Remember, it’s not hard. All you need is:

  1. To have an account with a major AI (Claude, ChatGPT, Gemini, Grok)
  2. To talk to it about your health in one big convo. Overshare. Lots of detail.
  3. To ask it to help you validate a concern or prepare for a visit.

That’s it! You’re now 80% of the way to revolutionizing healthcare.

Things that make this work even better

Like all things AI, you can go really crazy making this way better, but the main strategies are pretty simple. They are:

  1. Use projects. Projects are like a work computer for your AI. When it’s inside the project, it has the files it needs, the right context, and knows what is being asked of it. You can give it custom instructions (e.g. you’re my health assistant. Help me keep track of health concerns and help me get the most out of my visits) and upload files.
  2. Upload files. Specifically your medical records. I’m a One Medical patient and so can download a PDF of my records with a single click. You don’t need everything, just some stuff that seems most relevant.
  3. Create a summary. After a long conversation, the AI can write up a document, usually as an ‘artifact’ or ‘canvas’, both of which you can add back into the project, or to save as a PDF somewhere as a snapshot.

Concerns

Almost all concerns related to AI as clinical support can be addressed by “Think about it for 5 more minutes and use common sense.”

You may have concerns about privacy. I understand that. I recommend doing a bit of research on which companies are less great on privacy and then paying for the basic subscription to the one you like best. And by that I mean pay for Claude.

You may have concerns about diagnostic quality. I understand that. Don’t use AI for diagnosis. Use it to help you think through what something might be, how you would tell, and what home methods might work to treat it. AI will be better than all clinicians one day, probably soon. But that day could never come and AI as it is today will still make your care 90% better.

You may have concerns about sycophancy. I understand that. Ask the AI to help you evaluate a symptom. “I have a worse than usual headache and am worried, help me work through what it might be”, not “I think I have brain cancer, do I?” If something seems worrying or not right, say that to the AI. “This is still worrying me” or “this doesn’t make sense” or “that doesn’t seem right”.

You may have concerns about pissing off your doctor. I understand that. No matter what, you will not be the first person who does this to them. If you follow my tips, they’re more likely to be appreciative than cranky. If they’re not, reconsider them as your clinician.

If you’re doing this and having luck, or have questions. I’d love to hear them. Regardless of whether or not AI is a good doctor, it’s a phenomenal health assistant. Take advantage.