What is measured well
Step counting is the oldest and most reliable feature. Modern accelerometers count walking steps with reasonable accuracy for most people in most conditions; they miscount when you push a trolley, ride in a car on a rough road, or move your arms without walking, but as a day-to-day measure of activity volume it is sound. More importantly, it is actionable — the relationship between more daily steps and lower mortality is one of the more consistent findings in physical activity research, with the steepest gains at the low end.
Resting heart rate, measured overnight or at rest, is the other genuinely useful number. Optical sensors are accurate at rest and during steady activity, less so during high-intensity or irregular movement where a chest strap remains better. Resting heart rate tracked over weeks responds to training, illness, alcohol and poor sleep, which makes it a decent general indicator of how the body is doing. On devices with a cleared single-lead ECG feature, atrial fibrillation detection is a real capability that has led to genuine diagnoses — with the important caveat below.
What is estimated, and presented as measured
Energy expenditure is the clearest case. Your watch does not measure calories burned; it infers them from movement, heart rate and your entered characteristics, and independent evaluations have repeatedly found substantial errors, often in the tens of per cent. This matters most for anyone eating back the calories a device reports, which is an efficient way to undo a deficit. Treat the number as a relative indicator across similar activities, not as a quantity.
Sleep staging is the same story with a friendlier interface. Distinguishing light, deep and REM sleep properly requires measuring brain activity, and a wrist device is inferring stages from movement and heart rate variability. Total sleep time and bedtime consistency are tracked reasonably; the stage breakdown is a model output, and it varies between devices measuring the same night. Stress scores, recovery scores and readiness scores are proprietary composites, usually built on heart rate variability, with no external standard and no way to check them. They can be interesting as trends. They should never override how you actually feel, and if a low readiness score makes you skip a workout you had energy for, the device is costing you something.
False positives and the anxiety they cause
Screening a low-risk population with any test produces false positives, and this is not a flaw in the device so much as arithmetic. When a condition is uncommon in the people being tested, a large share of the alerts will be wrong even when the test is good. Irregular-rhythm notifications in young healthy users illustrate this well: the great majority who receive one turn out not to have atrial fibrillation, and each one still generates a call, a visit, possibly a monitor and a period of real worry.
The consequences are not trivial. Downstream testing costs money and time, incidental findings lead to further investigation, and a subset of people develop a persistent anxiety about their heart that is harder to treat than the false alarm was to generate. There is also the opposite error: a normal reading provides false reassurance. A single-lead ECG cannot exclude a heart attack, and paroxysmal atrial fibrillation can be entirely absent whenever you happen to check. If you have symptoms — chest pain, breathlessness, fainting, palpitations with dizziness — the device is not the relevant input. Our atrial fibrillation page covers what a real evaluation involves.
Continuous glucose monitors without diabetes
CGMs transformed type 1 diabetes care and are valuable in many people with type 2, particularly those on insulin. Extending them to people without diabetes is a different proposition and the evidence is thin. Glucose fluctuates in people with normal metabolism — after meals, with stress, with poor sleep, with exercise — and those excursions are physiology, not pathology. There is no established normal range for the metrics these consumer products report, no validated threshold for action, and no trial evidence that acting on them improves any outcome in a person without diabetes.
What a CGM does reliably provide is feedback, and some people find that genuinely motivating: seeing what a large plate of white rice does can change eating behaviour in a way that reading about it does not. That is a real benefit and worth acknowledging. The risks are equally real — pursuing an unnaturally flat curve, eliminating whole food groups on the basis of a spike, and in people vulnerable to it, feeding disordered eating. If you are curious, treat it as a short educational experiment with a defined end, and read our blood sugar guide for the interpretation. If your actual question is whether you have a glucose problem, the answer comes from an HbA1c or a fasting glucose test ordered by a clinician, not from a consumer sensor.
Using a wearable well
Pick two or three numbers and ignore the rest. For most people, daily step count, resting heart rate and time in bed cover almost all of the value, and adding more metrics mostly adds noise. Look at weekly and monthly trends rather than daily readings — nearly all of the useful signal in a wearable is in the trend, and nearly all of the anxiety is in the single reading. A resting heart rate that has drifted up over three weeks is worth noticing; one bad night is not.
Use the device to change one behaviour, not to accumulate data. Adding two thousand steps a day, going to bed at a consistent time, or getting a second strength session into the week are the kinds of changes a tracker can genuinely help with. And know when to take the thing off: if checking it has become a source of worry, if a readiness score is dictating decisions you would otherwise make sensibly, or if you have started managing the number instead of your health, a week without it is a reasonable experiment. It is a feedback device, not a diagnostic one.
Where wearables do have clinical value
Rhythm detection is the strongest case. Atrial fibrillation is common, often intermittent, frequently silent, and a major stroke risk factor — a condition where opportunistic detection genuinely helps, and where cleared consumer ECG features have produced real diagnoses that changed treatment. Wearables have also become useful in rehabilitation and in structured exercise programmes, where objective activity data helps a clinician see what is actually happening between appointments.
They are increasingly used in research too, providing continuous real-world activity and sleep data at a scale that would previously have been impossible. What they are not is a substitute for clinical measurement. Cuff blood pressure, laboratory glucose, formal sleep studies and twelve-lead ECGs exist because they are validated for the decisions they support. If a wearable prompts you to seek that testing, it has done its job. If it convinces you that you have already had it, it has done harm.