Diet · Patterns
Two different questions get argued as if they were one
One question is what a diet does to a body when the amount of food is held equal. The other is what happens when people are handed the rules and left to live. Those have different answers, and almost every argument about keto against vegan against carnivore is two people answering different ones at each other.
The Archive’s position
This is opinion, not a finding. Everything else on this page is the evidence.
Which diet pattern you pick matters far less than whether you can keep eating it, and far less than whether it gets enough protein into you. Choose on the basis of what you will still be doing in a year, not on a mechanism.
Why we hold it
- The trial that measured both variables directly found adherence correlated with weight loss at 0.60 and diet type at 0.07. Four diets built on contradictory theories all worked about equally, and how well people stuck to them explained nearly everything.
- When calories are matched in a metabolic ward, the differences shrink to almost nothing. Across 32 controlled feeding studies the metabolic advantage went to lower fat rather than lower carbohydrate, and it amounted to 16 grams of fat a day, which is a real measurement of a trivial effect.
- When people are overfed at 5, 15 or 25% protein, fat gain is identical. What differs is lean mass. That is the same answer the protein literature gives from the other direction, and it points at the same variable.
- The two most serious attempts to predict who suits which diet, by genotype and by insulin secretion, both failed in a 609 person trial designed specifically to test them.
- Every free-living pattern that produced weight loss did it by reducing how much people ate, wherever intake was actually measured. The patterns are delivery mechanisms for eating less, and they differ in how pleasant that is to live with.
Where this goes beyond the evidence
Equivalent for weight is not equivalent for everything, and this position is about body weight and body composition rather than health. At matched calories the patterns do diverge consistently on blood lipids, and they diverge in opposite directions: carbohydrate restriction lowers triglycerides and raises HDL while raising LDL, and fat restriction does the reverse. Those are real differences in measured risk markers and nothing here resolves what they mean for any individual. The long-term outcome evidence is also mostly observational, which means it is confounded by everyone who chooses a diet being different from everyone who does not, and where two large cohorts have examined the same question they have sometimes disagreed outright. Saying the pattern barely matters for the scale is well supported. Saying it barely matters is not, and we are not saying it.
Who this is not for
Anyone with a condition where a specific pattern is a treatment rather than a preference, including diagnosed coeliac disease, established chronic kidney disease, or epilepsy managed on a ketogenic protocol. It also does not apply to anyone whose reason for choosing a diet is not body composition at all.
How to read any diet study in one table
Before the result, find out which of these four the trial was. It changes what the number means more than anything else in the paper, and it is usually in the methods rather than the headline.
- controlled
- Energy intake was matched between groups, so a difference in the result is attributable to the composition of the diet rather than to how much was eaten.
- ad libitum
- Both groups ate freely. A difference here is the diet and its effect on appetite and adherence combined, which is a fair question but a different one.
- prescribed, self-reported
- Participants were told what to eat and reported back. Under-reporting is the norm rather than the exception, so the actual intake in each arm is unknown.
- not applicable, observational
- Nobody assigned anything. People who choose a diet differ from people who do not in ways that are not fully adjustable, so this shows association and not effect.
What the controlled trials found
Weight loss tracked adherence, not which diet it was
Measured in
160 overweight adults with hypertension, dyslipidaemia or high fasting glucose, randomised to Atkins, Zone, Weight Watchers or Ornish for a year
Four diets built on incompatible theories produced between 2.1 and 3.3 kg of loss at one year, none clearly ahead. Weight lost correlated with self-reported adherence at r = 0.60 and with which diet a person was assigned at r = 0.07. Every diet cut the LDL to HDL ratio by around 10%.
HumanComparison of the Atkins, Ornish, Weight Watchers, and Zone diets for weight loss and heart disease risk reduction: a randomized trial (2005)What this does not show
That the diets are interchangeable for every purpose. Adherence was self-reported, which is a weak measure, and a one-year trial cannot speak to what any of these does across decades.
Overfed on 5%, 15% or 25% protein, everyone gained identical fat
Measured in
25 healthy weight-stable adults living in an inpatient metabolic unit for 10 to 12 weeks, overfed by about 954 kcal/day for the final 8 weeks, body composition by DXA
Total weight gain differed a great deal, 3.16 kg on low protein against 6.05 and 6.51 kg. Body fat gain did not: it rose similarly in all three groups. The difference in scale weight was lean tissue, which the low protein group failed to add at all. The authors put it flatly: calories alone account for the increase in fat.
HumanEffect of dietary protein content on weight gain, energy expenditure, and body composition during overeating: a randomized controlled trial (2012)What this does not show
What happens in a deficit, which is the direction almost everybody cares about. This was overfeeding, in 25 people, under conditions nobody lives in.
Across 32 controlled feeding studies, cutting fat beat cutting carbohydrate
Measured in
A meta-analysis of 32 controlled feeding studies in which carbohydrate was swapped for fat at matched calories
Both energy expenditure, by 26 kcal/day, and fat loss, by 16 g/day, were greater on the lower fat diets. The direction is opposite to the popular claim that carbohydrate restriction confers a metabolic advantage.
ReviewObesity energetics: body weight regulation and the effects of diet composition (2017)What this does not show
Anything worth acting on in either direction. Sixteen grams of fat a day is a rounding error next to the differences adherence produces, and the honest reading is that the metabolic advantage is real, measurable, and far too small to matter.
Low fat against low carb for a year, and the two groups ended up eating the same
Measured in
609 adults with a BMI of 28 to 40, randomised to a healthy low-fat or healthy low-carbohydrate diet for 12 months, with 22 group sessions and no calorie target given
Weight change was 5.3 kg against 6.0 kg, a difference of 0.7 kg with a confidence interval crossing zero. The pre-registered attempts to find who each diet suits both failed: no interaction with a three-SNP genotype pattern (p = 0.20) and none with insulin secretion (p = 0.47).
HumanEffect 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 (2018)What this does not show
That personalisation is impossible, only that these two specific ways of attempting it did not work. Both arms were also taught to eat whole foods and cut added sugar and refined grain, so this compares two good diets rather than a good one against a bad one.
One study to stop citing
KETO-CTA followed 100 people whose LDL cholesterol rose sharply on a ketogenic diet and reported that plaque progression tracked their existing plaque rather than their ApoB. It was published in JACC Advances in 2025 under the title “Plaque Predicts Plaque, ApoB Does Not”, and it spread quickly because it appeared to answer the strongest objection to high-LDL ketogenic diets.
It was retracted in 2026, at the request of both the authors and the editors, after concerns were raised about the methodology that affected the reliability of the data. The notice states the errors were too great to fix with a correction. PubMed now flags the original as a retracted publication.
It is still cited. That is the practical reason this page exists in the form it does: a result that confirms what an audience already believes travels much further than the notice withdrawing it, and checking whether a paper still stands takes about thirty seconds. The retraction notice is here.
The individual patterns are being written
Keto, carnivore, vegan, Mediterranean, paleo and the whole against ultra-processed question each get their own page, with the calorie-controlled trials reported separately from the free-living ones and every citation checked against the source first. They are not published yet, and this page says so rather than showing you a thin version of them.
Questions people ask about this
- Is keto better than other diets for fat loss?
- Not when calories are held equal. Across 32 controlled feeding studies the small metabolic edge actually went to lower fat rather than lower carbohydrate, and it amounted to about 16 grams of fat a day. In free-living conditions keto often does produce more weight loss, because for some people it suppresses appetite and is easy to follow, which is a real advantage and a different one from the metabolic claim usually made for it.
- Why does the scale drop so fast in the first week of a low-carb diet?
- Glycogen binds water, roughly three grams of water per gram of glycogen, so depleting it drops scale weight quickly without a matching change in fat. Controlled trials that measured both have found carbohydrate restriction producing more scale weight loss while fat restriction produced more actual fat loss over the same period.
- Does carnivore have trial evidence?
- No randomised trial of a carnivore diet exists. The published human evidence is a self-selected survey of people who chose the diet and stayed on it, which cannot separate the diet from the people who suit it. That is a statement about the evidence, not a verdict on the diet.
- Can I find out which diet suits my genetics?
- The best-designed attempt to do this failed. DIETFITS pre-specified both a three-SNP genotype pattern and insulin secretion as predictors of who would do better on low fat or low carbohydrate, in 609 people over a year, and neither predicted anything. Commercial tests making this claim are ahead of the evidence.
- So does the pattern not matter at all?
- It matters for things other than the scale. At matched calories the patterns move blood lipids consistently and in opposite directions, and it matters enormously for whether a person can actually live on it. What the trials undercut is the specific claim that one pattern melts fat faster than another at the same calorie intake.