Key takeaways
- Wrist wearables read heart rate within a few percent and estimate energy burn with median errors from about 27% to over 90%.
- Heart rate is measured; calories are modelled from it. That one step is where the error enters.
- Cardio machines report gross calories, which include the 60–100 per hour you would have burned sitting still.
- Eating back tracked exercise calories double-counts activity your target already includes.
- Your real expenditure is best inferred from three weeks of intake and weight data, not from a device.
The calorie figure is the least reliable number your watch shows you. When researchers have put wrist wearables against laboratory indirect calorimetry — the gold standard, which measures oxygen consumed and carbon dioxide produced — heart rate came back within a few percent, while energy expenditure did not: median errors ran from roughly 27% on the best device tested to over 90% on the worst. The same wrist, the same minute, two very different qualities of number.
Why is heart rate accurate but calories are not?
Because only one of them is measured. A green LED on the underside of the watch reads blood-flow pulses through your skin, which gives a genuine measurement of heart rate. Nothing on the device measures energy. The calorie figure is the output of a model that takes your heart rate, age, sex, height and weight and predicts how much fuel a person like you would burn at that intensity.
The trouble is that heart rate is a weak proxy for energy expenditure. Your pulse rises with caffeine, heat, humidity, stress, poor sleep, dehydration, illness and simply standing up — none of which burn meaningful extra calories. It also stays low during genuinely demanding work like heavy lifting, where effort is brief and muscular rather than cardiovascular. The model cannot tell these apart, so it converts a stressful hot morning into calories you never spent.
| What the device does | How it is obtained | How much to trust it |
|---|---|---|
| Steps | Accelerometer, directly counted | High — good enough to act on |
| Heart rate | Optical measurement at the wrist | High at rest and steady cardio |
| Distance | Steps × stride, or GPS | High with GPS, moderate indoors |
| Active calories | Modelled from heart rate and your profile | Low — treat as a relative signal |
| Total daily calories | Modelled resting rate plus the above | Low — compounds two estimates |
The reliability gap sits at the point where a measurement becomes a prediction.
The errors do not cancel out
A device that is 30% high on some days and 30% low on others would still be usable — the noise would average away across a month. But wearable error tends to run in a consistent direction for a given person and a given activity, which means it accumulates. Thirty per cent of a 500 kcal reported burn is 150 calories a day pointing the same way, or about a thousand a week.
Do cardio machines overestimate calories burned?
Generally yes, and ellipticals are the usual worst offender. A machine knows its own resistance and speed but not how much of the work your arms are doing, how much you are leaning on the handles, or how efficiently you move — so it assumes a standard body doing standard work. Treadmills do better, because carrying your own body weight over a known distance is a far more constrained problem, and rowers and bikes with real power meters do better still.
Gross calories versus net calories
Almost every machine and app reports gross calories: the total your body used during that hour, including the 60–100 you would have burned sitting on the couch. Only the net figure — the excess above resting — is genuinely attributable to the workout. On a 45-minute session, that alone accounts for a 50–75 calorie overstatement before any modelling error.
~27%
Best-case median error
Energy expenditure, best wrist device tested against calorimetry
60–100
Resting calories per hour
Counted inside the 'calories burned' a machine shows you
<5%
Typical heart-rate error
The same devices, on the measurement they actually take
Fix the input you can actually control
Estimating what you burn is hard; recording what you eat is not. 2BIB scans meals with AI, tracks calories and macros, and recalibrates your TDEE from your real weight trend — so your target comes from your own results rather than a wrist model.
Should I eat back the calories my watch says I burned?
In most setups, no. If your calorie target was built by multiplying your resting rate by an activity factor — which is how nearly every calculator, including the 2BIB TDEE calculator, works — then your usual training is already inside the number. Adding the watch’s figure on top counts the same session twice.
Now stack the measurement error on that. A 400 kcal reported burn that was really 280, eaten back on top of a target that already assumed you trained, can erase most of a moderate deficit on a day you did everything right. It is one of the most common reasons a diet that looks correct on paper produces nothing on the scale — alongside the logging gaps that quietly erase a deficit from the other direction.
The double-count
The overstatement
The moving target
The better setup
What the numbers are still good for
None of this makes a fitness tracker useless. It makes one figure on it unsuitable for arithmetic. Used as a relative signal — this week against last week, this session against the same session a month ago — even a biased estimate is informative, because the bias is roughly constant for you.
- Steps are the honest metric. Directly counted, and a falling step count is one of the most common hidden causes of a stall.
- Resting heart rate trends track fitness and recovery better than any calorie readout.
- Session-to-session comparison works fine: 20% harder than last Tuesday is real information even if the absolute number is not.
- Consistency of training — days per week, minutes per session — predicts results far better than estimated burn does.
How to find your actual expenditure
Infer it from results. Your body is running the experiment continuously, and the scale reports the answer — you just need to hold the inputs still long enough to read it.
- 1Eat to one fixed calorie target for three weeks. Not a target that moves with your watch.
- 2Log completely, including cooking oils, drinks and weekends. An incomplete log makes the whole calculation meaningless.
- 3Weigh daily, compare weekly averages. Daily readings carry far too much water-weight noise to use.
- 4Do the arithmetic. Roughly 7,700 calories per kilogram of body fat: losing 0.5 kg a week on 2,000 calories a day implies you are burning somewhere near 2,550.
- 5Adjust once, then wait. One change every three weeks, not one every Monday.
That figure is worth more than any device output, because it is measured on the only body that matters. It is also the same feedback loop behind setting a calorie target in the first place, and behind why TDEE has to be recalculated as you lose weight.
The short version
Trust your tracker for steps, heart rate and consistency. Do not trust it for calories, and do not build a food target on top of it. Precision is available on the intake side of the equation, where you can read a label, scan a barcode and weigh a portion — so spend your effort there and let three weeks of weight data tell you what you are actually burning. For the wider picture, the fat loss guide covers how the two sides fit together, and our honest look at AI calorie counting applies the same scrutiny to the intake side.