What Counts as Ultra-Processed Food, and Why It Matters

Ultra-processed food has a precise, checkable definition, and it has one controlled human trial behind it that proves something: people eat more and gain weight on an ultra-processed diet than on a matched unprocessed one, inside the same two weeks, in the same bodies.^[1] The much bigger pile of evidence linking ultra-processed food to long-term disease is real, consistent, and still observational. It shows an association. Both of those facts are true at once, and most explainers blur them together.

What “ultra-processed” means

Ultra-processed is not a vibe. It comes from NOVA, a classification system that sorts foods into four groups by the degree and purpose of industrial processing.^[2] Nutrient content never enters the sorting.

• Group 1: unprocessed or minimally processed foods
• Group 2: processed culinary ingredients (oils, salt, sugar)
• Group 3: processed foods (canned vegetables, cheese, fresh bread)
• Group 4: ultra-processed foods

NOVA’s fourth group is the one doing all the work in headlines. It covers formulations made mostly or entirely from substances derived from food plus additives, with little if any intact whole food left in the finished product.^[2] Think of a soda, a packaged snack cake, or a reconstituted chicken nugget: ingredients extracted and recombined, not a food you’d recognize if you grew or raised it yourself.

The classification itself was created in 2010 by Brazilian epidemiologists as a tool for public health research and dietary guidelines.^[3] It was never built to grade an ingredient’s safety. It describes how far a food has traveled from its raw form, which turns out to predict some things and not others.

Why the category draws so much attention

Large cohort studies have repeatedly found higher ultra-processed food intake associated with increased risks of all-cause mortality, cardiovascular disease, and type 2 diabetes, with moderate-certainty evidence behind each of those links.^[4]

That’s a real signal. It’s also a signal from observational cohorts: people’s existing eating habits tracked over years, with no diet ever randomly assigned. Cohorts can show that people who eat more ultra-processed food get sicker more often. They can’t, by themselves, prove the food is what’s doing it.

Researchers are genuinely split on why the association shows up at all. Three competing explanations get argued over constantly:

• The additives and processing itself cause biological harm
• The energy density and palatability of these foods simply make overeating easier
• People who eat a lot of ultra-processed food tend to differ in other health-relevant ways, and the food is a marker, not a cause

No cohort study can referee that argument. Only a controlled trial can, and there have only ever been two.

The number everyone repeats: a 2024 umbrella review found that every additional 10 percentage points of ultra-processed food in someone’s diet tracked with a 12% higher risk of type 2 diabetes.^[5] It’s the same shape of relationship as pack-years and lung cancer: more exposure, more risk, without identifying which specific thing in the mix is responsible.

12% higher type 2 diabetes risk per 10 percentage points more ultra-processed food in the diet PMID 38418082

What the research base looks like

Most human evidence on ultra-processed food comes from large prospective cohorts. A 2024 umbrella review alone pooled 45 separate meta-analytic estimates, drawing on close to 10 million participant-outcomes across cohort, case-control, and cross-sectional designs.^[5]

Controlled trials are the rare exception. A 2019 NIH metabolic ward study was the first randomized controlled trial to directly test an ultra-processed diet against an unprocessed diet, with meals matched for calories, sugar, fat, and fiber.^[1] For years it stood alone.

That same 2024 umbrella review rated most of its own underlying evidence honestly: of 45 pooled estimates, 41 were graded low or very low certainty, with only four rated moderate and none rated high.^[5] A field built mostly on cohort data, by its own umbrella review’s admission, isn’t sitting on strong individual studies yet, even where the pattern across studies is consistent.

How researchers study this

The two study types work completely differently, and knowing which one you’re reading changes how much weight a claim deserves.

Cohort studies estimate intake with food frequency questionnaires or repeated dietary records, then sort each reported food into a NOVA group before comparing health outcomes across people who eat more or less of it.^[6] Nobody’s diet is assigned. Everyone just keeps eating what they already eat, and researchers watch what happens over years.

The controlled trial design is stricter and much smaller. In the 2019 metabolic ward trial, 20 adults lived on a research unit for two consecutive two-week periods, eating as much as they wanted from meals that were ultra-processed in one period and unprocessed in the other.^[1] Nothing else about the food changed. Same people, same unit, same instructions: eat however much you want.

The result was not subtle. Participants ate about 500 more calories per day and gained roughly 0.9 kg during the ultra-processed period, then ate less and lost about 0.9 kg during the unprocessed period.^[1]

What’s established versus what’s still assumed

Because the 2019 trial randomized diet order and matched meals for calories and macronutrients, the increase in intake and weight during the ultra-processed period can be attributed to the diet itself, within that small two-week-per-arm design.^[1] That’s a genuine causal finding. It is also a narrow one: one research unit, one country, one short exposure window.

Here’s the detail most coverage of this topic misses entirely:

A second trial exists, and most explainers don’t mention it. A 2025 randomized crossover trial tested ultra-processed and minimally processed diets, both built around the UK’s own national dietary guidelines, in 55 adults with overweight or obesity over 8 weeks per diet. It found significantly greater weight loss on the minimally processed diet again.^[7]

That second trial matters for a specific reason: it ran longer, enrolled nearly three times as many people, and used realistic guideline-based meals instead of a single tightly matched menu, and it still landed on the same side. That’s two independent trials, in two countries, pointing the same direction.

What neither trial can tell you:

• Whether the effect holds over months or years rather than weeks
• Whether it generalizes beyond adults with overweight or obesity
• Which specific component (additives, energy density, texture, or the eating pattern itself) is driving the extra intake

No published human trial has isolated that mechanism yet. The much larger cohort associations with chronic disease remain correlational on top of that, and can’t rule out that people who eat more ultra-processed food differ in other health-relevant ways the studies didn’t measure.

Where to read further

If you only read one primary source on this topic, make it the one with an actual experiment in it. The 2019 NIH metabolic ward trial, published in Cell Metabolism, is among the most-cited and most influential interventional studies on ultra-processed food intake, with more than 1,200 citations.^[1]

For the observational side, the 2024 umbrella review in The BMJ is one of the broadest syntheses published to date of associations between ultra-processed food intake and health outcomes, pooling those 45 separate meta-analytic estimates in one place.^[5] It’s the single best way to see the whole observational picture without reading 45 papers.

We’ve written before about how the ultra-processed label on oat milk measures less than people think, which is a useful case study in how a NOVA group 4 classification and an actual health risk can diverge. If you want the additive side of this argument specifically, we also cover the two food additives with real human trial evidence behind them.

Bottom line: ultra-processed describes a processing category, nothing about toxicity. The one thing two separate randomized trials now agree on is that an ultra-processed diet gets people to eat more and gain weight compared with a minimally processed one, and both trials proved it inside a matter of weeks. Everything about cancer, heart disease, and mortality risk beyond that still rests on cohorts that can show the pattern without pinning down the cause.

Reading a label and wanting to know how processed something is, rather than guessing from the ingredient list, is exactly what a scanner is for. EatMatrix reads the panel and flags the group 4 markers directly.

This article summarises publicly available regulatory assessments and peer-reviewed research about food ingredients. It is general information and is not intended to diagnose, treat, cure, or prevent any disease, and it is not personalised dietary or medical advice. Regulatory positions change; the EatMatrix app reflects the current assessments in our database. If you have an allergy, a medical condition, or are pregnant, talk to a qualified clinician or dietitian before changing what you eat.