Eating back exercise calories
The question
When your watch says a workout burned 500 calories, should you eat that many more today?
What the evidence says
Start with the number itself, because it is inflated twice before you ever decide what to do with it. The first inflation is arithmetic. A “500 calorie workout” includes the calories you would have burned sitting on the couch for that hour, roughly 10% to 40% of the gross figure depending on the session, and the standard resting value the tables use is itself about 20% too high (Byrne 2005). Add that gross number on top of a daily target that already covers your resting metabolism and you have counted the same calories twice. The second inflation is the device. A review covering 72 wearables put the average calorie error above 30% for every brand, and it has not improved across device generations. The failure is worst exactly where gym-goers look: when four current watches were tested against lab measurement, they overstated the cost of a lifting session by 83% to 116%, and their numbers were so weakly tied to true expenditure that for lifting they are close to noise. Steady running and walking fare much better, around 10% to 25% error. Heart rate, on the same wrists, is accurate to within 10%. Your watch is a good heart-rate monitor wearing a calorimeter costume.
People then inflate the number a third time in their heads. In the cleanest demonstration, moderately active adults walked on a treadmill until they had burned exactly 200 measured calories, then were asked to eat the equivalent. They estimated 825 calories and ate 557, and their guesses barely tracked what the session had actually cost. In the largest trial of its kind, 198 people over 24 weeks, the entire gap between expected and actual weight loss in the exercise groups came from eating 90 to 125 extra calories a day, not from any failure to burn.
Underneath all this sits a live scientific fight about whether exercise even adds to your daily burn one-for-one. The largest analysis, covering 1,754 adults, found total expenditure rose by only about 72 calories for every 100 spent on activity, with the rest absorbed by a drop somewhere else. A 2026 synthesis of aerobic training programmes put the shortfall far larger. Then a tightly controlled crossover the same year found the opposite: adding 250 calories a day of activity raised total expenditure by 272, full additivity, no suppression at all. There is even a methods paper showing that the statistical technique behind the headline compensation figures can conjure the effect out of data simulated to contain none. Three answers in five years, all in good journals.
What both camps agree on: exercise produces less weight loss than naive arithmetic predicts, and the compensation grows when you are in a calorie deficit. In people whose weight was stable, expenditure rose in step with activity; it flattened only in those losing weight. Compensation looks more like a response to under-eating than a response to exercise.
Whatever compensation exists, it is slow. The mechanisms with any evidence behind them are gradual: resting metabolism drifting down over weeks, organ mass changing, movement becoming more efficient. A 10-day study found full additivity; a 12-week one found resting metabolism falling; at 24 weeks the shortfall came from intake instead. Nobody has ever measured the expenditure response to one unplanned workout, because the gold-standard method averages over one to two weeks and cannot resolve a single day. So a single session probably does add close to its net cost. That is the best-supported reading, though it rests on mechanism and the absence of contrary evidence rather than on a direct measurement.
Which leaves the practical question: does eating those calories back on the day you trained do anything for you? No trial has tested it. The within-day fueling literature that gets cited in its favour is four or five cross-sectional studies of elite athletes, 25 to 31 people each, and none controlled for total intake, so an athlete who eats less mechanically spends more hours in deficit and the finding chases its own tail. Trials that shuffled a fixed weekly calorie total across days, through refeeds and diet breaks, consistently found no difference in body composition. Day-level matching earns its place in a narrow population: endurance training beyond roughly 10 hours a week, physique and weight-class sports, contest prep, female athletes ahead of male. For a recreational exerciser doing three to five moderate sessions, a 400 calorie workout is a small slice of the day, and where you land against any threshold that matters is set by what you eat over the week, not by which day the food arrives.
Key studies
| Study | Year | Design | n | Finding |
|---|---|---|---|---|
| Careau et al., Curr Biol | 2021 | Cross-sectional, doubly labelled water | 1,754 | Total expenditure rises 72.3% of the additive prediction (95% CI 0.626 to 0.820); compensation reaches 49% at the 90th BMI percentile |
| Yegian et al., J Physiol | 2026 | Crossover, 10-day blocks, DLW + calorimetry | 12 (+268) | +250 kcal/d of activity gave +272 kcal/d total, no resting suppression. Full additivity |
| Pontzer & Trexler, Curr Biol | 2026 | Framework + synthesis | many | Aerobic interventions raise total expenditure by only ~30% of prediction; compensation grows with intervention duration (p = 0.026) |
| Gonzalez et al., Adv Nutr | 2023 | Simulation + reanalysis | — | The derived-variable method produces apparent compensation from independent data; with measured variables the relationship is flat (r = 0.04) |
| Willis et al., MSSE | 2021 | Cross-sectional, DLW + accelerometry | 584 | Additive in stable or positive energy balance; flat only under deficit. No curvature overall (p = 0.920) |
| Martin et al. (E-MECHANIC), AJCN | 2019 | RCT, 24 weeks, DLW | 198 | The weight-loss shortfall came entirely from intake (+90.7 and +123.6 kcal/d), not from suppressed expenditure |
| Willbond et al., J Sports Med Phys Fitness | 2010 | Blinded randomised crossover, indirect calorimetry | 16 | For a measured 200 kcal session, participants estimated 825 kcal and ate 557 kcal |
| Fahrenholtz et al., Scand J Med Sci Sports | 2018 | Cross-sectional, 7-day records | 25 | More hours in energy deficit in athletes with menstrual dysfunction. Total intake not controlled for |
| Loucks & Thuma, JCEM | 2003 | Randomised crossover, controlled feeding, 5 days | 29 | LH pulsatility unaffected at 30 kcal/kg FFM/d, disrupted below. The origin of the “30” threshold, from a 5-day exposure in sedentary women |
| Siedler et al., J Hum Kinet | 2023 | RCT, continuous vs interrupted deficit | 38 | No difference in body composition or resting metabolism from redistributing the same deficit |
| Lee et al., Sensors | 2026 | 4 watches vs indirect calorimetry | 62 | Resistance training: Apple +112%, Garmin +116%, Samsung +83%. Correlation with truth r = 0.10 to 0.34, ICC 0.14. Heart rate on the same protocol: ICC 0.95 |
| Germini et al., JMIR | 2022 | Systematic review, 72 devices | — | Energy expenditure error above 30% for every brand. Heart rate under 10% |
| Byrne et al., J Appl Physiol | 2005 | Ventilated-hood calorimetry | 769 | Measured resting rate is 0.84 kcal/kg/h, not the conventional 1.0. The standard convention overstates resting expenditure by about 20% |
| Kozey et al., J Phys Act Health | 2010 | Indirect calorimetry, 2,555 activities | 252 | Standard MET values misclassify intensity 12.2% of the time, worst in low-fit, overweight and older people |
| Riou et al., Nutrients | 2015 | Systematic review | 928 | Mean compensation 18% (SD 93%), approaching 84% in interventions around 80 weeks |
| Hahn et al., J Eat Disord | 2024 | Longitudinal, adjusted for baseline | 138 | Using a tracking app for weight management predicted increased disordered weight-control behaviours (β = 0.894, p = 0.012) |
Confidence rating and why
Moderate. The strongest parts are either definitional or repeatedly measured: gross session calories contain resting metabolism and cannot be stacked on a daily baseline without double-counting, wearable calorie figures carry over 30% error in every published review, and people reliably overestimate what they burned while underestimating what they ate back. That compensation exists and grows with time and energy deficit is agreed by both sides of the field. What caps the rating is that the two numbers you would most want are the two the literature cannot supply: the size of the compensation effect, where three good papers give three different answers and the leading method has a demonstrated statistical artifact, and the response to a single workout, which no study has ever measured directly.
What to do with this
Do not eat back the calories your watch reports, especially not from a lifting session, where the figure is essentially uninformative. Set your calorie target from your overall training pattern across the week, then let individual sessions ride.
If a week of hard sessions leaves you genuinely hungrier, honour that with food. Appetite integrated over days is a better sensor than the wrist number.
For steady running, cycling or walking the device figure is a fair rough gauge, still likely 10% to 25% off, so treat it as context rather than a food budget. Trust the heart-rate data; that part the watch does well.
The people who genuinely need to match intake to training day by day are endurance athletes past roughly 10 hours a week, physique and weight-class competitors, and female athletes in heavy training. If that is not you, your weekly average decides the outcome, and the watch’s calorie line is the least reliable number it shows you.
What we do NOT claim
- That exercise doesn’t burn calories or doesn’t help. Total expenditure rises with activity. It just rises by less than arithmetic predicts, and by much less than your watch says.
- That we know the size of the compensation effect. Estimates run from zero to around 70%, and the honest position is a wide range.
- That the watch number is useless for everything. Heart rate is accurate, and steady-state cardio estimates are usable as a rough guide. The lifting estimate is not.
- That a single under-fuelled training day causes harm. The threshold everyone cites comes from studies where women were underfed for four to five consecutive days.
- That eating more on a training day helps a recreational exerciser. No trial has tested it against a steady intake with the weekly total held equal.