Should You Eat Back Exercise Calories? Why Apps Fail
Your wearable's calorie estimate is deeply flawed, but blindly adding those calories back to your plate is even worse for your training.
~6 MIN · KEXBI RESEARCH
Your wearable's calorie estimate is deeply flawed, but blindly adding those calories back to your plate is even worse for your training.
You finish a hard hour on the trails. The watch flashes a number: 850 calories burned. You open your food app and it quietly does you a favour, raising today's target by roughly that amount. Dinner just got bigger. It feels earned.
It should not. Of every number a wearable produces, the calorie figure for a workout is the one it is worst at, and handing it back to you as extra food is the least defensible thing a nutrition app can do with it. The evidence for that is not controversial. What the honest alternative looks like is where it gets interesting.
Connect a watch to a calorie app and, in the naive version, it takes the burn your device reports for a session and credits it back to your daily allowance so you can eat to match. It is intuitive. Calories in, calories out. You spent some, so you can afford them.
To be accurate about the category, though, not every app does this, and the differences are the tell.
| App | Add-back stance | What that means for the day |
|---|---|---|
| MyFitnessPal | Full add-back, by design | Sync a 500 kcal workout against a 2,000 kcal goal, target rises to 2,500. Worked example from MFP's own help centre. |
| Lose It | Conditional add-back | Adds back only once your total daily burn passes a target threshold. Same mechanism, delayed. |
| Cronometer | Above-baseline only | Credits only expenditure above your baseline for the session hour, so the same hour is not counted twice. Cleaner bookkeeping. |
| Noom | 50 percent add-back | Halves the session credit deliberately, explicitly to stop users overcompensating with food. A population-level guard. |
| MacroFactor | Zero add-back | Refuses to add exercise calories at all. Learns your energy needs from the relationship between intake and weight trend. |
| KEXBI | Verify, split, protect EA | Read the number in context, reconcile against KEXBI's own metabolic model, then keep two numbers: undiscounted for Energy Availability, compensation-adjusted to check today's actual against the plan you already had. |
The apps closest to defensible got there by distrusting the workout number. The question is why.
Your device tracks your heart rate well and turns it into calories badly. Those are different tasks, and wearables are good at the first and poor at the second, because converting heartbeats into energy depends on physiology the sensor cannot see: your efficiency, your training status, your real body composition.
One honesty point that makes the argument stronger, not weaker: the error does not always run high. Depending on the device, the person and the activity, a watch can over-read or under-read. Some popular devices tend to overestimate, which is the case that inflates your dinner, but the deeper truth is simply that the figure is not precise enough to spend food against. An add-back model bakes that imprecision straight into your target.
There is a bookkeeping error hiding underneath the accuracy problem. Some workout-calorie figures are "gross": they include the resting metabolism your body was always going to burn during that hour just to keep you alive. By convention, resting metabolism is about one MET, roughly one calorie per kilogram of body weight per hour (Compendium of Physical Activities, Ainsworth and colleagues 2011, updated 2024) [4]. Your daily calorie target already includes your resting metabolism across the whole day. If an app adds a gross workout figure on top, it counts that resting slice twice.
Here is the nuance the better engines get right and the naive ones miss: it depends entirely on which number you were handed. Apple Health's Active Energy is defined as activity energy above your basal rate, so on paper it is not the gross figure the double-count story assumes. Garmin's "Activity Calories" can include resting energy, while its "Active Calories" do not. A value passed through Strava can be the platform's own estimate or a partner device's, with no single convention.
The uncomfortable part is that the definition does not settle it, because a field is only as trustworthy as the app writing into it. Apple Health is a pipe, not a source. One person's Health library can hold sessions written by Garmin, WHOOP, Oura, a treadmill app and the Watch itself, all landing in the same field, and they do not agree with each other. We tested 544 sessions from our own users against an independent metabolic model, holding out the activity types where such models are known to be unreliable, and the writers spread across the range: one treadmill app read clearly gross, while the single largest writer of real training data landed squarely between the two conventions and could not be classified either way.
So the correct thing to remove is not always "resting metabolism"; it is whatever expenditure your daily plan already assumed for the time you were training. Establishing that means checking each source against physiology rather than trusting a field name, and accepting that for some sources the honest answer is still "we do not know yet." Almost no consumer app does any of this. It is the difference between accounting and guessing.
Even a correct net figure may overstate what a session truly costs you, because humans do not appear to burn extra energy in simple proportion to activity. Spend more on exercise and the body may spend less elsewhere, through lower unconscious movement and metabolic adjustment. The largest dataset on this, Careau and colleagues 2021, analysed 1,754 adults from the international doubly-labelled-water database and estimated that people offset about 28 percent of activity energy on average, with the effect larger in people carrying more body fat [5]. It builds on Pontzer's constrained model of daily energy expenditure [6].
This is the reason to hedge, and to hedge openly. As of 2026 the science is actively contested. A doubly-labelled-water study of 75 adults (Howard and colleagues, PNAS 2025) found an additive relationship, with more activity simply meaning more total burn and no drop in resting metabolism [7]. A short longitudinal experiment (Yegian and colleagues 2026) found much the same [8]. The Careau figure is a population average from a cross-sectional analysis, not a licence to subtract a fixed 28 percent from your Tuesday run.
So compensation belongs in the case as a real and important phenomenon that gives you a third reason not to treat the watch number as truth, and it does not belong as a precise correction anyone can apply per workout. The two reasons above stand on their own regardless of how this debate settles.
Now flip the user, because this is the part most coverage misses. Take an athlete who is fuelling, not dieting. The identical inaccuracy does the opposite damage.
Energy Availability, the number that governs whether training becomes progress, is your intake minus the energy training cost, measured against your lean mass [9,10]. If a system trusts an inflated device figure and treats it as energy spent, it manufactures a shortfall that is not real. A well-fuelled athlete can be flagged as under-fuelled on paper, or a genuine shortfall can be read as a crisis.
Chronically low Energy Availability is the state that blunts muscle protein synthesis, stalls adaptation and slows recovery [10,11]. Point the wrong number one way and it feeds a dieter calories they did not spend. Point it the other way and it invents a deficit that eats an athlete's progress. Same session, same faulty source, two ways to get the day wrong.
KEXBI treats the device figure as an observation to be checked, not an instruction to be obeyed. Three things happen to a workout before it reaches your plan.
First, it is read in context. KEXBI records which app wrote each number, not just which service it arrived through. That matters because a known data fault, like a cycling power figure misread into the wrong units, is caught rather than passed on. It is also what makes reason two answerable at all: you cannot decide whether to subtract resting metabolism from a figure until you know which app produced it. That per-writer record is in place today, and the same subtraction is applied to every writer for now; when the evidence in reason two hardens on who emits gross versus net, the wiring differentiates them without a rewrite.
Second, that figure is reconciled against KEXBI's own model of the session, built from the activity type, your effective body mass and the duration using published metabolic values. If device and model roughly agree, the measured value is kept for the extra signal it carries. If the device figure is implausible, the kind of wild over-read a watch produces for a heavy lifting session, the model takes over. Neither is treated as gospel; the point is to catch the estimates that are clearly wrong before they touch your day.
Third, the honest number is used to protect your fuelling, not to license eating more. That turns out to need two numbers rather than one, because the same session is being asked two different questions.
The first question is safety. What did your body actually spend above rest? Energy Availability is scored on that undiscounted figure, because the floors it is judged against were established on undiscounted exercise energy. Shrinking a number before comparing it to a published threshold would make a truly low day look acceptable.
The second question is the plan. Your fuel plan is not built from today's watch number: it was built a week ago from the session you had planned. The compensation-adjusted figure is what today's actual gets reconciled against that plan. If the mismatch is meaningful, KEXBI offers you a rebuild. The compensation discount lives here, on a plan check, not on the safety threshold, because applying a contested population average to a safety threshold would be exactly the overreach reason three warns against.
A note on honesty, since it is the whole point of the piece: KEXBI's checked figure is a better-founded estimate, not a claim to have measured the one true calorie cost of your run, which no consumer device can do. What it can do is refuse to launder an unreliable number into either a bigger plate or a fake deficit, and keep correcting against the outcomes that actually matter for you.
The thing to hold onto is that none of these apps is built to protect Energy Availability. Every one is, at its core, a weight instrument, and that is the lens that sorts them.
MacroFactor is the most sophisticated and the closest to defensible. It refuses to add exercise calories back at all, and instead learns your energy needs from your intake and weight trend, which sidesteps the trust-the-watch problem entirely. For general weight management that is a good tool.
For an endurance athlete it is the wrong instrument. A weight trend lags exactly when the risk is highest: ramp your mileage for an event and your needs jump at once while the model takes weeks to catch up, so you under-fuel at peak demand. And weight is decoupled from Energy Availability, so a tool that only watches the scale is blind to the very state it most needs to see. You can hold a stable weight while deep in low Energy Availability, and the model can even deepen it, reading a suppressed metabolism as lower expenditure and trimming your target again.
Noom is honest about the problem and blunt in its answer: it credits back only half of a session, explicitly to stop people overeating on a watch estimate. For a dieter that is a fair guard. Apply the endurance lens, though, and it is the same category error at a smaller size. Fifty percent is a population guess, not calibrated to you or to what the session actually cost, and halving real training energy is a quick route to under-fuelling. Right instinct, wrong frame.
Cronometer does the cleanest bookkeeping of the add-back apps, crediting only expenditure above your baseline so the same hour is not counted twice, and its micronutrient depth is real, but it still hands the credit back as food and still measures nothing about Energy Availability.
MyFitnessPal and Lose It sit at the naive end, full add-back that lifts your food target whenever the watch claims a big burn, with all the dieter licensing and athlete blindness that implies.
KEXBI's distinct place is simple to state: it is built to fuel training, not just to predict weight, so it treats the workout number as a fuelling input to be verified and protected around, rather than a reward to be paid out.
"Eat back your exercise calories" sounds like fairness. It rests on trusting the least trustworthy number your watch produces, often too high, sometimes double-counting the metabolism you would have spent anyway, for a session your body may partly offset. For someone losing weight, it quietly refills the deficit they are working to hold. For someone training, the same wrong number, pointed the other way, invents a shortfall that costs them the adaptation they are chasing.
The honest version is not complicated to want, only to build: work out what the session plausibly cost, check it against the physiology, know where the number came from, and use it to make sure you are fuelled for what comes next, rather than to justify a bigger plate. That is the number worth counting, and it is the one KEXBI is built to get right.