Hall 2021: Keto vs Plant-Based — Who Actually Ate More?

The prediction was specific. Carbs spike insulin. Insulin drives hunger. Hunger makes you overeat. Cut the carbs, and appetite handles itself.

Kevin Hall's research group at the NIH designed the most controlled test of that prediction ever attempted. They admitted twenty adults — eleven men, nine women, ages eighteen to thirty-nine, BMI ranging from lean to obese — into a locked metabolic ward for a continuous four weeks.

No trips home. No food from outside. Every meal prepared by NIH nutrition staff, every leftover weighed to the gram.

Each person ate two diets in random order, two weeks each. One was plant-based and low-fat — rice, lentils, vegetables, fruit, with 75% of calories from carbohydrates. The other was animal-based and ketogenic — meat, eggs, cheese, with 75% of calories from fat. Both diets offered as much food as anyone wanted. Eat until you're full. Eat again whenever you're hungry.

Both diets rated equally pleasant. Both rated equally familiar. Nobody preferred one over the other.

Then the intake data came back.

On keto, every single person ate more. Not some of them. Not most. All twenty. On average, the keto side consumed 689 extra calories per day — roughly an entire extra dinner's worth of food, appearing on no plate, registering in no hunger rating. The statistical confidence was as high as it gets.

The diet that was supposed to kill hunger produced an invisible extra meal every single day. And nobody felt any different.

One in a Million

You might think that's just an average — some people ate a little more on keto, some ate a lot more, and it averaged out to 689. But this wasn't a trend. It was unanimous.

All twenty participants ate less on the high-carb side. Every single one. The probability of that happening by chance — of flipping a coin twenty times and getting heads every time — is less than one in a million.

And this held regardless of which diet people started with. The researchers checked for order effects — whether going keto-first or plant-based-first changed the outcome. It didn't. No significant effect of diet order. No significant effect of sex. The direction was the same for everyone, no matter how the study was arranged.

The hunger ratings make it stranger. On a hundred-point scale, participants rated their hunger, fullness, satisfaction, and eating capacity after both diets. Not a single measure showed a significant difference. Identical hunger on both sides — yet one side quietly consumed an extra dinner every day.

The prediction said carbs would drive hunger and overeating through insulin. The insulin part worked exactly as predicted — the high-carb diet tripled insulin levels after meals compared to keto. But the eating went the opposite direction. More insulin. Less food. The mechanism fired. The prediction failed.

What the Scale Was Actually Measuring

Here's where it gets personal for anyone who's ever celebrated a keto weigh-in.

The keto side lost more weight. In two weeks, participants on keto dropped 1.77 kg — nearly four pounds. On the low-fat plant-based side, they lost 1.09 kg. If you stopped at the scale, keto won.

But the researchers didn't stop at the scale. They used DXA scans — the gold standard for separating what the body actually lost into fat versus everything else.

Of that 1.77 kg keto lost, 1.61 kg was fat-free mass — muscle, water, and glycogen. That's 91%. The remaining 0.18 kg of fat loss wasn't even statistically significant. The scale was tracking a loss the body didn't want to keep making.

Meanwhile, the low-fat side barely moved the scale. But underneath, it was burning body fat at more than triple the rate — 51 grams of fat per day compared to keto's 16. The low-fat fat loss was statistically significant. The keto fat loss was not.

If you turned keto's two-week scale result into a pie chart, you'd see a massive slice labeled muscle and water, and a sliver labeled fat that the statistics couldn't distinguish from zero. The weigh-in everyone screenshots was tracking the wrong metric.

More Protein In, More Muscle Out

This is the part that makes even keto-friendly researchers pause.

The keto diet contained more protein — about 135 calories more per day than the plant-based side. In the standard model of muscle preservation, more protein should mean less muscle loss. Protein is the building block. More blocks, more protection.

That's not what happened. The keto side ate more protein and lost more muscle. The low-fat side ate less protein and preserved its fat-free mass almost entirely — a loss of just 0.16 kg that wasn't statistically significant.

The researchers confirmed the protein loss with a second measurement. Twenty-four-hour urinary nitrogen — a direct marker of protein being burned for fuel — showed that the keto side was metabolizing body protein despite consuming more of it in their diet.

In deep ketosis, the body increases a process called gluconeogenesis — it converts amino acids into glucose to keep blood sugar stable. The body was burning protein for fuel, even while more protein was coming in on the plate. The protein-sparing promise that often accompanies keto didn't survive the metabolic ward.

The Three Strongest Objections

If this were a bad article, this is where it would declare keto dead and move on. The study's own authors didn't do that, and neither will this page.

Three objections deserve serious consideration — and the study's own data partially answers each one.

First: two weeks isn't long enough. Keto advocates argue that full adaptation takes four to eight weeks. It's a fair point — and the study data engages with it directly.

By the second week, participants' blood ketone levels had reached 1.8 millimoles per liter — deep nutritional ketosis by any clinical definition. Their calorie intake on keto did drop by 312 calories per day from week one to week two, suggesting adaptation was actively occurring.

But even after that adaptation, the gap remained: 544 fewer calories per day on the plant-based side in the final week alone. A longer trial might narrow the gap. But narrowing a gap is not reversing a direction.

Second: the crossover design may carry effects. Harvard's David Ludwig — one of the carbohydrate-insulin model's most prominent defenders — published a 2025 letter in Nature Medicine arguing that switching diets without a washout period could skew the results.

Hall replied in the same issue, noting no significant diet-order effect was found — the result held regardless of which diet came first. The strongest scientific critics of this study are engaging with it in the field's top journal.

Third: the diets differed in more than just macros. The plant-based side had half the energy density of the keto side — 1.1 versus 2.2 calories per gram. It had more fiber, and it was entirely plant-sourced while the keto side was almost entirely animal-sourced.

Any of these variables could explain part of the intake gap. This is the most technically valid criticism, and the study's own Discussion acknowledges it openly.

But here's the distinction that resolves all three objections at once. The carbohydrate-insulin model made a directional prediction: high-carb food would drive MORE eating through insulin. That specific direction went the wrong way in every participant.

Duration might explain the size of the gap. Confounders might explain the size of the gap. But neither explains why the prediction's direction was wrong. The confounders affect magnitude. The direction was just wrong.

Not Just One Ward

A separate 2025 study tested the same prediction from a different angle. Liu and colleagues at Shenzhen University gave 120 healthy adults meals that deliberately varied in glycemic index — the exact variable the carbohydrate-insulin model says should drive hunger and overeating [1].

Glucose and insulin followed the predicted pattern perfectly — higher-glycemic meals produced higher blood sugar and insulin responses. But hunger didn't follow. Participants reported identical hunger across all three glycemic groups.

And at the individual level, a weak correlation between insulin and subsequent eating went in the opposite direction the model predicts — higher insulin was associated with eating less, not more.

Two different designs. Two different populations. Two different countries. The CIM's hunger prediction failed in the same direction both times.

The Clip and the Evidence

The carbohydrate-insulin model hasn't survived as a scientific consensus — it's survived as a content format. The idea that carbs spike insulin and insulin drives hunger has been compressed into confident sixty-second clips that millions of people have absorbed as metabolic law.

This study didn't test whether keto is a bad diet. Both diets caused weight loss. Both had metabolic advantages the researchers documented — keto lowered fasting triglycerides and reduced glucose variability; the plant-based side lowered LDL cholesterol and improved fat loss.

The study tested one specific prediction made by one specific model — and that prediction went the opposite direction in every person who walked through the ward.

This hunger test was one of nine predictions about carbs that controlled trials have now examined. Which of the model's nine predictions survived and which didn't — from the metabolic edge to GI to fiber — is the question the rest of the evidence answers.

The question was never keto versus plant-based. The question was whether the mechanism behind the clip was real. Twenty people. Identical hunger. Seven hundred extra calories per day on the side that was supposed to suppress appetite. All twenty the same direction.

The clip made a prediction. The ward returned a verdict.

That still leaves the model's other half unanswered. The CIM doesn't just predict that carbs drive hunger — it also predicts that cutting carbs unlocks a metabolic-rate advantage, burning more calories at rest. A different controlled trial tested that exact claim. The gap between what was promised and what was measured is worth knowing.