Most people have a year of health data sitting on their wrist and no idea what any of it means.
Your watch knows your sleep stages. Your phone counts your steps. The Oura ring or the Whoop band tracks heart rate variability, stress, recovery, and resting heart rate down to the minute. You glance at the daily score, decide whether it confirms or contradicts how you feel, and move on. The deeper picture, the patterns across weeks and months, never gets looked at.
AI changes the math. Export 90 days of data, drop a redacted version into Claude or ChatGPT, and ask the right questions. Patterns that would take a quantified-self obsessive a year to find surface in about ten minutes.
This article is about doing that safely. Two things matter most. Strip the identifying parts before uploading. And be clear about what AI can and cannot tell you about your own health.
Why most health data sits unread
Three reasons.
The dashboard is built for the wrong question. Apple Health, Fitbit, and Oura all show you today's number. Today's number is barely useful. The question worth answering is "what's different about this week compared to the last six," and no consumer wearable shows you that without you doing the work.
The data lives inside the app that captured it. Apple Health doesn't talk to Oura. Oura doesn't talk to your therapy notes. You become the manual integration layer between them.
Analysis takes hours you don't have. Even if you exported the data into a spreadsheet, a real analysis would mean writing formulas, building charts, and learning enough about your own physiology to interpret what you're seeing. AI compresses that into a ten-minute conversation, which is the difference between something getting done and something getting added to the list.
The redaction step (this matters)
Health data is sensitive. Treat it that way before uploading.
Most wearables let you export a CSV or JSON file. You don't need to upload the raw file. You need a stripped version with just the columns AI actually uses to find patterns.
Strip these before uploading: any field that contains your name, account ID, device serial number, GPS coordinates, exact addresses, or anything tied to a medical record number. If your data includes labeled events like "doctor's appointment" or "ER visit" or specific medication names, replace those with generic placeholders like "appointment" or "health event."
Keep these: date, the metric (sleep duration, steps, resting heart rate, etc.), and the value. That's the entire data shape AI needs to spot patterns.
Five minutes of CSV cleanup. Almost every privacy risk gone.
If your data has anything tied to specific medical conditions, medications, or therapy notes, don't upload it at all. Talk to a doctor instead. AI is a pattern tool, not a clinical one.
Audit 1: Sleep patterns
The first useful pass is sleep, because almost every other health metric is downstream of how you actually sleep.
Drop your 90 days of redacted sleep data into Claude or ChatGPT with this prompt:
"Here's 90 days of my sleep data: total sleep time, deep sleep time, REM time, time in bed, and wake time. Tell me: 1) what's my average total sleep across the period, 2) which days of the week are best and worst, 3) is there a trend (improving, declining, stable), 4) what's the gap between my best week and my worst week. Show numbers. Don't interpret them medically. Just describe the patterns."
The "don't interpret them medically" line is important. AI's default tendency is to add helpful commentary like "this could indicate" or "you may want to consult a doctor about." Strip that out. You want the patterns, not the unqualified medical guesses.
What you do with the patterns is your call. Most people find their worst sleep days correlate with something they already suspect. Late dinners. Stressful work days. A specific drink. The data confirms what your body has been trying to tell you.
Audit 2: Energy and activity patterns
The second useful pass is the relationship between movement and how you feel.
"Here's 90 days of my step count, active minutes, and resting heart rate. Tell me: 1) which weeks had the highest activity, 2) which weeks had the highest resting heart rate (a stress proxy), 3) is there a pattern between activity weeks and rest heart rate weeks. Describe what you see in numbers. Don't suggest causes. Just show the data."
Again, the "don't suggest causes" line keeps AI honest. Correlation is not causation. AI is excellent at finding correlations and terrible at distinguishing real ones from coincidence. Your job is to look at the pattern and decide if it matches what you already know about yourself.
A common finding: the weeks people log the most exercise are often followed by weeks of lower resting heart rate, then a return to baseline. That's not a discovery. It's confirmation of a thing your trainer would have told you. The value is in seeing it in your own data.
Audit 3: What changed
The third pass is the one that earns the whole exercise.
"Here's 90 days of my data. Compare the first 30 days to the most recent 30 days. What's measurably different across all metrics? Show the change in absolute numbers and percentages. Don't speculate about causes. Just show what changed."
This is the audit that catches the things you'd miss. The slow drift in average sleep duration. The gradual rise in resting heart rate. The activity that quietly dropped off two months ago. These are the patterns you can't see in any daily dashboard because each daily number looks fine in isolation.
The "what changed" view is the difference between owning a wearable and getting something useful from it.
This is not medical advice
A clear boundary worth saying out loud.
AI finds patterns in numbers. It cannot diagnose anything. It cannot tell you whether a pattern is dangerous. It cannot replace a doctor who can actually examine you, order tests, and put the data alongside your full health context.
If anything the audit surfaces concerns you (a clear downward trend in sleep, a sustained rise in resting heart rate, anything that doesn't match how you feel), bring the numbers to a doctor. Don't ask AI what to do about it. The Vora 2026 wearable data guide put it directly: "This content is for informational purposes and does not constitute medical advice." That framing applies here too.
If your situation involves chronic pain, mental health symptoms, eating concerns, sleep that's persistently bad despite your best efforts, or anything that interferes with daily life, the answer is a professional, not an AI prompt.
The cult vs the curious
There's a version of this where you spiral into checking your data three times a day, optimizing every variable, and quietly developing a worse relationship with your body than you started with.
Don't do that version.
The point isn't to become a quantified-self obsessive. The point is to take ten minutes once a quarter, look at the patterns, and see if anything obvious deserves attention. Then close the dashboard and live your life. The wearable is a tool, not a referee.
A good signal you're using this well: the audit takes 10 minutes a quarter and you barely think about your data the rest of the time. A bad signal: you're running new audits weekly and changing your habits based on every blip. If you find yourself in the second mode, take the watch off for a week.
If you want the broader workflow of using AI for daily habits beyond health tracking (drafting in your voice, weekly planning, the second-draft critique loop, end-of-day brain dumps), I just published Your AI Operating System: The Beginner's Field Guide to Letting AI Do Your Busywork on Gumroad. Volume 1 of my AI for Real Life library. Volume 9 will go deep on AI for health, habits, and self-care, where this audit workflow becomes one of the central tools. Launch price for Volume 1 is $19, and existing buyers get every future volume free as I release them.
A year of health data isn't insight. It's raw material.
Ten minutes of redacted analysis turns it into something you can actually use.
Tags: Health, Artificial Intelligence, Productivity, AI Tools, Privacy