Ultra-processed food: what the definition covers and what the trials show

Ultra-processed does not mean processed. It means assembled in a factory from extracted substances and additives, rather than cooked from food. When men were fed matched menus and left to eat as much as they liked, they took in 813.5 more calories a day on the ultra-processed one, and gained 1.1 kg in a week.¹

What the definition actually covers

The research uses a system called NOVA, which sorts food by how much and why it was processed, not by what nutrients it contains. It has four groups.

Unprocessed and minimally processed food comes first: vegetables, fruit, grains, meat, eggs, milk — and those same foods dried, frozen, ground or pasteurized. Then culinary ingredients: the oils, butter, sugar and salt pulled out of foods for use in kitchens. Then processed foods, which are the first two combined — canned vegetables, cheese, cured meat, fresh bread.

Ultra-processed foods are a different kind of object. They are built mostly from substances extracted or derived from food: protein isolates, modified starches, hydrogenated oils, high-fructose syrups. Additives then make the result look, taste and keep like food — emulsifiers, colors, flavors, sweeteners. Sweetened drinks, packaged snacks, reconstituted meat, most breakfast cereals and most factory bread sit here.

Two things follow. Processing on its own is not the marker: canning, freezing and fermenting all fall outside the group. And because the marker is the formulation, a product can be ultra-processed while looking unremarkable on its nutrition label. A home-cooked meal, meanwhile, can be heavy in fat, sugar and salt and still count as minimally processed.

So the category overlaps with plain diet quality without being the same thing. The research has real trouble telling the two apart, and one large review closed on exactly that question.⁵

How well anyone can measure it

The definition gets applied after the fact, to food records that were never designed to capture it. A review of 30 studies shows what that costs.⁶

The questionnaires repeat themselves reasonably well, scoring about 0.46 to 0.94 when the same people are measured twice.⁶ Checked against fuller dietary methods, they came out only modest to moderate, at about 0.47 to 0.72.⁶ Different classification systems agreed with each other at about 0.36 to 0.84.⁶ In plain terms, two teams coding the same diet will put a fair share of foods in different boxes.⁶

Keep that in view for everything below. In the United States ultra-processed food supplies most of the calories people eat, and older adults in one national survey averaged 50% of their energy from it.² The exposure is enormous and the ruler is blunt.

The feeding trials

Studies that follow people around cannot separate the food from the people eating it. That is why a handful of small feeding trials matter more than their size suggests.

The design is simple. House people. Offer them meals that are either ultra-processed or not. Let them eat as much as they want. Then switch them to the other menu.

At the NIH Clinical Center, 20 weight-stable adults aged 18 to 50 were admitted for four weeks. Each ate freely from a diet drawing 80% or 0% of its calories from ultra-processed food for two weeks, then switched straight to the other.² That trial's headline result on calories was published separately and is not among the records here. What is here is a companion analysis of it, and a replication.

The companion analysis tested taste. Two weeks on either diet left both sweet and salt detection thresholds unchanged, and preferences unchanged with them.³ One signal did show. After the ultra-processed diet, salt preference tracked upper blood pressure, at 0.59 on a scale where 1 would be a perfect match, and tracked body weight and body mass index too.³ Whatever makes people eat more of this food, it is not a change in what they can taste.

The replication came from Tokyo. Nine men with overweight or obesity ate ultra-processed and non-ultra-processed menus for one week each in random order, with two weeks between.¹ The two menus were built to match on total calories and on the balance of protein, fat and carbohydrate. On the ultra-processed week the men ate 813.5 more calories a day, gained 1.1 kg more, and chewed noticeably fewer times per calorie.¹ The authors' conclusion was blunt: ultra-processed food causes significant weight gain.¹

Two trials, both small and short. They point the same way. Because the second one matched the menus on paper, the extra eating cannot be explained by the ultra-processed food simply being richer. What neither trial can say is why.

What the long-term studies add, and what they cannot

Most evidence here comes from following large groups of people over years, and two reviews have sorted that literature by how much weight it can bear.

The first went through 39 earlier analyses drawing on 122 articles and 49 health outcomes.⁴ Twenty-five outcomes were linked to eating more ultra-processed food. Only two cleared the bar for convincing: decline in kidney function, at 1.25 times the odds, and wheezing in children and teenagers, at 1.42.⁴ Five more — diabetes, overweight, obesity, depression and common mental disorders — were rated one notch below that.⁴ Not one study anywhere in it reported a benefit.⁴

The second graded how certain each result was, and only a short list reached moderate certainty. For every extra 50 g a day, death from any cause rose 2% and heart disease 4%. For every extra 10% of daily calories, type 2 diabetes rose 12% and colorectal cancer 4%.⁵ Obesity, fatty liver, inflammatory bowel disease and mental health showed similar-looking numbers with weaker certainty behind them.⁵ Its authors ended on the honest question: is it the processing that raises risk, or is ultra-processed intake simply a way of measuring a poor diet?⁵

Long-term risk increase per additional ultra-processed food exposure% higher risk

Death, any cause (per +50 g/day)+2%

Heart disease (per +50 g/day)+4%

Type 2 diabetes (per +10% of calories)+12%

Colorectal cancer (per +10% of calories)+4%

A third review looked at the reviews themselves. Of 14 it examined, it judged 2 to be of moderate quality; the other 12 were low or critically low.⁷

Where the evidence thins

The trials are tiny and brief: 9 men for one week, 20 adults for two, all of them somewhere every meal was handed to them.^1,2 They establish that people overeat these menus in that setting. They do not establish which property of the food does it, whether it lasts once novelty wears off, or what happens to somebody doing their own shopping.

The long-term studies have the opposite problem. They are large and they run for years, and they cannot separate the food from the income, smoking, activity and everything else that travels alongside it. Under both sits a classification that different coders apply differently.⁶

What remains unknown

Nobody has measured which feature drives the extra eating. Tokyo points at chewing, and so at texture and speed. The taste analysis rules out a shift in taste sensitivity within two weeks.^1,3 Neither one isolates energy density, additives, sheer palatability, or the form the portion comes in.

Whether a reformulated product with more protein or fiber escapes the effect is a separate question, and these trials did not ask it. No trial has run long enough to show that weight gained in a week or two stays on. And none has swapped ultra-processed for minimally processed food at equal calories to see whether any of the diseases the long-term studies describe actually shift.⁴