Why your health app's advice doesn't fit you — and what that costs

Short answer

Most health apps personalise the wrapper, not the advice. The tone changes, the plan gets a new name, the score moves with your sleep — but the underlying logic is the same whether you're a healthy 28-year-old, a woman in menopause, someone on beta-blockers, or someone whose body crashes two days after ordinary activity. The industry is careful about the risk of saying something too specific and wrong. It barely counts the risk of saying something so general it's wrong for millions of people.

Originally published on Founder And The City, Jane's newsletter about building in health tech. This version is written for the person on the receiving end of the advice rather than for the people shipping it.

And if you have followed the plan properly for years and it hasn't worked — that isn't laziness, that isn't a discipline problem, and it is not your fault. It's the most commonly misattributed failure in this whole category, because the advice was never built for your body in the first place, and nothing in the interface tells you that. You can do everything right against a plan that was quietly wrong for you.

Why the app that recommends music knows you better than the one tracking your health

Start with the uncomfortable comparison. A streaming service can usually guess what you want to hear. A health app still struggles with a woman in menopause losing bone and muscle, a fifty-two-year-old man with hypertension on beta-blockers, or someone whose body crashes 24 to 72 hours after ordinary activity.

That isn't because music matters more. It's that music personalisation is a preference problem and health personalisation is a care problem.

Music learns from dense, continuous, low-stakes feedback — what you play, skip, save, and never want to hear again. Health learns from sparse, noisy, fragmented, high-stakes data, and usually without knowing the thing that matters most: who this body actually is, what else is going on in it, and what this person can safely do.

Adaptive is not the same as personalised. Plenty of apps do adapt: bedtime suggestions shift, readiness scores move with your sleep, training plans adjust to cycle phase. Those are real. But genuine personalisation in a health sense would mean the core logic changes according to sex and life stage, existing conditions and their combinations, medications that alter exercise response, your own long-run baseline, functional limits, and the actual conditions of your life.

A 2023 overview of personalised mobile health tools put the distinction plainly: these systems mostly rely on behaviour-change theory, gamification, and motivational messaging, and personalise the content rather than the functionality.

Tailoring content isn't the same thing as personalising care.

Three ways generic advice goes wrong

This gets abstract fast, so here are three concrete versions of the same failure.

Menopause. A woman in her early fifties opens a supportive app. It tells her to walk more, do yoga, breathe deeply, sleep well, be kind to herself. She does all of it, for years. What the app never says — because it isn't built to — is that bone needs something heavier than walking and harder than yoga. The Royal Osteoporosis Society is explicit that keeping bones strong through and after menopause takes both weight-bearing impact exercise and muscle-strengthening exercise. She doesn't crash. Nothing goes visibly wrong. She just loses bone quietly, doing exactly what she was told was good for her.

Hypertension and beta-blockers. A man starts a training app. He has hypertension and takes beta-blockers. The app doesn't ask about either. He straps on a monitor and trains to heart-rate zones. But beta-blockers blunt heart rate by design — the American Heart Association notes that target heart rate may need recalibrating with a clinician, because these drugs affect people differently. Guidance from Exercise is Medicine adds that on these medications, perceived exertion is the safer guide than the number on the watch. None of that is in the app, so he chases zones that don't apply to him: sometimes overshooting without knowing, sometimes undertraining because the screen says he's fine.

Post-exertional malaise. A woman with long COVID has good days and bad days. Her tracker congratulates her: streak achieved, ring closed, goal exceeded. On the days she pushes for the green ring, she pays for it two days later. The largest peer-reviewed long COVID synthesis, in Nature Reviews Microbiology, states directly that graded exercise therapy is explicitly not advised as a treatment and can cause injury. An entire class of standard app logic — step nudges, scores that reward steady upward progression, "build capacity" framing — isn't merely unhelpful here. It can make people worse. If this describes you, post-exertional malaise and pacing are the relevant reading, not any recovery score.

In all three, the app never said anything obviously false. It said something general. And general was the wrong tool.

The industry is afraid of the wrong mistake

Here's the asymmetry, stated plainly.

Health tech is terrified of commission error — being too specific and getting it wrong. Telling someone on beta-blockers to chase a zone that doesn't apply. Telling someone with an energy-limiting condition to push through. The fear is rational: the liability is real, and regulation makes staying broad and lifestyle-shaped the price of remaining a wellness product at all.

But the same industry is remarkably relaxed about omission error — saying something so generic that it's wrong for millions of people while still sounding safe.

The working assumption, never said out loud, is that bad personalisation is dangerous and absent personalisation is neutral.

It isn't neutral. It's just uncounted. The cost of being too specific is visible — a complaint, a letter, a feature pulled. The cost of being too generic is invisible: the bone that was never preserved, the flare that was provoked, the training zone chased for two years, the preventive change never started because the advice on screen was too vague to act on. Nobody invoices anyone for those.

The "average healthy user" is mostly fictional

It's tempting to read the three scenes above as edge cases. They aren't.

Roughly three-quarters of US adults have at least one chronic condition, and about half have more than one. Even among 18-to-34-year-olds — the group these products design for hardest — around six in ten already have one. And of the adults with no diagnosed condition, the question becomes how many are actually metabolically healthy: a large NHANES-based analysis published in the Journal of the American College of Cardiology found only 6.8% of US adults met criteria for optimal cardiometabolic health. Roughly one in fifteen.

The trend runs the wrong way too. Between 2013 and 2023, the share of US adults aged 18–34 with multiple chronic conditions rose by about a quarter.

So the healthy optimiser — stable physiology, room to push, no medication interactions worth modelling — is the edge of the user base, not its centre. The centre is someone managing one or more conditions, often invisibly, often undiagnosed, often without the vocabulary for what their body is doing.

Why this happened, and why it isn't anybody's fault exactly

It's worth understanding, because it tells you what will and won't change.

The evidence base is uneven. Women remain meaningfully underrepresented in exercise and sport science. One large audit across six major journals found female participants averaged around a third of subjects, with only a small fraction of studies looking exclusively at women. A great deal of what reaches consumers as "women's wellness" is adjusted male physiology rather than a separately built evidence base — and adding a cycle tab doesn't close that gap.

Medicine is organised around single diseases. Real adult bodies increasingly carry several at once. The 2022 Nature Reviews Disease Primers synthesis on multimorbidity says it about as directly as a clinical journal will: guidelines, training and delivery focus on single diseases, and that focus can produce care that is sometimes inadequate and potentially harmful for people with several. It's far easier to build a product for "sleep" than for a person living with hypertension, perimenopause, migraine, anxiety and a parent in care — but the second is most of the user base.

Regulation draws the line there. Software that maintains or encourages a healthy lifestyle, unrelated to disease, sits outside the device definition. Get disease-specific and you enter a much harder world — which we wrote about in more detail in why your watch won't show you your blood pressure. So the market learned the predictable lesson: "optimise your sleep" is safe to ship; "because you're on beta-blockers, this heart-rate logic may be wrong for you" is not. The wellness category is, by design, the un-individualised one.

And the conditions of your life outweigh most of it anyway. The WHO's 2025 report on social determinants states that the circumstances in which people live and work can outweigh genetics, healthcare access or personal choices in shaping outcomes. A plan that quietly assumes stable sleep, spare time and uninterrupted adherence can be safe in every clinical sense and still be impossible for the person it was built for.

None of these is malice. Every layer independently tilts toward the simpler product, the simpler claim, the simpler user. The result is predictable — and the cost isn't zero just because nobody counts it.

Neither the population average nor a sample of one

If tailoring isn't personalisation, the obvious fix seems to be the opposite extreme: forget averages, model every body on itself. Pure n=1.

That doesn't work either, and not for the reason people assume. The problem isn't computing power. It's that comparing you only to yourself leaves you without orientation. When you feel worse than you did three months ago, "different from your own baseline" tells you something changed. It doesn't tell you whether what's happening is what happens to bodies like yours, whether people in your situation got better, or what they did differently. For most people, being your own experiment of one isn't empowerment. It's being left alone with a dashboard.

Jane's argument is that the useful middle is phenotype-level: not a marketing persona and not a single diagnosis, but a group of people genuinely similar in the ways that matter — symptoms, physiology, function, medications, life stage, and real-world circumstances.

The reason that's the right unit isn't methodological elegance. It's that people learn by analogy to people like themselves. Everyone who has ever joined a long COVID group, a perimenopause forum or a fibromyalgia community was looking for exactly this: the right reference class. Other bodies doing something close enough that pattern recognition becomes possible. Recognition is what makes everything downstream work — better questions, better experiments, better conversations with a doctor.

What to do while the industry catches up

Practical, and none of it requires waiting for anyone.

Find out whether your medication changes the numbers. This is the single highest-value question if you take anything for blood pressure or heart rhythm, because heart-rate-based guidance may simply not apply to you. Ask specifically, and ask what to use instead.

Treat perceived exertion as real data. How hard something felt is information, not a failure to be objective — and in several situations it's more reliable than what the device reports.

Judge advice by whether it names your situation. Guidance that would read identically to a twenty-eight-year-old athlete and to you is not guidance about you. That isn't a reason to distrust everything; it's a reason to check specifics before acting on them.

Track your own pattern rather than your compliance. Whether you closed the ring matters much less than whether the thing you changed made you feel and function better over weeks. Those often point in opposite directions.

Be careful with anything that rewards steady upward progression if you have an energy-limiting condition. That design assumes capacity builds linearly with effort, which for some bodies is exactly backwards.

When to see a doctor — and what to ask

Nothing above replaces a clinician, and the specific questions worth asking are surprisingly concrete.

Book an appointment before starting or significantly increasing exercise if you have a diagnosed heart condition, uncontrolled blood pressure, diabetes, or an energy-limiting condition — and if you take beta-blockers or other rate-controlling medication, before you use any heart-rate-based training plan at all.

Seek urgent help for chest pain, unusual breathlessness, fainting or near-fainting during exertion, or an irregular heartbeat that's new to you.

What to ask. Whether your medication changes what your heart rate means during exercise, and what target or method to use instead. What kind of exercise your situation actually calls for, and specifically whether resistance and impact work belong in it. Whether there's anything you should not do. And what a sensible starting point looks like given where you are now rather than where a generic plan assumes you are.

What to bring. A short written list of what you've been doing, for how long, and what happened — including anything that reliably makes you feel worse. Concrete beats adjectives.

If you're waved off with "just be active" — that is the same generic answer in a consulting room. It's reasonable to ask the question again with specifics: "Given this medication and this condition, what should I actually use to gauge effort?"

Start understanding your body

Learn what affects your energy, stress, sleep, and daily state. Get the app.

Explore your data

How Welltory fits — and what we're changing

We're part of the category being criticised here, so this section is an accounting rather than a pitch.

Two things Jane has publicly said we're stopping. We stop designing as though our average user is a healthy optimiser — the data says the healthy optimiser is the edge of our user base, not its centre, and the roadmap should reflect that. And we stop calling personalisation what isn't personalisation: adaptive messaging isn't personalisation, a score that moves with your sleep isn't personalisation, and Plan A versus Plan B isn't either.

She is equally direct about what we haven't done. We do not yet adapt the product well enough for post-exertional malaise, for menopause, or for migraine combined with IBS — three of the situations we see most often among our own users. This is a statement of intent, not a description of a finished product.

What we do honestly today is narrower, and worth saying plainly:

We show your own patterns over time — sleep, stress load, resting heart rate, recovery — rather than scoring you against a population average.

We can tell you whether something you changed actually did anything, which is a smaller and more useful claim than telling you what to change.

We're working condition by condition rather than all at once. The first community cohort is for people with post-exertional malaise, run jointly with the people in it and with medical advisors involved.

What we can't do is diagnose anything, tell you whether a medication is affecting your numbers, or replace a clinician on any of the questions above. And if our advice ever reads as though it would apply equally to anyone, that's the failure described in this article, happening here.

Read more from Jane

Jane writes about health tech, evidence and the gap between what the industry claims and what it delivers in Founder And The City. The original goes considerably further than this version — into the economics of omission, why almost every institutional layer rewards simplicity, the arithmetic of what generic advice leaves on the table, and the full version of what Welltory is committing to.

Discounts for blog readers: up to 36% off

See what affects your energy, stress, sleep, and daily state with Welltory

Explore your data