Rapid
Wearable tech has a strange sales problem: the feature a device is advertised on is often the one with the weakest published evidence. Heart rate is measured well enough to be useful. Calories burned are not. Blood oxygen readings on a wrist drew an FDA safety communication and, in 2025, draft guidance aimed at fixing them. And "readiness", "stress" and "sleep quality" scores are vendor algorithms that no one outside the company can check.
This page sorts the sensors by how much of their output survives independent testing, so the money goes to the parts of the device that work.
1. What is actually inside a wearable
Almost every wrist device is built from the same short list of parts. Knowing them explains both the strengths and the failure modes.
- Optical heart rate (PPG). Green LEDs shine into the skin and a photodiode watches how much light comes back. Each heartbeat changes blood volume, which changes the light. This is the workhorse sensor and it also produces the raw signal for the blood-oxygen and stress features.
- Accelerometer and gyroscope. Motion. This is the most reliable thing in the device and the reason step counts and sleep-window estimates are roughly trustworthy.
- ECG electrodes. A few devices add two electrical contacts that let you complete a single-lead ECG by touching the case with your other hand. That is a genuinely different measurement from PPG, not a marketing rename.
- Skin temperature and bioimpedance. Used for cycle tracking and body-composition estimates. Both are indirect — they infer a body state from a signal that other things also move.
The important structural point: the sensor produces a raw signal, and the algorithm turns it into a number. Faults can come from either, and vendors document only the first.
2. The reading that holds up: heart rate
Wrist heart rate is the one feature with a large independent evidence base, and the best known study is still the sharpest. Shcherbina and colleagues at Stanford tested seven wearable devices — Apple Watch, Basis Peak, Fitbit Surge, Microsoft Band, Mio Alpha 2, PulseOn and Samsung Gear S2 — against continuous telemetry and indirect calorimetry in 60 volunteers while sitting, walking, running and cycling.
Their heart-rate finding was broadly reassuring: six of the seven devices achieved a median error below 5% during cycling, and the Apple Watch had the lowest overall error of the group.
Read that sentence carefully, because both halves of it matter. Heart rate at rest and on a bike is good enough for training. And "during cycling" is doing a lot of work, which brings us to the next point.
3. The reading that does not: calories burned
The same study measured energy expenditure — the calorie number on your watch face — and produced one of the most quotable sentences in consumer-wearable research:
"No device achieved an error in EE below 20 percent."
Not one, across seven devices and four activities. Bear in mind that this is the median error: half of the measurements were off by more than that. Energy expenditure is not directly measurable from a wrist at all — it is estimated from heart rate, motion and a model of your body, and that estimate carries the error of every one of those inputs plus the model's own assumptions.
The practical reading is not "throw the watch away". It is that calorie totals are the wrong thing to budget with eat back. If you eat back what the watch says you burned, you are budgeting against a number with a documented error floor above 20%, and you will not know which direction it erred.
4. Where the error concentrates
The Stanford cohort was deliberately diverse in age, height, weight, skin tone and fitness level, which is why it could show not just that error exists but who absorbs it. Device error was higher for:
- Walking — the study found the lowest error for cycling and the highest for walking. That is the awkward one, because walking is what most people actually do most of the time. Arm swing and a loosely seated sensor both inject noise.
- Higher body mass index. More tissue between the LEDs and the blood vessels means a weaker signal.
- Darker skin tone. Melanin absorbs green light, which is exactly what PPG uses. The optics are less favourable, and this is a hardware property, not a user error.
- Men, in that cohort, likely through the same body-composition mechanism.
An earlier study reported by the American College of Cardiology described error ranging from ±15 to ±34 beats per minute depending on the activity. A reading that can be off by 34 bpm at the moment you care most — the hard interval — is a trend line, not a measurement.
5. Blood oxygen: the FDA has now intervened
Blood oxygen (SpO2) is derived from the same optical sensor, and it is the feature most often sold as a health capability rather than a fitness one. It is also the one regulators have publicly qualified.
In February 2021 the FDA issued a safety communication, Pulse Oximeter Accuracy and Limitations, and its own product page lists the factors that affect accuracy: poor circulation, skin pigmentation, skin thickness, skin temperature, current tobacco use, and use of fingernail polish. That is a list of things about the wearer, not about the software.
In March 2025 the FDA went further with draft guidance proposing stricter accuracy standards, larger and more representative study cohorts, and objective assessment of skin pigmentation in device testing — explicitly to address measurement bias in darker-skinned patients.
What to do with this: a consumer SpO2 number is not a medical measurement, a single low reading is not a diagnosis, and a normal reading does not rule anything out. Changes over days, alongside how you feel, are more informative than any one value. If you are ill and worried about your oxygen level, that is a question for a clinician and a clinical device, not a watch.
6. ECG alerts: real, narrow, and not a diagnosis
Single-lead ECG is the one advanced feature that is a different measurement rather than a bigger number. Several wrist devices carry regulatory clearance for an ECG app and for an irregular-rhythm notification feature.
Two limits are consistently undersold in the marketing. First, a notification tells you the rhythm looked irregular; it does not tell you what the rhythm is, and it is designed for people with no known heart condition. Second, a cleared feature is cleared for a specific purpose, and using it as ongoing reassurance for a diagnosed condition puts you outside what it was validated for. A notification is a reason to talk to a doctor, not a result.
7. Sleep, stress and readiness scores: unverifiable by design
These are the features the product pages lead with in 2026, and the honest description is that they are vendor algorithms. The inputs are real — movement, heart rate, heart-rate variability, skin temperature. The output is a number produced by a model that is not published, not independently reproducible, and tuned to the vendor's own definition of a good night.
That does not make them useless. It makes them useful in exactly one way: as a personal baseline. If your own score, measured by the same device in the same way, moves, something about your routine changed. Comparing your score to a friend's, to a population average, or to a threshold the app invents is comparing numbers from different models.
Watch for the pattern where a low-confidence feature wears the language of a clinical one — "sleep stages" for what is inferred from motion and heart rate, or "stress level" for a heart-rate-variability index. The words are doing more work than the measurement.
8. How to decide, feature by feature
- Buy for: heart rate at rest and during steady cycling, step and activity volume, notifications, battery life, comfort, and whether it works with the phone you own. Those are the parts that survive testing.
- Add a chest strap if you train to heart-rate zones. A strap measures electrical activity directly instead of inferring it through skin optics, and it costs far less than the upgrade you would otherwise buy to get a "better" watch.
- Ignore calorie accuracy when comparing devices. Do not eat back the number.
- Treat SpO2 and ECG as nudges, not measurements, and never as reassurance.
- Treat sleep, stress and readiness as trends on your own device only. If two devices disagree, that is expected, not evidence either is broken.
- Prefer the older model. Sensors improve slowly; year-on-year changes are usually software, a bigger screen and a battery claim. Last year's device with the same sensor array is the better purchase.
The one-line version: a wearable is an excellent activity and heart-rate logger, a poor calorie counter, and a health monitor only in the narrow, regulated sense the fine print describes. Buy it for the first thing and you will not be disappointed.
FAQ
Are smartwatch calorie counters accurate?
No. In the Stanford study that tested seven wrist devices against laboratory calorimetry (Shcherbina et al., 2017), no device achieved an error in energy expenditure below 20%, and that was the median error — half of the measurements were worse. Calorie figures on a watch are estimates built from heart rate, motion and a body model, so do not eat back what they report.
How accurate is heart rate on a smartwatch?
Better than any other metric it reports. Six of the seven devices in that same study achieved a median heart-rate error below 5% during cycling, and the Apple Watch had the lowest overall error. Accuracy drops for walking, for higher body mass index, for darker skin tone, and during high-intensity intervals, where error has been reported from ±15 to ±34 beats per minute depending on activity.
Is blood oxygen on a smartwatch reliable?
Treat it as a rough indication rather than a medical measurement. The FDA issued a safety communication in February 2021 listing skin pigmentation, skin thickness, skin temperature, poor circulation, tobacco use and nail polish among the factors affecting pulse oximeter accuracy, and in March 2025 proposed stricter accuracy standards and more representative testing across skin tones. A single reading should never be used to rule anything in or out.
Is an Apple Watch ECG a real ECG?
It is a genuine single-lead ECG — an electrical measurement, not an optical estimate — and a device that carries clearance for it has been assessed for a specific purpose, such as irregular-rhythm notification in people without a known condition. It is not a diagnostic tool, it does not identify what an irregular rhythm is, and it is not a substitute for monitoring a diagnosed condition.
Why do my sleep score and stress score seem arbitrary?
Because they are produced by unpublished vendor algorithms. The underlying signals — movement, heart rate and heart-rate variability — are real measurements, but the score converts them into a number defined by the company. That makes a score meaningful only against your own baseline on the same device, and not comparable between devices or people.
Which wearable tech is actually worth paying for in 2026?
The parts with independent evidence: heart rate at rest and during steady aerobic exercise, step and activity volume, notifications, battery life and comfort. If you train to heart-rate zones, a chest strap is a better use of money than a more expensive watch. Calories, blood oxygen, sleep stages and readiness scores are the features where the evidence is weakest and the marketing is loudest.
Are cheap fitness trackers worse than expensive smartwatches?
Not necessarily, for the measurements that matter most. The Stanford comparison found one premium device with the lowest overall error and another with the highest, so price did not determine accuracy. A mid-range tracker with good wrist fit plus a chest strap will beat an expensive watch for training heart rate.
Does wearing a wearable improve your health?
The trustworthy benefit is behavioural rather than medical: step counts, activity streaks and heart-rate feedback help people move more. The parts that claim to detect or diagnose conditions are narrow, regulated features with specific limits. Use the device to change your habits first and to make health decisions last.
