Does ultra-processed food make you eat more?

In a metabolic ward where food was provided and intake measured, people ate 508 kcal a day more on an ultra-processed diet and gained 0.9 kg in two weeks. That is the most reliable finding in this area. The same paper shows why the simple reading is wrong: the energy density of what was actually eaten was not matched, and the authors say the 85 percent higher energy density of the non-beverage foods likely contributed.

Is it the processing itself?

The trial's own data points at energy density and eating rate instead. People ate 17 kcal/min faster on the ultra-processed diet, and reported no difference at all in how pleasant or satisfying the food was, which undercuts the hyperpalatability story. The authors state that no causal relationship between ultra-processed food and human obesity has yet been established.

So can a diet high in ultra-processed food be healthy?

The 2025 trial is the best answer anyone has. When both an ultra-processed and a minimally processed diet were built to follow national dietary guidelines, both produced weight loss and minimal processing won by about one percent of bodyweight over eight weeks. That suggests composition matters far more than processing status, which is useful if cooking from scratch is not realistic for you.

Why do the two trials disagree?

They do not, quite. In 2019 a typical ultra-processed diet produced weight gain. In 2025 a guideline-compliant ultra-processed diet produced weight loss. Together they say that what is in the food drives most of the effect and that ultra-processed is a category broad enough to contain both.

Whole-food2 human studies cited

Whole food versus ultra-processed

Q: What about the studies linking it to diabetes?

They are cohorts: 14 of them, 692,508 people, hazard ratio 1.24 comparing highest with lowest intake. The direction is consistent and the confounding is severe, because people who eat the most ultra-processed food differ in income, education, time and cooking facilities, none of which adjustment fully removes. The cohorts give length where the trials give control, and neither gives both.

Written by Aaron CuhaReviewed Sep 2026

Also called: NOVA, UPF, Minimally processed

The landmark ward trial found 508 kcal a day more eaten on ultra-processed food. Its own text shows the mechanism was energy density and eating rate rather than processing. A 2025 replication where both diets followed dietary guidelines shrank the gap to one percent of bodyweight.

What it actually is

Not a diet with rules, but a claim about which variable matters. The argument is that degree of industrial processing, rather than macronutrient ratio, explains most of what diet trials find. Everything here depends on the NOVA classification, which is a published definition rather than a measured quantity, so the definition is a load-bearing assumption in every study below.

The mechanism its advocates propose

Industrial processing produces food that is energy dense, soft, quick to eat and engineered for palatability, which defeats the satiety signals that regulate intake. The strong version says processing itself is the cause. The weak version says processing correlates with energy density, eating rate and nutrient profile, which are the actual causes. Those are different claims and the best trial distinguishes between them.

What happened when calories were controlled

Strongest design available: 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.

Calories were deliberately not controlled, because intake is the outcome. What was matched was what participants were offered. The important finding is that the matching was incomplete in exactly the way that matters.

Hall 2019 admitted 20 weight-stable adults to the NIH Clinical Center and randomised them to two weeks ultra-processed then two weeks unprocessed, or the reverse, eating as much as they wanted. Meals were designed to match presented calories, energy density, macronutrients, sugar, sodium and fibre. Participants ate 508 kcal/day more on ultra-processed (p = 0.0001), gained 0.9 kg on it and lost 0.9 kg on the unprocessed diet.
The same paper undercuts the simple reading. Energy density of what was actually consumed was not matched: 1.36 against 1.09 kcal/g (p < 0.0001). The presented meals matched only because beverages carried dissolved fibre supplements, and the authors state that the roughly 85 percent higher energy density of the non-beverage foods likely contributed to the excess intake.
Palatability was not different either. Participants reported no significant difference in pleasantness (p = 0.13), familiarity (p = 0.57), satisfaction (p = 0.42) or capacity to eat (p = 0.25). The hyperpalatability explanation is not supported by the subjective measures in the trial most often cited for it.
Eating rate was different: 17 kcal/min and 7.4 g/min faster on ultra-processed (both p < 0.0001), correlating with the intake difference (r = 0.45, p = 0.047). So the mechanism this trial supports is closer to energy density and eating rate of non-beverage foods than to processing as such. The authors write that no causal relationship between ultra-processed food consumption and human obesity has yet been established.
Dicken 2025 is the replication that reframes everything: 55 adults, two eight-week ad libitum diets in crossover, both built to follow the UK Eatwell Guide, one minimally processed and one ultra-processed. Both produced weight loss. Minimally processed minus 2.06 percent of bodyweight, ultra-processed minus 1.05 percent, difference 1.01 percent (95% CI 0.14 to 1.87, p = 0.024, Cohen's d minus 0.48).

What happened when people just ate it

The observational layer is large and consistent in direction. A 2025 systematic review pooled 14 prospective cohorts covering 692,508 participants and found a hazard ratio of 1.24 for incident diabetes comparing highest against lowest ultra-processed consumption, with 1.13 per 10 percent increase in intake. Heterogeneity was substantial. These are cohorts, so they carry the usual problem in an acute form: people who eat the most ultra-processed food differ from those who eat the least in income, education, time, cooking facilities and much else, and no adjustment set fixes that. The reason this layer still matters is that the randomised trials are two and eight weeks long and the cohorts run for years.

Protein, and whether it confounds the result

Protein is the quiet finding in the ward trial and it is worth extracting. The 508 kcal/day excess came from carbohydrate (plus 280 kcal/day) and fat (plus 230 kcal/day) and not at all from protein (minus 2 kcal/day, p = 0.85). Protein intake was effectively identical on both diets. That fits the protein leverage idea, that people eat until a protein target is met and consume more total energy when protein is dilute, and the authors themselves suggested future studies should match diets more closely for protein and non-beverage energy density.

The measured intakes behind the protein figures are on the protein page.

What reliably moves

Marker	Direction	From
Energy intake	Up by about 508 kcal a day on ultra-processed	The headline finding, from a metabolic ward with food provided and intake measured. The most reliable number in this block.
Body weight	Up 0.9 kg in two weeks on ultra-processed, down 0.9 kg on unprocessed	Weight change correlated with energy intake at r = 0.8. In the 2025 replication where both diets followed dietary guidelines, both arms lost weight.
Body fat mass	Up 0.4 kg on ultra-processed, down 0.3 kg on unprocessed	From the full text. The larger fat free mass swings are attributed by the authors to fluid shifts, with sodium intake differences correlating with fat free mass change (r = 0.63, p = 0.004).
Eating rate	Faster on ultra-processed	Plus 17 kcal/min and 7.4 g/min, correlating with the intake difference. This, with energy density, is what the trial's own data supports as the mechanism.
Reported palatability	No difference	Pleasantness p = 0.13, familiarity p = 0.57, satisfaction p = 0.42. The most popular explanation for this trial's result is not supported by the trial.
Incident type 2 diabetes	Up, observationally	Hazard ratio 1.24 comparing highest against lowest intake across 14 cohorts and 692,508 people, and 1.13 per 10 percent increase in intake. Cohorts, with heavy confounding by socioeconomic factors.

Long term, and hard outcomes

No randomised trial has run longer than eight weeks, and none has measured an event. The cohorts supply the long-term signal and cannot supply causation. The most useful thing to carry from this block is the comparison between its two trials rather than either alone. In 2019 an ultra-processed diet produced weight gain over two weeks. In 2025 an ultra-processed diet built to follow national dietary guidelines produced weight loss over eight weeks, and the advantage of minimal processing shrank to about one percent of bodyweight. Those are not contradictory results. Together they say that the nutritional composition of ultra-processed food, not its processing status by itself, drives most of the effect, and that a guideline-compliant ultra-processed diet behaves very differently from a typical one. That is a more useful and more actionable conclusion than the headline either trial generated.

Citations

Human2019
Ultra-processed diets cause excess calorie intake and weight gain: an inpatient randomized controlled trial of ad libitum food intake
Cell Metabolism
20 adults, NIH ward, two weeks per diet in crossover. Energy intake 508 kcal/day greater on ultra-processed (p = 0.0001), from carbohydrate and fat but not protein. Weight plus 0.9 kg on ultra-processed and minus 0.9 kg on unprocessed. Consumed energy density was not matched (1.36 against 1.09 kcal/g) and eating rate was faster; palatability did not differ.
Human2025
Ultraprocessed or minimally processed diets following healthy dietary guidelines on weight and cardiometabolic health: a randomized, crossover trial
Nature Medicine
55 adults, two eight-week ad libitum diets both following the UK Eatwell Guide. Both produced weight loss: minimally processed minus 2.06 percent and ultra-processed minus 1.05 percent, difference 1.01 percent (p = 0.024, Cohen's d minus 0.48). Extensive competing interests declared on both sides of the topic.
Review2019
Ultra-processed foods: what they are and how to identify them
Public Health Nutrition
The NOVA classification reference on which every study in this block depends. A published definition rather than a measured quantity, which makes it an assumption carried into all of this evidence.
Review2025
Ultra-processed food consumption and risk of type 2 diabetes: a systematic review and updated meta-analysis of prospective cohort studies
Metabolism
14 prospective cohorts, 692,508 participants. Highest against lowest ultra-processed consumption HR 1.24 (1.14 to 1.34), I squared 69 percent. Per 10 percent increase in intake HR 1.13 (1.08 to 1.18). Observational, with strong socioeconomic confounding.

What people report

These are uncontrolled self-reports, not evidence. They are here because they tell you what to expect and what to watch for, which the trial literature does not. They cannot tell you whether anything works.

Reduced snacking and less frequent hunger when switching to home-cooked food, reported very widely and consistent with the energy density and eating rate findings.
Substantially more time spent cooking and shopping, the most common reason people stop.
Confusion about what counts, which is a fair response to a classification where wholemeal bread and a fizzy drink share a category.
Higher grocery cost reported often, though the trial evidence says nothing about cost.
Slower eating described as the change people notice most, which happens to be the variable the ward trial measured as differing.

Sources: General nutrition communities and the growing popular literature on ultra-processed food, some of it written by authors who hold commercial interests in the topic. Note that this applies in both directions: the competing interest statements in the 2025 trial include book royalties on ultra-processed food, broadcasting fees on the subject, pharmaceutical employment and shareholding, and clinic ownership. Uncontrolled self-report otherwise.

Who this is wrong for

Nobody, since this is not a prescription, and one caution is worth stating: treating a NOVA category as a health verdict will mislead you in both directions. Wholemeal supermarket bread and a sugar-sweetened drink both classify as ultra-processed.
Anyone for whom minimally processed cooking is not realistic on time, money or facilities. The 2025 trial shows a guideline-compliant ultra-processed diet still produced weight loss, which is a genuinely useful result for people whose constraint is not willpower.
Anyone using this framework to avoid fortified foods. B12-fortified products are classified as processed and are the practical route to adequacy for many people, particularly on plant-based patterns.
Anyone reading the cohort data as causal. The socioeconomic confounding in ultra-processed intake is among the strongest in nutritional epidemiology.

Questions

Does ultra-processed food make you eat more?: In a metabolic ward where food was provided and intake measured, people ate 508 kcal a day more on an ultra-processed diet and gained 0.9 kg in two weeks. That is the most reliable finding in this area. The same paper shows why the simple reading is wrong: the energy density of what was actually eaten was not matched, and the authors say the 85 percent higher energy density of the non-beverage foods likely contributed.
Is it the processing itself?: The trial's own data points at energy density and eating rate instead. People ate 17 kcal/min faster on the ultra-processed diet, and reported no difference at all in how pleasant or satisfying the food was, which undercuts the hyperpalatability story. The authors state that no causal relationship between ultra-processed food and human obesity has yet been established.
So can a diet high in ultra-processed food be healthy?: The 2025 trial is the best answer anyone has. When both an ultra-processed and a minimally processed diet were built to follow national dietary guidelines, both produced weight loss and minimal processing won by about one percent of bodyweight over eight weeks. That suggests composition matters far more than processing status, which is useful if cooking from scratch is not realistic for you.
Why do the two trials disagree?: They do not, quite. In 2019 a typical ultra-processed diet produced weight gain. In 2025 a guideline-compliant ultra-processed diet produced weight loss. Together they say that what is in the food drives most of the effect and that ultra-processed is a category broad enough to contain both.
What about the studies linking it to diabetes?: They are cohorts: 14 of them, 692,508 people, hazard ratio 1.24 comparing highest with lowest intake. The direction is consistent and the confounding is severe, because people who eat the most ultra-processed food differ in income, education, time and cooking facilities, none of which adjustment fully removes. The cohorts give length where the trials give control, and neither gives both.