Calories and eating

Calories eaten versus calories burnt

You set a deficit of 200 a day. Then you walk for an hour and lift for another. Your actual deficit is somewhere near 600, and nothing on your phone told you that.

This is the most common way a sensible plan turns into an aggressive one without anybody deciding to make it aggressive. The weight comes off faster than expected, which feels like winning, and the part that suffers is the part that does not show up on the scale for another month.

Published 2 September 2026. This is a summary of published research, not medical or dietary advice.

In short

13 min read

Your calorie number is an intake target, not a deficit target. Eat the 2,200. The deficit is what the training adds on top of it, and shaving the intake down to 1,900 to try to create one doubles a gap that was already there.
At 80 kg, an intended 200 calorie deficit plus an hour of walking and an hour of lifting is really 585 to 745 on a training day. Three to nearly four times what was chosen.
No device on you can measure this. In the anchor study no wrist wearable achieved an energy expenditure error below 20%, and during weights the estimates barely correlate with reality at all.
Protein first, then carbohydrate. Under restriction the reviews put protein at 2.3 to 3.1 g per kg of fat-free mass. Carbohydrate is what the deficit squeezes hardest and what the sessions are actually spending.
Judge it on the rate over weeks, not on the target. Losing faster than about 1% of body weight a week means the real deficit is bigger than the intended one, and the correction is to eat more rather than to train less.

What that training actually costs

Take an 80 kg adult, an hour of level treadmill walking at 5 to 6 km/h, and a 45 to 60 minute weights session. Here is the whole calculation, with the sources for each line, because the number at the bottom is the entire point of this page.

The 2024 Adult Compendium of Physical Activities, which is the reference table the whole field uses, assigns 3.8 METs to level walking at 4.5 to 5.5 km/h and 4.8 METs at 5.6 to 6.3 km/h. So 5 to 6 km/h straddles the two. On the Compendium’s own convention that is 304 to 384 calories for the hour at 80 kg.

But that is a gross figure, and gross figures include the resting metabolism you would have spent sitting on the sofa. At 80 kg that is about 80 calories an hour whatever you are doing, so the walk only added 224 to 304. Between a fifth and a quarter of the number you were about to celebrate was going to happen anyway.

For lifting, the gap is wider still. Mookerjee, Welikonich and Ratamess measured 24 young adults through a five-exercise session by indirect calorimetry and reported both numbers: 167.9 calories gross, 88.3 net. Roughly half. That is one study on a short upper-body protocol, so treat the ratio as indicative rather than fixed, but the direction is not in doubt, and the reason is obvious once stated. A lot of a weights session is standing between sets.

Scaling that up against the longer sessions in the literature, such as the 289 calories Rustaden and colleagues measured for a heavy-load session in 18 women, a 45 to 60 minute session in an 80 kg adult lands somewhere around 130 to 190 calories net. That is an interpolation between studies with different protocols and populations, and it is published here as a range for that reason. A scoping review of 166 studies found reported values for resistance exercise “varied widely between studies” and that measuring oxygen alone misses the anaerobic contribution anyway.

Add the afterburn, which is real and small. In 13 trained men measured over the two hours after a session, the excess came to 30 to 50 calories. Not the hundreds it is usually sold as.

Training day, 80 kg	Added, net
60 min walking, 3.8 to 4.8 METs	224 to 304
45 to 60 min resistance	130 to 190
Afterburn, two hours	30 to 50
Added to the day	385 to 545
Real deficit, on top of the 200	585 to 745

Three to nearly four times what was intended. And the 2013 American guideline on managing overweight in adults prescribes a deficit of 500 to 750 calories a day under clinical supervision. Which is to say: somebody aiming at 200 has walked into the top of a clinically supervised band, without the supervision, and without knowing they are in it.

Why the machine cannot tell you this

The obvious response is to read the number off the treadmill or the watch and use that. The evidence on this is unusually consistent and unusually damning.

Stanford researchers put 60 adults through sitting, walking, running and cycling wearing seven wrist devices at once, against indirect calorimetry as the criterion. Six of the seven got heart rate within 5% during cycling. Not one achieved an energy expenditure error below 20%, and the error was highest during walking. A later review of 158 publications across nine brands put it flatly: for energy expenditure, no brand was accurate. A 2026 living review of 82 Apple Watch studies covering 430,052 participants found heart rate accuracy good, step counting moderate, and energy expenditure error “inconsistent and frequently large”.

During weights it collapses entirely. Sixty-two men wore four smartwatches through resistance exercise against ECG and indirect calorimetry. Heart rate held up, correlating at 0.64 to 0.97. The calorie estimates correlated at 0.10 to 0.34. That is close enough to no relationship that the number is decorative.

Treadmill consoles are harder to indict, and it is worth being honest about that: there is no adequate modern validation of console displays specifically. What is documented is that the nearest equivalent, an elliptical trainer’s own calculation, overestimated measured expenditure in 26 adults, and that the ACSM walking equation most machines implement overestimated real expenditure by 9.2% at moderate relative workloads and 20.3% at maximal in 103 adults, and got worse after those adults had trained for 24 weeks. Consoles also report gross calories, and many do not ask your body weight at all.

Left to guess unaided, people do worse still. Sixteen adults were blinded to a treadmill bout that indirect calorimetry measured at 200 calories and asked to estimate it. They said 825. Then, invited to eat back exactly what they had burned, they ate two to three times the true cost. The standard deviation on that estimate was enormous, so the exact multiple is soft. The direction has been replicated often enough to build a page on.

And only about three quarters of it counts

Even the calories you genuinely burn do not all land. Pooling doubly labelled water and basal expenditure from 1,754 adults, researchers found energy compensation averaging 28%: about 72% of the extra calories burned in activity showed up as extra calories burned that day. The body quietly reclaims the rest.

Some of that reclaiming is behavioural rather than metabolic. Two thirds of studies find non-exercise movement falls when structured training starts. You sit down more, take the lift, fidget less. A 2026 review comparing additive and constrained models reported that in aerobic exercise interventions total daily expenditure rose by only around 30% of what the additive arithmetic predicted.

Two details from that same review cut in opposite directions and both belong here. Compensation appears reduced with resistance training, so the weights session is closer to genuinely additive than the walk is. And it appears amplified when aerobic exercise is paired with dietary restriction, which is exactly the scenario on this page. The walking is worth less than it looks; the lifting is worth more.

There is a fuller version of this argument on the steps page, including why walking so often moves the scale less than the arithmetic promised.

So what does an accidental 700 actually cost you?

Less than the internet says, and more than nothing. The honest answer requires being careful about what the studies found rather than what they get quoted as finding.

Garthe and colleagues took 24 elite athletes training with weights four times a week and assigned them slow or fast weight loss. Both groups lost about 5.6% of body weight. The slow group, losing 0.7% a week, gained 2.1% lean body mass by DXA. The fast group ended at minus 0.2%, statistically unchanged.

Read that carefully, because it is almost always reported wrongly. The fast group did not lose muscle. It failed to build any while training hard for two months, which is a real cost but a different one. The group also never reached its assigned rate, and there were 24 people in the study. The governing position stand puts it more cautiously than the internet does: fat-free mass and performance may be better preserved below 1% of body weight a week.

Where the effect gets sharper is when protein is low at the same time. Twenty resistance-trained men spent two weeks at 60% of their usual intake. At 1.0 g of protein per kg they lost 1.6 kg of lean mass. At 2.3 g/kg they lost 0.3 kg. Same deficit, five times the tissue loss, decided by one macro. The high-protein group also reported feeling worse, with higher fatigue scores, which is the kind of finding that usually gets left out.

And there is a threshold people quote that deserves a caveat rather than a repetition. Energy availability, meaning intake minus exercise cost divided by fat-free mass, is where sports nutrition looks for trouble. The figure of 30 calories per kg of fat-free mass comes from a 2003 study in which 29 women spent five days at various levels and hormonal pulsatility was undisturbed at 30 but disrupted below it. It is a useful heuristic. It is not a cliff edge, and the 2025 update to the Female Athlete Triad consensus is explicitly moving away from the idea of a threshold at all, in favour of a continuum.

For our 80 kg person at roughly 18% body fat, eating 2,200 calories with a net training cost of 465, that arithmetic gives 26.4. Eating 2,600, it gives 32.7. The entire answer turns on which activity multiplier set “maintenance” in the first place. Which brings us to the real problem.

Every input is wrong by more than the thing you are adjusting

This is the part worth sitting with. Line the errors up against the 200 calories the person believes they are controlling.

Under-reporting of what you ate. 15% on a single 24-hour recall, 28% on a food frequency questionnaire, measured against doubly labelled water across five validation studies. On 2,200 calories that is 330 to 615.
The device's estimate of what you burned. At least 20% on a roughly 500 calorie training day, so 100 or more, with the direction unpredictable. One study found the same watch overestimating in women and underestimating in men.
Compensation you never see. Around 28% of the exercise cost is reclaimed, so a 500 calorie session is worth about 360.
The maintenance figure itself. The convention that one MET equals one calorie per kg per hour overestimated real resting expenditure by 20% across 769 adults. On a 2,400 calorie estimate that is hundreds.

Every single input carries an error larger than the number being adjusted. The conclusion is not that the deficit is unknowable. It is that a deficit cannot be set. It can only be inferred backwards, from what the scale actually did over several weeks.

Which is why the useful reading is the trend, and why the first week tells you almost nothing. If you are dropping faster than about 1% of body weight a week once past the water, your real deficit is bigger than your intended one, and the correction is to eat more rather than to train less.

Which is where the macros stop being optional

A deficit you did not choose is survivable. A deficit you did not choose, with the fuel for two hours of daily training squeezed out of it, is where the training quality goes.

Here is the squeeze, in arithmetic. The position stand puts carbohydrate at 6 to 10 g per kg per day for someone training one to three hours daily, which at 80 kg is 480 to 800 g. Now try to fit that into a real intake. At 2,200 calories, with protein at 2 g/kg and fat at the 20% floor, what is left for carbohydrate is 280 g, or 3.5 g/kg. At 2,600 calories it is 360 g, or 4.5 g/kg.

Both land in the band the position stand describes as appropriate for light activity, for someone training two hours a day. And notice what created that gap. It was not how the macros were divided. It was the size of the deficit. No rearrangement of the plate closes it.

Carbohydrate, with the honest caveat

The usual claim is that low carbohydrate wrecks your lifting. It is weaker than that, and this page will not pretend otherwise.

A systematic review of 49 studies concluded that carbohydrate intake in itself is unlikely to affect strength training performance in a fed state, for sessions up to ten sets per muscle group. Thirteen of 19 acute studies found no benefit. None of the seven short-term manipulation studies found one. Fifteen of 17 long-term studies found training progress uninfluenced by carbohydrate intake. Most pointedly, a trial in 22 trained men matched a carbohydrate breakfast against a viscous placebo for back-squat repetitions and found no difference between them, while both beat water. Some of what people feel is expectation.

So why protect it at all? Because of where the benefit does show up. A meta-analysis of 21 studies found a clear effect on session volume, concentrated in sessions longer than 45 minutes, and moderated by the number of maximal-effort sets rather than by the carbohydrate dose. Two hours of daily training sits exactly on that boundary. Meanwhile a meta-analysis of 20 studies found a single resistance session lowers thigh muscle glycogen by around 104 mmol/kg, and in ten elite weightlifters a high-volume session cut muscle glycogen by 38%, with a substantial fraction of the glycogen sitting between the contractile filaments nearly depleted in fast-twitch fibres.

The fuel is genuinely being spent. Whether spending it costs you performance is conditional, and the conditions are longer sessions, higher set counts, training fasted, and training twice in a day. That is an argument for keeping carbohydrate where you can, not for claiming a decrement nobody has demonstrated in your situation.

Fat, and what the floor is actually for

The position stand recommends 20 to 35% of energy from fat and says intakes of 20% or less do not benefit performance. Read the stated reason, because it is not the one usually given: athletes are discouraged from chronic low fat intakes because the reduction in dietary variety is likely to reduce intakes of fat-soluble vitamins and essential fatty acids. That is a micronutrient argument, not a demonstrated performance threshold.

At 80 kg and 2,200 to 2,600 calories, 20% is 49 to 58 g a day. As for the popular claim that low fat intake damages testosterone, the meta-analysis behind it covers six studies and 206 men, its confidence interval barely excludes no effect, its own authors call for confirmatory trials, no study linked the hormonal change to any performance or body composition outcome, and a correction has since been published. It is not quoted here for those reasons.

Protein, which is the one that is actually load-bearing

A systematic review of resistance-trained athletes under energy restriction put likely requirements at 2.3 to 3.1 g per kg of fat-free mass, scaling upward with leanness and with how severe the restriction is. The denominator matters and is often dropped: that is fat-free mass, not body weight. For our 80 kg person with about 65 kg of it, that is roughly 150 to 200 g a day.

The familiar figure of 1.6 g/kg does not govern this case. It comes from a meta-analysis of 49 studies in people who were not in an energy deficit, it describes supplementation on top of habitual intake, and its breakpoint carries a wide confidence interval. Deficits raise the requirement. That is the whole reason the athlete-specific reviews exist.

Should you eat the exercise back?

Fuel8 does not add a calorie credit when you train, and the evidence for that choice is the first two thirds of this page. The number a device gives you is wrong by at least 20%, is close to uncorrelated with the truth during weights, and is only about 72% real even when it is right. People handed that number eat two to three times what they burned. Adding it to a target systematically over-refunds.

But it would be dishonest to stretch that into “eat the same on a two-hour training day as on a rest day”, because that is not what sports dietetics does. The position stand’s whole framework subtracts exercise cost from intake first and then asks whether what remains is adequate per kg of fat-free mass. It places a floor under intake rather than a credit on top of it. And its carbohydrate targets rise with training load by design.

The defensible position is narrower than either slogan. Do not add a number nobody can measure. Do notice that a hard training day is a poor day to be at your lowest intake, and that the correction is usually smaller than it feels. In the trial that restored menstrual function in exercising women, the increase that did it averaged 330 calories a day.

What to do with all this

Stop treating your target as your deficit. It is your intake. The deficit is whatever it turns out to be after training, compensation and everything you logged approximately, and you will only ever see it in hindsight.

Judge it on the rate, over weeks. Losing faster than roughly 1% of body weight a week, once past the first fortnight, is the signal that the real deficit is larger than the intended one. It is the only instrument here that is not carrying a 20% error.

Protect protein first, then carbohydrate. Protein is the macro with direct experimental evidence for preserving tissue under restriction. Carbohydrate is the one the deficit squeezes hardest and the one your sessions are actually spending. Fat has a floor worth respecting for reasons of dietary variety more than performance.

And if the choice is between eating a bit more and training a bit less, eat. The training is producing the stimulus that decides where the lost weight comes from. The deficit is already larger than you meant it to be. That inference is ours, drawn from the studies above rather than tested directly, and it is the reason this page exists.

Fuel8 is on the App Store.

A food diary for iPhone. It tracks protein, carbs and fat alongside calories, shows a weight trend rather than a daily verdict, never adds exercise calories back, and has no streaks to break.

Download on the App Store

Or leave your name and email for new recipes and what we add next. No more than that.

This form is the website, not the app. Your name and email go to Grow More Services L.L.C-FZ, the company that builds Fuel8, and are stored in Mailchimp so we can email you about new recipes and what we add next. This list is separate from your optional Fuel8 account. Unsubscribe in one click, any time.

Common questions

If I set a 200 calorie deficit and then train, what is my actual deficit?

Much larger. For an 80 kg adult, an hour of level walking at 5 to 6 km/h is 3.8 to 4.8 METs on the 2024 Adult Compendium of Physical Activities, or roughly 304 to 384 gross calories, of which about 80 is the resting metabolism you would have spent sitting still. A 45 to 60 minute resistance session adds somewhere around 130 to 190 calories net, interpolating from the measured studies, and the two-hour afterburn adds 30 to 50 more (An et al., 2026). That is an extra 385 to 545 calories on a training day, so the intended 200 is really something like 585 to 745.

Can I trust the calorie number on my watch or the treadmill?

Not as a quantity. In a Stanford study of 60 adults wearing seven wrist devices against indirect calorimetry, no device achieved an energy expenditure error below 20%, and error was highest during walking (Shcherbina et al., 2017). A review of 158 publications across nine brands concluded that for energy expenditure, no brand was accurate (Fuller et al., 2020). During resistance training it is worse: across 62 men wearing four smartwatches, watch-estimated expenditure correlated with indirect calorimetry at r = 0.10 to 0.34 (Lee et al., 2026). Heart rate on the same devices is good. It is the calorie conversion that fails.

How fast should weight come off if I am also lifting?

The position stand of the American College of Sports Medicine, the Academy of Nutrition and Dietetics and Dietitians of Canada states that fat-free mass and performance may be better preserved in athletes who keep weekly loss below 1% of body weight (Thomas, Erdman & Burke, 2016). In 24 elite athletes training four times a week, losing 0.7% a week was associated with a 2.1% gain in lean body mass, while a faster rate left lean mass unchanged at minus 0.2% (Garthe et al., 2011). Worth being precise about that: the faster group did not lose lean mass, it failed to gain any.

Do I need to hit my carbs to train properly?

Less absolutely than it is usually put. A systematic review of 49 studies concluded that carbohydrate intake in itself is unlikely to affect strength training performance in a fed state for workouts up to ten sets per muscle group, with benefits appearing mainly in fasted subjects and higher-volume sessions (Henselmans et al., 2022). A meta-analysis of 21 studies did find a benefit, but concentrated in sessions longer than 45 minutes (King et al., 2022). Two hours of daily training sits right on that boundary, which is a reason to protect carbohydrate rather than proof that cutting it will hurt you.

How much protein when in a deficit and training?

A systematic review of resistance-trained athletes under energy restriction put likely needs at 2.3 to 3.1 g per kg of fat-free mass per day, scaling up with how lean you are and how hard the restriction is (Helms et al., 2014). Note the denominator is fat-free mass, not body weight. In 20 resistance-trained men on a two-week 40% restriction, 2.3 g/kg of body weight was associated with 0.3 kg of lean mass loss against 1.6 kg at 1.0 g/kg (Mettler, Mitchell & Tipton, 2010). The same study reported higher fatigue ratings in the high-protein group, which is worth knowing.

Should I eat back the calories I burn training?

Not from a device number, because that number is unreliable and only partly real. Pooling doubly labelled water data from 1,754 adults, energy compensation averaged 28%, meaning about 72% of the extra calories burned in activity showed up as extra calories burned that day (Careau et al., 2021). And people asked to eat back a treadmill bout measured at 200 calories ate two to three times its true cost (Willbond et al., 2010). But that is an argument against a credit, not an argument for eating the same on a two-hour training day as on a rest day. Sports dietetics does the opposite: it indexes carbohydrate to the day's training load.

More in this series: what happens when you eat too little, why week one is the fastest week, and how many steps a day.

Sources

This page summarises published research and is not medical or dietary advice. The worked example uses an 80 kg adult to show the arithmetic and is not a prescription. Compendium values are population averages and the paper itself states they do not reflect precise individual expenditure. Studies report group averages and individual results vary. If you have a medical condition, are pregnant, or have any history of disordered eating, speak to a qualified professional before changing how you eat or train.