Accuracy
Your calorie count is a guess
Of every figure a wearable reports, calorie burn has the widest gap between how confident it looks and how well it is actually known. The validation research on this is unusually blunt.
Median error, same study
Same devices, same wrists, same afternoon. The sensor is fine — it is the calorie model sitting on top of it that is not.
You finish a workout and your band tells you it cost 486 calories. Not 480. Not “roughly 500”. Four hundred and eighty-six — a number precise enough to budget dinner around, produced by a device that has no way of knowing how much energy you used.
Of all the figures a wearable reports, calorie burn is the one with the widest gap between how confident it looks and how well it is actually known. The validation literature on this is unusually blunt.
What the studies found
The most-cited evaluation is a Stanford study that put seven consumer devices against laboratory reference measurement. Median error in energy expenditure ranged from 27.4% for the best device to 92.6% for the worst. Not one of the seven came in under 20%.
The same study measured heart rate on the same devices, at the same time. The Apple Watch's median heart rate error was 2.0%.
That contrast is the whole story. The hardware on your wrist is good. What it is being asked to do with calorie estimation is something the hardware cannot do.
| Study | Finding | Direction |
|---|---|---|
| Stanford, seven devices | Median error 27.4%–92.6% | Varied by device |
| Apple Watch vs metabolic cart | 24.3% relative error | Over for women, under for men |
| Fitbit Charge HR 2 | 18.6% men, 24.2% women | Underestimated both |
| Wheelchair users, treadmill | Both devices off | Apple under, Fitbit over |
| Cycling, four brands | ~13% of Fitbit trials implausible | Over 450% of reference |
Look at the direction column. The errors do not even point the same way — one device overestimates for women and underestimates for men; another underestimates for everyone; a third produced readings more than four times the true value in roughly one trial in eight. You cannot correct for a bias that changes sign depending on who is wearing it.
A number that is wrong by a consistent amount is still useful. A number that is wrong by an unpredictable amount in an unpredictable direction is decoration.
Why it is such a hard problem
Measuring energy expenditure properly means measuring gas exchange — how much oxygen you consume and how much carbon dioxide you produce. That requires a mask, a metabolic cart and a laboratory. No wrist device measures any of it.
So the band models it instead, from inputs it does have:
- Heart rate, which it reads well, but which relates to energy use differently in every person.
- Movement, from the accelerometer — excellent for steps, poor for load carried, incline, or anything your wrist does not do.
- Your profile: height, weight, age, sex. Four numbers standing in for your entire metabolism.
From that it estimates a resting metabolic rate and adds the cost of activity on top. Both halves carry error, and the big one is personal variation. Two people of identical height, weight, age and sex can have resting metabolic rates differing by several hundred calories a day, and movement economy that differs just as much. The model has no way of knowing which of them is wearing it.
Resistance training is the worst case. Heart rate during lifting is driven partly by effort and partly by intrathoracic pressure and static muscle contraction, and your wrist barely moves. A model built on heart rate and arm movement is being asked a question outside its design. Treat strength-session calorie figures as close to meaningless.
Why this matters most if you are eating back calories
Nobody is harmed by a wrong number on a screen. The damage happens when the number enters a decision.
If you are in a 500-calorie daily deficit and your band credits you 486 for a session that actually cost 300, you have quietly erased a third of your week. Do that four times and the deficit is gone, and the obvious conclusion — that your body is not responding — is wrong. The arithmetic was.
The error runs the other way too. A device that underestimates can have you eating less than you should while training hard, which is its own problem.
Don't
- Eat back the calories your band reports
- Compare the figure with a friend’s device
- Trust calorie numbers from lifting
- Pick exercise by what burns most
- Treat the daily total as a budget
Do
- Track duration and effort instead
- Compare this week with your own last week
- Use time-in-zone for cardio sessions
- Pick exercise you will repeat
- Judge intake by weight trend over weeks
What the number is still good for
It is not worthless. It is worthless as an absolute measurement, which is the only way most people use it.
Within a single device, worn the same way by the same person, the error is at least somewhat consistent — so the comparison between your Tuesday and your Thursday carries more information than either figure does alone. Used as a relative index of how hard a week was, it is reasonable. Used as a count of calories, it is not.
-
Use duration and intensity as the primary record
Forty minutes at a moderate effort is a fact. The calorie figure attached to it is a model output. Train and log the fact.
-
Compare yourself to yourself, weekly
Your own numbers across weeks share whatever bias your device has, which makes the comparison far more meaningful than the raw value.
-
Let the scale and the mirror settle the energy question
Body weight trending over three to four weeks tells you what your actual energy balance is doing. No wearable can.
-
Discount strength sessions entirely
If your app lets you hide the calorie field for resistance work, hide it. It is the least reliable output of the least reliable metric.
-
Watch the inputs you trust more
Your resting heart rate trend and step count are both measured far better, and both tell you more about whether the week went well.
The sensor is honest. It is the arithmetic afterwards that is making things up.
Where we land on this
Mesh 2.0 reports activity, but we lead the app with duration, intensity and your heart rate trend rather than a calorie total, because those are the parts the hardware actually knows. An estimate presented with two significant figures implies a precision nobody has.
It is the same reason the band has no screen: a number you cannot check at a glance is a number you are less likely to make a bad decision with. See how the app presents it, or read the full Mesh 2.0 specification.
Questions people actually ask
How accurate are the calories on a fitness tracker?
Not very. A Stanford evaluation of seven consumer devices found median energy-expenditure errors ranging from 27.4% to 92.6% across tasks, with no device coming in under 20%.
Heart rate measured by the same devices in the same study was far better — the Apple Watch’s median heart rate error was 2.0%. The sensor is good; the calorie model on top of it is the weak part.
Does Apple Watch or Fitbit measure calories more accurately?
Neither is reliable, and their errors do not even point the same way. In treadmill testing against a metabolic cart, the Apple Watch showed about 24.3% relative error, overestimating for women and underestimating for men, while the Fitbit Charge HR 2 ran 18.6% for men and 24.2% for women and underestimated for both.
In a cycling study across four brands, roughly 13% of Fitbit trials produced implausible estimates of over 450% of the reference value. Because the bias changes direction, you cannot simply correct for it.
Why can’t a watch measure calories properly?
Because measuring energy expenditure properly means measuring gas exchange — oxygen consumed and carbon dioxide produced — which needs a mask and a metabolic cart. No wrist device measures any of that.
Instead it models calories from heart rate, accelerometer movement and your height, weight, age and sex. Two people identical on those four inputs can differ by several hundred calories a day in resting metabolic rate alone.
Should I eat back the calories my tracker says I burned?
No. This is where an inaccurate number does real damage. If you are in a 500-calorie deficit and your band credits you 486 for a session that cost 300, you have erased a third of your week without noticing.
Judge intake by body weight trending over three to four weeks instead. That reflects your actual energy balance; a wearable estimate does not.
Are calorie estimates worse for weight training?
Considerably. Heart rate during resistance work is driven partly by static muscle contraction and intrathoracic pressure rather than aerobic demand, and your wrist barely moves, so both of the model’s main inputs are misleading at once.
Treat calorie figures from strength sessions as close to meaningless, and hide the field if your app allows it.
What should I track instead?
Duration and intensity, which are facts rather than model outputs. Time-in-zone for cardio. Your resting heart rate trend, which is measured directly and tells you more about how the week actually went.
If you want to use the calorie figure at all, use it only to compare your own weeks against each other on the same device, where whatever bias it has is at least shared.
Sources
- Shcherbina et al., Accuracy in wrist-worn, sensor-based measurements of heart rate and energy expenditure in a diverse cohort, Journal of Personalized Medicine, 2017 — seven devices; median energy expenditure error 27.4% (Fitbit Surge) to 92.6% (PulseOn); no device below 20%; Apple Watch median heart rate error 2.0%.
- Assessment of accuracy of overall energy expenditure measurements for the Fitbit Charge HR 2 and Apple Watch, 2019 — Apple Watch relative error 24.3%, overestimating for women and underestimating for men; Fitbit 18.6% for men and 24.2% for women, underestimating for both.
- Accuracy of the Apple Watch Series 4 and Fitbit Versa for assessing energy expenditure and heart rate of wheelchair users, JMIR Formative Research, 2024 — Apple Watch underestimated and Fitbit overestimated energy expenditure; both underestimated heart rate.
- Body fat, skin tone, and the accuracy of smartwatch caloric expenditure estimates, PLOS ONE — four-brand cycling comparison; bias lowest for Apple; approximately 13% of Fitbit trials produced estimates above 450% of reference values.
Figures are quoted as reported by the cited studies. Validation research covers specific device models and firmware versions tested under specific protocols, so newer hardware and other activities may perform differently. This article is general information about consumer health technology and is not medical or dietary advice. If you are managing weight for a medical reason, that is a conversation for a doctor or a registered dietitian rather than a wearable.
Mesh 2.0
Wear less.Know more.
A screenless band that tracks your heart rate, sleep and BioFuel — then stays out of the way. Available now across Pakistan.