Lose ten percent of your body weight and your muscles will thank you about equally, whether you got there on bacon or broccoli. But your liver kept score — and it is not a tie. In late August in Cell Metabolism, a Washington University feeding trial gave "a calorie is a calorie" its fairest test yet: three diets, every meal supplied, every group losing the same ~10%. Muscle insulin sensitivity rose ~50% everywhere. Then the story broke: hepatic insulin sensitivity improved two to three times more on the ketogenic diet — and the 73%-fat menu produced no worse cholesterol than the heart-healthy ones. Equal on the scale. Not equal in the liver.
The takeaway — if you only read this box
What happened: a fully catered trial — three diets, every meal supplied, every group losing the same ~10% of body weight. Muscle insulin sensitivity improved ~50% in all three groups.
Where they split: the liver. Liver fat fell ~67% on the ketogenic menu vs ~45% on the other two; hepatic insulin sensitivity improved two to three times more; prediabetes resolved in roughly half the keto group, ~29% on Mediterranean, ~7% on plant-forward.
Does keto work? For the liver, under these tightly controlled conditions — clearly. As a lifetime 73%-fat prescription — no, and nobody needs one. The finding is that what your plate contains is a dial that weight-loss medicine has left set to "whatever".
What this is not: not "keto is safe forever" (a five-month trial can't say that), and not "keto lowers cholesterol" (LDL came out equal — which still kills the old scare).
Three menus, every meal supplied
Most diet trials run on trust: people are told what to eat, asked to remember what they ate, and by the end the groups have drifted together. That is how "the best diet is the one you can stick to" became a scientific conclusion. This trial removed the trust.
Petersen, Klein and colleagues (Cell Metabolism, 2026) randomised adults with obesity, prediabetes and fatty liver — the disease package now known as MASLD, metabolic dysfunction-associated steatotic liver disease — to one of three menus: very-low-carbohydrate ketogenic (roughly 73% fat and 4% carbohydrate), Mediterranean (about 50% carbohydrate and 35% fat), or very-low-fat plant-forward (about 70% carbohydrate and 15% fat). The team supplied every meal for roughly four to five months and checked in weekly with a dietitian; adherence was not hoped for, it was catered.
With calories titrated so every group lost the same ~10%, and adherence guaranteed by the kitchen, the only variable left standing was macronutrient composition — precisely the one the "calories are calories" school insists does not matter. They measured what scales cannot: liver fat by MRI, and muscle versus liver insulin sensitivity separately, with two-step clamps and 24-hour blood sampling.
The muscle got the memo. The liver didn't.
Start with the reassuring part: muscle insulin sensitivity improved ~50% in every group — keto, Mediterranean and plant-forward alike. On this endpoint, "any diet works" is simply true.
The liver is where the three diets part company. Hepatic insulin sensitivity — the liver's ability to respond to insulin — improved two to three times more in the very-low-carbohydrate group than in either of the others (p<0.001). MRI-measured liver fat fell about 67% on the ketogenic diet versus about 45% on the other two, the researchers reported. Hepatic de novo lipogenesis — the liver's internal fat factory, run on surplus carbohydrate — fell most in the same group, as did HbA1c and 24-hour glucose and insulin.
Prediabetes resolved in roughly half of the ketogenic group — blood glucose back to normal — against about 29% in the Mediterranean group and about 7% in the plant-forward group, the university reported. Same pounds, same kitchens, same supervision: the difference between 50% and 7% is not willpower or total energy. It is what the plate contained.
The LDL bomb that didn't go off
The ketogenic diet has spent a decade tarred as a saturated-fat bomb; a 73%-fat menu invites the reflex. Here is the result that should annoy the most people: at equal weight loss, there were no between-group differences in LDL cholesterol, apolipoprotein B or 24-hour triglycerides — the 73%-fat group was statistically indistinguishable from the 15%-fat plant-forward group. The measurement was thorough: apolipoprotein B (the particle count that matters more than the cholesterol inside it) and a full day of serial triglycerides, not one fasting sample.
Read that precisely: keto did not lower anyone's LDL — its lipids simply ended up equal to everyone else's, at the same weight loss. The specific fear, that this diet uniquely damages the lipid profile, failed to appear. That is not a lifetime safety certificate. It is a controlled, five-month result.
The organ the dogmas forgot
The "any diet works" doctrine is not a straw man; it is a celebrated trial result. In Gardner's DIETFITS study (JAMA, 2018), 609 adults spent a year on a healthy low-fat or healthy low-carbohydrate diet. Low-fat lost 5.3 kg; low-carb lost 6.0 kg — a 0.7 kg difference (95% CI −0.2 to 1.6), statistically indistinguishable, with no genotype or insulin-secretion interaction. The field's conclusion was fair: the best diet is the one you can stick to. Calories matter; composition does not.
The problem is the endpoint the doctrine was tested on. Body weight is a whole-body average that blends muscle and liver into one number — and the muscle, up ~50% in every group, swamps the ledger. The organ where metabolic obesity actually kills is the liver: a fatty liver is not passive storage. Hepatic insulin resistance keeps fasting glucose high, fat begets inflammation and fibrosis, and MASLD feeds cardiovascular disease. A trial that only weighs bodies cannot see it.
The liver is carbohydrate-sensitive in a way muscle is not: under chronic excess of insulin and glucose it manufactures fat de novo — the very pathway that fell fastest on keto. The dogma was right about the muscle; it was just measured on the wrong organ.
Watch the plate, not just the scale
In the GLP-1 era, the most common route to a ten percent loss is a prescription, carrying an implicit promise: the injection is the intervention, the plate mere garnish. The investigators push back: the drugs deliver the weight loss, but diet composition can add benefits on top.
Weight loss as medicine is established: in DiRECT (The Lancet, 2018), 46% of people with type 2 diabetes were in remission at twelve months on a structured low-calorie programme — against 4% of controls — after losing about 10 kg on average. The new trial adds the layer underneath: at the same deficit, composition decides how much liver-level disease resolves. Drug-scale effect, no prescription required.
So if you are losing weight — by drug or diet — and you carry liver fat or prediabetes, the scale is the wrong instrument: ask what your macronutrients are doing, not just the number. A monitored very-low-carbohydrate period is a legitimate, measurable liver intervention, though long-term responses vary and a clinician should be involved. Keto is not mandatory; the Mediterranean and plant-forward groups still cut liver fat by ~45%. Composition is a dial — currently set to "whatever".
What this trial can't tell you
Now the honesty. This was a five-month trial, and not a large one — a strong signal, not a decade of evidence. No one yet knows whether remissions hold, whether the benefit survives maintenance calories and real-world kitchens, or how much depended on catered meals with a dietitian on call. Long-term LDL responses to very-low-carbohydrate diets vary individually, so this paper licences neither "keto is safe for everyone" nor a lifetime of 73% fat. These were adults with obesity, prediabetes and fatty liver — the results belong to that population, not to healthy livers or athletes.
No one needs to eat 73% fat forever. But next time someone tells you a calorie is a calorie — or that the Mediterranean diet is the only honourable answer — ask which organ they mean. For the muscle, they are right, and that is good news. For the liver, where metabolic disease actually does its damage, the menu matters as much as the deficit, with drug-scale effect sizes, measured under conditions most nutrition research only dreams of. Mechanism-respecting nutrition is not soft science. It is science that finally fed everyone the same meals.
Sources
- Petersen, M.C., Smith, G.I., Farabi, S.S., et al. "Effect of diet macronutrient content on the cardiometabolic response to weight loss: a randomized clinical trial." Cell Metabolism (2026). DOI: 10.1016/j.cmet.2026.07.020 (online 27 Aug 2026).
- Washington University School of Medicine. "Keto diet cut liver fat by 67% in a clinical trial." Press release via ScienceDaily, 28 Aug 2026.
- Gardner, C.D., Trepanowski, J.F., Del Gobbo, L.C., et al. "Effect of Low-Fat vs Low-Carbohydrate Diet on 12-Month Weight Loss in Overweight Adults and the Association With Genotype Pattern or Insulin Secretion: The DIETFITS Randomized Clinical Trial." JAMA 319(7), 667–679 (2018). DOI: 10.1001/jama.2018.0245.
- Lean, M.E.J., Leslie, W.S., Barnes, A.C., et al. "Primary care-led weight management for remission of type 2 diabetes (DiRECT): an open-label, cluster-randomised trial." The Lancet 391(10120), 541–551 (2018). DOI: 10.1016/S0140-6736(17)33102-1.