AI for health, without another app

Most health apps sell the same thing: a tidy interface that reads your own data back to you, for €9.99 a month. They didn’t collect the data. They didn’t draw the labs. They just wrap it in a nicer font and charge rent.

There’s a quieter, free alternative hiding in plain sight: use the general-purpose AI you already have as your health intelligence layer. No new app. No subscription. You learn the method once, and you keep the answers, not the vendor.

the architecture, not the app, is the thing

Apps promise to combine your wearable, your labs, and your notes into one tidy dashboard. In practice most of them just repackage the data and charge for the spreadsheet. A better approach splits the work into three roles — research, ledger, protocol — and lets you assemble them with tools you already own.

Why bother? Two reasons. First, you stop being locked into one company’s view of your body. Second, when a better AI ships next month (it will), you can swap it in without rebuilding your life’s health record. Your data stays where you put it.

the three roles, in plain English

Imagine three quiet helpers. One reads research and tells you what’s strong, what’s promising, and what’s being sold to you on Instagram. One holds your data — sleep averages, last bloodwork, current meds, what you ate this week. One turns those two things into a small, testable plan. None of them need to live in a single product.

The ledger can literally be a Google Doc. The research helper can be a free chat tool with web search. The protocol helper can be the same chat tool with a different prompt. The AI is the interpreter. You are the editor.

what AI is good at — and what it isn’t

AI is genuinely useful for summarising, pattern-spotting, and generating hypotheses. (“Hey, my sleep got worse three weeks before my period — is that a thing?” Yes, it’s a thing.) It is not great at causality, nuance, or rare clinical cases.

A small habit that fixes most of the misuse: ask it to label its claims. Strong, promising, or anecdotal. If it can’t tell you which, that’s your answer.

And remember: garbage in, sensible out. AI cannot rescue inconsistent dates or mislabelled meds. That’s a ledger problem, not an AI problem.

the playbook — without paying anyone

1. Collect — export what you already have. Activity logs, glucose or BP exports, the PDF your GP gave you, supplement bottles. CSV, PDF, plain text. Don’t buy anything new yet.
2. Store — put it all in one folder you control. A Google Drive folder, an iCloud folder, a USB stick. That folder is your ledger.
3. Query — open a free long-context chat tool. Paste in a chunk. Ask for a summary, a trend line, a hypothesis. Ask for plain English.
4. Validate — ask the same tool to find the strongest evidence for and against its hypothesis, and to label confidence. Cross-check anything load-bearing with a primary source.
5. Act — turn one validated finding into one small, time-boxed experiment. Two weeks. Log adherence and how you felt back into the ledger. Boring. Repeatable. Yours.

Five steps. Zero subscriptions. The whole thing fits on a Sunday afternoon and costs you a cup of tea.

the trade-offs, honestly

• Convenience vs control — apps are easier. They’re also closed. Pick which one you actually want.
• Features vs privacy — deep integrations need vendor access. Sometimes worth it; sometimes not.
• Speed vs accuracy — longer prompts help, but they also need a little patience. AI is faster than your old Sunday spreadsheet either way.

If you’re a practitioner, this approach gives you an audit trail and stops you depending on whichever wellness platform is currently being acquired. If you’re managing a chronic condition, it gives you portability — you can take your record to a new clinician without exporting from seventeen different apps.

“Treat the AI as an interpreter of your record. Not as the record itself.”

A few prompts that punch above their weight. Ask the research helper: “Summarise current evidence on X. Strong, promising, or anecdotal? Cite sources.” Ask the ledger helper: “Make me a timeline of these notes by month.” Ask the protocol helper: “What three low-risk experiments could I run this fortnight to test this hypothesis?” And always end with: “Show me what you’re uncertain about and what you might be missing.”

Privacy housekeeping: only export what you’re comfortable keeping locally. Use end-to-end encrypted storage where you can. Keep a copy somewhere that isn’t the same platform. Boring discipline. Pays off the day you change tools.

Start small. One export, one conversation, one experiment. Two weeks. Then evaluate. That tempo gives you real learning without the heroic life-overhaul that nobody actually finishes.

The honest case isn’t that AI will replace apps. It’s that you can use AI as the intelligence layer above a ledger you control. That architecture privileges sovereignty, transparency, and the ability to upgrade. It scales from curious to cautious to clinician. And it’s simple enough to start with one file and one Sunday.