Low Carb vs Low Fat for Fat Loss: What the Trials Actually Show

TL;DR: This question has been answered. Gardner et al. (2018) randomised 609 people to healthy low fat or healthy low carb for a full year with no calorie target, and the two groups finished 0.7 kg apart, which was not statistically significant. Johnston et al. (2014) pooled 48 trials and 7,286 people and found a 0.02 kg difference at twelve months. Meanwhile the spread inside each group ran from about 30 kg lost to 10 kg gained. Your macro split is not what decides this. Protein intake, total calories and whether you can actually live on the plan are.

---

Every couple of years the internet re-runs this argument and nothing new gets said. So here's the conclusion first, and then the evidence that earns it.

Carbohydrate and fat are close to interchangeable for fat loss once protein and total energy are matched. Not "roughly similar." Statistically indistinguishable across the best-powered trials we have. If you're still choosing between them as though one does something the other can't, you're optimising the wrong variable while the one that matters sits untouched.

The trial that was built to settle it

Christopher Gardner's DIETFITS trial is the closest thing to a definitive answer in this space, and it was run by people who expected a winner.

609 adults, ages 18 to 50, BMI between 28 and 40, randomised to a healthy low-fat or healthy low-carbohydrate diet for twelve months. Twenty-two small-group sessions with health educators. And nobody was given a calorie target. Participants were coached to push their fat or their carbs as low as they could sustainably hold, and to choose whole foods either way.

The groups genuinely separated on macros. At twelve months the low-fat arm was eating 212.9 g of carbohydrate and 57.3 g of fat per day. The low-carb arm was at 132.4 g of carbohydrate and 86.2 g of fat.

The results:

| Group | 12-month weight change | |---|---| | Healthy low fat | 5.3 kg lost (95% CI, 4.7 to 5.9) | | Healthy low carb | 6.0 kg lost (95% CI, 5.4 to 6.6) | | Difference | 0.7 kg, not statistically significant |

Seven hundred grams. Over a year. On two diets that differed by 80 g of carbohydrate a day.

The trial also tested the two most popular personalisation stories of the era: a three-gene insulin-resistance genotype pattern, and baseline insulin secretion. Neither predicted who would do better on which diet. The genotype hypothesis that launched a dozen DNA-diet companies did not survive its own confirmatory trial.

The variance is the real finding

Here's the number that should reframe the whole debate. Within both arms, individual twelve-month results ranged from roughly 30 kg lost to 10 kg gained.

Sit with that. The gap between the best and worst responder inside a single diet group was tens of times larger than the gap between the two diets. Whatever drives fat loss outcomes is operating almost entirely at the individual level, not the macronutrient level.

A 2023 secondary analysis of DIETFITS in the American Journal of Clinical Nutrition went looking for what separated the responders. It found adherence and overall diet quality, not assigned diet class, tracked with weight loss success. The people who did well are the people who stuck to the plan and ate reasonably clean food while doing it. That's the mechanism.

Johnston et al. (2014) arrived at the same place from a different direction. Forty-eight randomised trials, 7,286 participants, every named diet program with usable data. At six months low carb showed 8.73 kg and low fat 7.99 kg. By twelve months they'd converged: 7.25 kg against 7.27 kg. The authors' conclusion was blunt, recommending any diet a patient will adhere to.

Why low carb looks faster for a fortnight

There's a real effect underneath the low-carb hype, and it isn't fat.

Every gram of stored muscle glycogen holds roughly three grams of water. Cut carbohydrate hard and you deplete glycogen within a few days, and the water leaves with it. A trained lifter can shed three or four kilos this way inside a week without losing a meaningful amount of fat tissue.

Kevin Hall's group demonstrated this cleanly in a 2021 Nature Medicine inpatient study. Twenty adults lived in a metabolic ward and ate ad libitum on both a very low-fat plant-based diet and an animal-based ketogenic diet, two weeks each. The ketogenic arm dropped more scale weight. The low-fat arm lost more body fat by DXA. Faster on the scale, slower on the tissue that matters.

That same study produced the finding that annoyed everybody. On the low-fat diet, people voluntarily ate 689 kcal per day less than on the ketogenic diet, and rated their hunger, fullness and satisfaction about the same. Whatever you believe about insulin and appetite, the ward data ran the other way.

So if you start a low-carb cut and the scale plunges in week one, that's glycogen and water, and it comes straight back the day you eat a normal amount of rice. It's the same trap that makes daily scale weight so misleading during a cut.

What changes if you lift

Almost everything above comes from general-population weight-loss research. Add serious resistance training and the picture shifts a little, though not in the direction most people assume.

Protein is the variable that matters, and it matters far more than the carb-versus-fat split. Set it first, defend it, and let the remaining calories fall where they like. For most lifters in a deficit that means roughly 1.6 to 2.2 g per kg of bodyweight, and the reasoning behind that range is in how much protein you actually need per day.

Beyond protein, the evidence on carbohydrate and hypertrophy is thinner than gym lore suggests. Recent systematic reviews find no significant effect of carbohydrate intake on training-induced fat-free mass gains, and most long-term strength comparisons between higher and lower carbohydrate conditions come back null. Ketogenic diets combined with lifting do show a small lean-mass decrement in meta-analysis, though a chunk of that is intramuscular water rather than contractile tissue.

Where carbs earn their place is work capacity on high-volume sessions. Depleted glycogen won't wreck a heavy triple, but it shows up in the back half of a twenty-set leg day. If your training is high volume, keeping carbs up is a performance decision, not a fat-loss one.

How to actually choose

Stop treating this as a physiology question. It's a logistics question, and it has three inputs.

What do you already eat without thinking? If your default meals are rice, oats and fruit, going keto means rebuilding your entire food environment for a 0.7 kg edge that isn't statistically real. If you'd rather eat eggs, meat and olive oil and skip breakfast, forcing 300 g of carbohydrate on yourself is the same mistake in reverse.

What kills your adherence? Some people white-knuckle through hunger fine but can't handle restriction rules. Others need the bright line. Fat is the more satiating macro per calorie for some people and carbohydrate is for others, and no test tells you which you are. Two weeks of honest tracking will.

What does your training demand? High-volume hypertrophy work runs better with carbohydrate available. Low-volume strength work barely notices.

Then set the numbers in this order: protein first, calories second, and the carb-to-fat split last, with whatever's left over. Our macro calculator runs that sequence, and the full logic for structuring a deficit around it is in the cut protocol guide.

The uncomfortable version

A year-long, 609-person, purpose-built trial could not separate these diets. Forty-eight trials and 7,286 people could not separate them. The genotype test that was supposed to tell you which camp you belong to failed its own validation.

What separates people is adherence, and adherence is personal. The diet that works is the one still running in month six, which is almost never the one that felt most impressive in week one.

If you want to know which split is genuinely working for you rather than which one the internet is arguing about this quarter, you need your own data: intake, bodyweight trend and lifting performance in one place, over months rather than weeks. That's the loop Protokl is built around. It adjusts your targets off your measured response instead of a formula, so the macro question answers itself with your numbers instead of somebody else's trial.