Ultra-processed food
The more ultra-processed food people eat, the more disease and earlier death they tend to have, and in short feeding trials people overeat on it without noticing. Whether processing itself does harm beyond the calories, sugar, salt and fat these foods carry is plausible but still unproven — which changes what is worth worrying about.
Ultra-processed food (UPF) is the category of industrial formulations that supply 53% of adults' calories in the United States and about 54% in the United Kingdom — and 62% in American children.[1] [2] The US adult share fell significantly between the 2017–2018 and 2021–2023 survey cycles, from 56% to 53% — the first significant fall the national survey has recorded. The direction of the evidence is reliable: high UPF intake is consistently associated with worse health, and cutting it — especially sugary drinks and processed meats — is sound advice. The certainty is thinner than headlines imply: most of the disease associations are observational, and the graders who have looked at them formally rate almost all of that evidence "low" or "very low".
What counts as "ultra-processed"
The widely used NOVA classification (developed at the University of São Paulo) sorts foods by how they were made, not by their nutrient content. There are four groups:
| NOVA group | Description | Examples |
|---|---|---|
| 1 — Unprocessed / minimally processed | Whole foods altered only by drying, freezing, fermenting, pasteurisation | Vegetables, fruit, legumes, plain oats, milk, eggs, raw meat, plain yogurt |
| 2 — Culinary ingredients | Substances pressed, refined, or milled from Group 1 | Olive oil, butter, refined sugar, salt, honey |
| 3 — Processed foods | Group 1 combined with Group 2 ingredients to extend shelf life | Tinned fish, freshly baked artisan bread, salted nuts, hard cheese |
| 4 — Ultra-processed (UPF) | Industrial formulations containing at least one ingredient never or rarely used in home kitchens (isolated proteins, hydrogenated or interesterified oils, high-fructose corn syrup, hydrolysed proteins) or a cosmetic additive (emulsifiers, artificial flavours, colours, sweeteners, thickeners) | Soft drinks, packaged breads, reconstituted meat products, instant noodles, flavoured yogurts, packaged cookies, most "protein bars" |
One qualifying item is enough to put a food in Group 4.[3] In practice two markers do the flagging: an ingredient you cannot buy at a supermarket for home cooking — soy protein isolate, maltodextrin, interesterified oils, hydrolysed proteins — and the cluster of cosmetic additives at the end of an ingredient list: emulsifiers, thickeners, artificial colours, flavour enhancers, non-caloric sweeteners.
The classification has serious critics. Independent academics argue NOVA conflates formulation (which ingredients are used) with processing (industrial treatment), and that much of the harm may reduce to the underlying nutrient profile.[4] The reproducibility problem has been measured directly: when 159 food and nutrition specialists sorted 120 real supermarket products into NOVA groups with the ingredient lists in front of them, agreement was only fair — a kappa statistic of 0.32, where 1.0 would be perfect and 0 would be chance.[5] The counter-evidence is that the harm signal often persists after adjusting for nutrient quality, suggesting processing may matter on top of the nutrition label — though not in every cohort, as the sub-category data below show.[6]
What the evidence actually shows
Moderate — a dose-response relationship replicated across many cohorts, though they disagree substantially on its size, and two small controlled-feeding trials suggesting that calories, salt, sugar and fat do not explain the whole effect; but the disease associations are observational and graded mostly "low" or "very low" certainty, and the strongest trial is contested.
Mortality and chronic disease
A 2024 umbrella review pooling 45 meta-analyses across roughly 10 million people found high UPF exposure was associated with 32 distinct adverse outcomes spanning cardiometabolic, cancer, and mental-health domains.[7] The caveat the headline obscures is that the review graded every result twice: once for how credible the association looks (a five-step scale topped by "convincing"), and separately for how much confidence the underlying evidence deserves (GRADE, the standard system for that judgement). Four outcomes reached the top "convincing" tier: cardiovascular-disease mortality (about 50% higher risk, but GRADE very low), type 2 diabetes (about 12% higher per dose increment, GRADE moderate), anxiety (about 48% higher, GRADE low) and common mental disorders taken together (about 53% higher, GRADE low). All-cause mortality, heart-disease mortality and depression sat one rung down at "highly suggestive", all at low certainty. Across all 45 pooled analyses only four earned moderate certainty and none earned high.
Headline numbers from updated 2025 dose-response analyses:
- Highest versus lowest UPF intake carries about 15% higher all-cause mortality — the pooled estimate from the largest dedicated dose-response meta-analysis to date (18 cohorts, 1.15 million adults).[8] The 95% confidence interval was 1.09–1.22 (in plain terms: the range where the true effect most plausibly lies — here, a mortality increase somewhere between roughly 9% and 22%). The often-quoted 62% figure comes from a single Spanish cohort of university graduates in which just 335 people died, giving a very wide plausible range — anywhere from 13% to 133% higher risk.[9]
- The per-increment estimate is genuinely unsettled. That same 18-cohort meta put each 10% rise in the energy share from UPF at about 10% higher mortality (HR 1.10, 1.04–1.16), but with very high between-study heterogeneity (I² = 91%). A separate meta built for an eight-country mortality model put it at about 3% per 10% (RR 1.03, 1.02–1.04).[10] So the per-10% figure is best read as roughly 3–10%, not a settled value. That eight-country model attributed between about 4% (Colombia) and 14% (UK and US) of premature deaths to UPF — a striking figure that depends entirely on treating the observational associations as causal.
- A systematic review of 41 cohorts covering roughly 8.3 million adults put each extra 100 g/day at about 14.5% higher hypertension risk, 5.9% higher cardiovascular-event risk, and 19.5% higher digestive-disease risk. This is a conference abstract from ACC Asia 2025, not a peer-reviewed publication — treat the precise numbers as provisional.[11]
- A 2025 Lancet three-part series (44 authors) concluded UPFs harm health across major organ systems and called for policy action.[12] It is an advocacy framing led by the researchers who originated NOVA, and the journal published five critical letters with the authors' reply in March 2026.[13]
Those pooled figures are averages over cohorts that disagree with one another a great deal: for the highest-versus-lowest comparison, 83% of the variation between the 18 studies was more than chance alone would produce (I² = 83%). The sharpest illustration comes later on this page — two US cohorts of nurses and doctors, followed for more than three decades, put that same top-versus-bottom contrast at 4%, not 15%. Three things pull an estimate down that far. How far apart the top and bottom groups actually are: in those cohorts the heaviest quarter ate about 7.4 servings of ultra-processed food a day and the lightest about 3.0, a much narrower gap than "highest versus lowest" implies in a general-population sample. How much of the difference between heavy and light eaters is smoking, income and overall diet rather than processing: in these same cohorts, dropping smoking alone from the statistical model made the association with mortality much stronger, which is a direct measure of how much work confounding does here. And how thoroughly those things were measured: diet was re-surveyed every four years for thirty years in a population that is unusually health-conscious to begin with.[14] The practical reading is that the 15% describes how much worse the health of heavy ultra-processed eaters is in the population at large — much of which may be the company these foods keep — while the 4% is closer to what someone who already eats reasonably well and does not smoke should expect from cutting back further. Neither figure on its own shows that processing did the damage.
Brain and mental health
Moderate for the association, weak for causation — anxiety and common mental disorders are two of the umbrella review's four convincing-tier associations, but at low certainty; confounding is hard to rule out and there are no long-term trials.
Cognition is thinner still. A 2026 systematic review found only five studies on UPF and cognition; three of the five reported worse cognitive outcomes. But the reviewers' own closing recommendation is that future work "should incorporate a measure of overall diet quality to aid in determining whether the effect of UPF is independent of dietary pattern or influenced by it" — so whether this is about processing or about the poor diets ultra-processed food travels with is exactly what the review says is still unknown.[15]
Autoimmune disease
Weak / preliminary — a plausible mechanism plus one borderline cohort association, no trial evidence.
A 2026 review on UPF and rheumatic disease links emulsifiers, thickeners, synthetic colourants, and microplastic exposures to gut-barrier disruption, endotoxemia (bacterial toxins leaking into the bloodstream), and a shift toward inflammatory immune cells (a Treg/Th17 imbalance).[16] That is a plausible mechanism for a real but modest signal: among 207,000 UK Biobank adults followed twelve years, the top fifth of UPF intake had 17% more new rheumatoid arthritis than the bottom fifth — an association weak enough that its plausible range barely clears no effect (hazard ratio 1.17, 95% CI 1.01 to 1.36).[17]
The 2025 UCL trial: important, but contested
Moderate (single trial), contested — a well-conducted randomised trial, but small (55 randomised, 50 analysed), short, nine-tenths female, and confounded by the fact that the UPF diet was about a quarter more energy-dense; three formal critiques followed.
A randomised crossover trial run by University College London is the field's most discussed recent evidence.[18] Fifty-five adults — nine in ten of them women, average BMI 33 — spent eight weeks on each of two diets built to the UK's Eatwell Guide, one made of ultra-processed foods and the other of minimally processed foods, with a four-week break in between; 50 provided data for the main analysis. They could eat as much or as little as they liked.
Results:
- The minimally processed diet produced roughly twice the weight loss (−2.06% vs −1.05% body weight) — about a 1-kg difference over eight weeks.
- Weight loss on the minimally processed diet was largely from fat mass, with lean muscle preserved.
- Participants on the minimally processed diet reported better craving control, and their food diaries showed a spontaneous deficit of about 330 kcal a day — though only about two-thirds of them returned diaries.
The trial should not be read as settling the question. Notably, both diets produced weight loss — which the authors say ran counter to what they had expected from the observational literature, though the registration itself specified only that the two diets would be compared, not which way the result would go — and both followed healthy-eating guidelines. Three formal critiques appeared together in the February 2026 issue of Nature Medicine, and between them they make the case that processing did little beyond making food more palatable and energy-dense:
- The diets were nutritionally mismatched in a way that could explain the result without invoking processing.[19] On the trial's own corrected figures the ultra-processed diet was about a quarter more energy-dense than the minimally processed one — 1.60 versus 1.25 kcal per gram, after a December 2025 correction to the trial's supplementary data.[20]
- The ~1-kg difference is of limited public-health importance and does not justify processing-based dietary guidance, and the missing-data and carryover handling are questionable.[21]
- The body-composition signal may be a bioimpedance artifact (water and glycogen shifts), with the cholesterol benefit attributable to oat fibre rather than processing.[22] Body fat was measured with a bioimpedance scale and never with a DXA scan, and the trial's own data show total body water fell about half a kilogram more on the minimally processed arm.
The authors replied that energy density is intrinsic to ultra-processing and should not be "matched away".[23] Worth flagging for balance: several trial authors have obesity-drug industry ties (one is an Eli Lilly employee; another has Novo Nordisk and Eli Lilly links), and a co-author authored a popular anti-UPF book.
The other feeding trials
The UCL result aligns directionally with the foundational 2019 NIH metabolic-ward trial, in which a UPF diet drove about 500 kcal/day of involuntary excess intake[24] — a finding pointed at, though not replicated, by a small Japanese pilot: nine men ate about 800 kcal a day more on a one-week ultra-processed diet than on a matched unprocessed one, and chewed less per calorie, though with nine participants the true figure could be anywhere from 340 to 1,280 kcal.[25] A later analysis of 2,733 meals from the NIH trial found energy density, eating rate and hyper-palatable foods consistently tracked how much people ate at a meal; additives were never tested.[26] Hall, who led that trial, is also a co-author of the UCL one, so the two are not fully independent lines of evidence.
Within the POUNDS Lost weight-loss trial, a statistical model applied to 356 participants' diet recalls estimated that trading 10% of calories from ultra-processed for minimally processed foods went with modestly greater weight and body-fat loss over six months — an observational look inside a trial, so an association rather than proof that the swap caused it.[27]
One more trial is often cited here, and shows less than it appears to. In 2026, 27 young adults were provided two 14-day diets, 81% versus 0% ultra-processed, matched not only for calories, macronutrients, fibre and added sugar but for energy density too. Because both diets were fully provided and calorie-matched, the trial could not test passive overeating at all; what it tested was appetite afterwards, at a single buffet meal following each period. There, intake was no different — with one exception the authors flag: in an exploratory split by age, the eighteen- to twenty-one-year-olds did eat more after the ultra-processed period, which is what the paper's own conclusion points at.[28] So it is evidence that a fortnight of ultra-processed eating does not obviously reprogram later appetite in most adults — not evidence about the overeating effect itself.
The takeaway: in short feeding trials people do passively overeat on ultra-processed diets — but the effect tracks energy density and how fast the food is eaten, and no trial has yet isolated processing from either. That processing harms independent of nutrients, calories and texture is plausible, not proven.
Why processing itself might harm: gut, inflammation, and the brain
Weak / mechanistic — mostly animal, cell, and observational data, with very few human randomised trials; these pathways are plausible but not established in people.
Eating speed and satiety
Industrial processing physically pre-digests food. Less chewing is needed; gastric emptying speeds up. Calories arrive faster than the gut's satiety hormones (glucagon-like peptide-1, GLP-1; and peptide YY, PYY) can signal fullness — so people passively overconsume. The cleanest test of this is a 2-week randomised crossover in 41 healthy-weight young adults that held palatability, portion size, calories served, and UPF content constant and varied only food texture: people ate 369 kcal a day less on the slower-to-eat version (95% CI 221 to 517), and the gap did not shrink over the fortnight.[29] Its lead author also co-wrote the first of the critiques of the UCL trial above and declares speaking fees from Ferrero, PepsiCo, General Mills, Mondelez and the World Sugar Research Organization, so read it as a consistent viewpoint rather than independent corroboration. It remains strong evidence for the eating-rate mechanism — and, read the other way, evidence that "processing" per se may be doing less work than texture.
Microbiome disruption and "leaky gut"
UPF-heavy diets reduce microbial diversity and deplete two bacteria that keep the gut lining healthy in different ways: Faecalibacterium prausnitzii, which ferments fibre into butyrate — the short-chain fatty acid colon cells burn as fuel — and Akkermansia muciniphila, which grazes the mucus layer and in doing so helps regulate its thickness and the tight junctions sealing the intestinal wall. This comes from a narrative review rather than a systematic one.[30]
Several common additives have mechanistic or preliminary evidence for direct effects — though most of it is from animal models, cell studies, or observational cohorts, with very few human randomised trials.[31] The human data are thin but no longer negligible — see the two randomised emulsifier trials described below the table. Randomised trials of non-caloric sweeteners are genuinely mixed.
| Additive | Where you find it | Mechanistic / preliminary effect |
|---|---|---|
| Emulsifiers (carboxymethylcellulose, polysorbate 80) | Ice cream, baked goods, sauces, plant-based milks | Erode the gut mucus layer; promote inflammation; linked to inflammatory bowel disease and metabolic syndrome |
| Non-caloric sweeteners (sucralose, saccharin, aspartame) | Diet sodas, "low-sugar" yogurts, protein bars | All shift the stool and oral microbiome at doses below the regulatory limit; only saccharin and sucralose measurably worsened glucose responses in the best trial, and the effect was strongly person-specific — see Sugar substitutes |
| Maltodextrin | Sports drinks, packaged sauces, "fat-free" foods | Thins the protective mucus layer and increases bacterial adhesion to the gut lining — in rats. Human trials show shifts in bacterial composition only, with no inflammation endpoint measured |
| Carrageenan | Plant-based milks, processed cheese, deli meats | Laboratory colitis models use degraded carrageenan (poligeenan), not the food-grade additive — but a 2026 randomised trial in 60 adults found food-grade carrageenan raised intestinal permeability compared with participants' own baseline (not compared with placebo), with no change in inflammatory markers. Genuinely contested |
| Azo dyes (Red 40, Yellow 5) | Sweets, drinks, cereals | Alter microbiota and induce colitis in mice — inactivating the azo bond abolishes the effect. A randomised trial in 297 UK children found small increases in hyperactivity, but the test drinks also contained sodium benzoate, so the dyes cannot be isolated[32] |
| Titanium dioxide (TiO₂) | Whitening agent in candies, sauces, supplements | Accumulates in gut epithelium; oxidative stress and DNA damage |
Titanium dioxide is the one additive here that a major regulator has acted on. The EU withdrew it from the permitted list in August 2022, after its food-safety authority concluded the additive "can no longer be considered safe" because genotoxicity could not be ruled out and no safe daily intake could be set. It remains permitted in the United States, where a petition to repeal its listing has been pending since 2023.
The proposed next step is that a thinned mucus layer lets bacterial fragments (lipopolysaccharide, flagellin) cross into the bloodstream, where they trip the innate immune system's bacterial sensors, release inflammatory messengers (interleukin-6, tumour necrosis factor-α) and drive chronic low-grade systemic inflammation — the pathway implicated in insulin resistance, atherosclerosis, and neurodegeneration. In animals that chain is well demonstrated. In people it is much thinner than the mechanism implies. The best-controlled emulsifier trial fed seven adults a high dose of carboxymethylcellulose for eleven days: it did perturb the microbiome and deplete short-chain fatty acids, but the antibodies to bacterial fragments it used to measure permeability, the serum inflammatory markers, and the gut-inflammation marker faecal lipocalin-2 were all unchanged.[33] A larger 2026 trial randomised 60 adults to four weeks of one of five emulsifiers and found the same absence of any inflammatory signal; the one exception was a permeability signal — food-grade carrageenan modestly raised intestinal permeability compared with baseline, though not compared with placebo.[34]
A candidate cardiometabolic biomarker
Imidazole propionate is a gut-bacterial metabolite made from the amino acid histidine that interferes with insulin signalling; it runs higher in people with type 2 diabetes and impairs glucose tolerance in mice.[35] A 2026 prospective analysis of 7,432 US health professionals found people in the top fifth for blood imidazole propionate had 82% higher coronary heart disease risk than the bottom fifth. The dietary lever turned out not to be the precursor — histidine intake showed no association — but fibre: pectin intake was the strongest negative predictor of the metabolite, which only rose with high histidine and an imidazole-producing microbiome when fibre was low.[36] No study has measured ultra-processed food intake against imidazole propionate directly, so the link to UPF is still inferred — but running through low fibre, which is a defining feature of these diets, it is a far more concrete hypothesis than it was.
Brain reward and addiction-like patterns
Ultra-processed foods cluster around nutrient combinations that rarely occur in nature — fat with sodium, fat with sugar, carbohydrate with sodium. That is not marketing folklore: it has a quantitative definition — and when its authors applied it to the USDA's food-composition database, 62% of the 7,757 items in it qualified.[37] Foods meeting it raise how much people eat at a meal independently of energy density and eating rate.[38] Chronic consumption is proposed to blunt dopamine reward signalling — animal work shows downregulation of dopamine D2 receptors, and human studies show reduced striatal response to sugary drinks after prolonged intake, a tolerance pattern that parallels substance use.[39] "Food addiction" remains a research construct measured by questionnaire and is not a recognised psychiatric diagnosis. The same review points to two further processing-linked exposures — advanced glycation end-products formed during high-heat processing, and bisphenol A from packaging — though no human intervention study has isolated either from overall UPF intake.
Accelerated biological aging
Weak / preliminary — the signal is consistent across the aging-biomarker studies, but the biological-age data are cross-sectional and the only direct DNA-methylation evidence is a 15-person pilot.
This is the longevity-relevant headline. Heavy UPF consumption shows up in composite biological-age scores and telomere measurements as faster biological aging, and the signal shrinks but does not disappear once overall diet quality is taken into account. Read "biological age" here as a blood-marker composite rather than a DNA-methylation clock.
A large cross-sectional analysis of 16,055 US adults (from the National Health and Nutrition Examination Survey, NHANES) used PhenoAge, a score modelled from chronological age plus nine routine blood analytes, and found:[40]
- Each 10% rise in energy from UPF added ~0.21 years (about 2.4 months) to biological age.
- Top quintile of UPF intake (68–100% of energy from UPF): nearly a full year of biological aging beyond chronological age vs the bottom quintile (0.90 years, 95% CI 0.60 to 1.20).
- The association was only modestly attenuated by adjusting for overall diet quality.
Because the analysis is a single snapshot, it shows correlation rather than a proven aging effect.
A pilot study sequenced blood DNA from just 15 women — the highest and lowest ultra-processed eaters out of 30 recruited — and found 80 stretches of DNA with different methylation between the two groups, most of them under-methylated in the high-UPF women. At that sample size this is hypothesis-generating and nothing more.[41]
Frailty is the better-measured end of this. A 2025 meta-analysis of 246,261 people found high UPF intake went with about a 40% higher risk of frailty in the studies that followed people over time (RR 1.40, 95% CI 1.25 to 1.58); the link to weak muscle strength was much smaller, appeared only in snapshot studies (RR 1.13, 1.06 to 1.20), and was absent in the two studies that followed people forward.[42] On telomeres, the evidence is one cross-sectional study: among 886 older Spanish adults, those eating more than three UPF servings a day had roughly double the odds of short telomeres (OR 1.82, 1.05 to 3.22).[43] Whether displacing protein and phytochemicals is what drives the muscle signal is a plausible guess, not something these studies tested.
Is it causal? The skeptic's case
Most of the disease evidence is observational, and there are real reasons the associations might overstate any true causal effect:
- Healthy-user bias and residual confounding. People who eat the most UPFs also tend to smoke more, exercise less, earn less, and face more food insecurity — all independent drivers of poor health that no statistical adjustment fully removes. Robinson and Jones calculated an E-value of 1.55 for the link between UPF intake and weight gain: an unmeasured factor connected to both — depressive symptoms, a tendency to overeat, or poor access to nutritious food are the examples they give — would need only a risk ratio of about 1.55 with each, over and above the confounders the studies already adjusted for, to explain the whole association away.[44]
- Negative-control signals. In the 450,000-person EPIC cohort, higher UPF intake was also associated with dying in an accident — about 12% higher risk per increment, an outcome no diet plausibly causes. The researchers report it precisely as a signal that residual confounding is at work.[45]
- Reverse causation. People who are already unwell may shift toward cheaper, more convenient processed foods.
Weighed against this, the field does show consistency across populations, a dose-response gradient, and biological plausibility. But experimental evidence is limited to short, small trials, and UPF is associated with almost every disease studied — which argues for caution about reading every association as direct cause and effect.
Not all UPFs are equally bad
Moderate — the sub-category breakdown comes from large, long-followed cohorts, though still observational.
The category is broad — from a diet soda to a whole-grain industrially baked bread fortified with vitamins. In two US health-professional cohorts totalling 114,000 people — followed for a median of 34 years in the women's cohort and 31 in the men's — the top quarter of UPF intake carried 4% higher all-cause mortality than the bottom — 1,536 versus 1,472 deaths per 100,000 person-years. But the whole signal came from causes other than cancer and cardiovascular disease, for which there was no significant association at all. And within any given band of overall diet quality the UPF association largely vanished, while diet quality predicted mortality within every band of UPF intake. This is the 4% set against the pooled 15% above, and the two are not in contradiction: the gap is mostly the narrower spread of intakes in these cohorts and the unusually thorough accounting for smoking and diet quality. Nor is the vanishing-within-diet-quality result the last word — a 2026 analysis of 47,999 US adults, adjusting for a broader index of diet quality, found the mortality association only partly weakened and still present.[46] What was left in the health-professional cohorts was concentrated in a few sub-categories:[47]
- Processed meats and other meat/poultry/seafood-based ready-to-eat products — the strongest signal (all-cause HR 1.13, against 1.09 for sweetened beverages)
- Sugar-sweetened and artificially sweetened beverages
- Dairy-based desserts
- Ultra-processed breakfast foods
A few NOVA-4 categories show no detectable harm rather than benefit. Among 267,000 European adults followed for a median of 11 years, industrially made breads and cereals and plant-based alternatives were not associated with going on to develop two or more of cancer, cardiovascular disease and type 2 diabetes — hazard ratios of 0.97 per standard-deviation rise in intake for each, neither statistically significant, with the bread-and-cereal result what the authors call inverse only "with borderline certainty".[48] "No harm detected" is not "protective": no cohort has shown these foods to be actively beneficial. And the finding does not hold everywhere — in the US cohorts above, ultra-processed breads and breakfast foods carried a small but statistically clear excess mortality (hazard ratio 1.04). Intact cereal fibres, calcium, B12 fortification and live cultures are a plausible reason a fortified product might fare better than a soft drink, but that is an explanation offered for the pattern, not something these studies measured.
The practical reading: processed meat and sugary drinks are where cutting back pays most clearly. A few "less-bad" UPFs — a fortified plant milk, a packaged wholegrain bread — are pragmatic transitional foods for a busy adult, but they remain compromises against a diet anchored in whole foods.
What to actually do
- Run the kitchen test. If an ingredient on a label is something you cannot buy in a normal grocery store for home cooking — soy protein isolate, high-fructose corn syrup, hydrogenated oils, dough conditioners, hydrolysed proteins — the product is ultra-processed.
- Scan the bottom of the ingredient list. Cosmetic additives cluster there: emulsifiers (soy lecithin, carrageenan, polysorbate 80), thickeners (xanthan gum, maltodextrin), artificial colours, flavour enhancers, non-caloric sweeteners. Their presence flags a Group 4 food regardless of front-of-package marketing.
- Ignore the marketing. "All-natural", "fortified", "plant-based", "high-protein", "gluten-free" tell you nothing about processing.
- Make the easy swaps first. The biggest wins are concentrated in a few categories:
| If you currently eat | Try instead |
|---|---|
| Sugar-sweetened or diet soda | Plain or sparkling water; tea; black coffee |
| Flavoured/sweetened yogurts | Plain yogurt or kefir + fruit + honey |
| Boxed breakfast cereals | Steel-cut or rolled oats; plain yogurt with nuts |
| Mass-produced sliced bread | Fresh bread (flour, water, salt, yeast only) or whole-grain sourdough |
| Margarine / vegetable spreads | Olive oil, butter, mashed avocado, nut butter |
| Processed meats (deli, sausage, bacon) | Roasted whole cuts; tinned sardines; eggs; legumes |
| Packaged snacks (chips, cookies, bars) | Nuts, fresh or dried fruit, hard cheese, dark chocolate |
| Frozen ready meals | Batch-cooked one-pot meals from whole ingredients |
- Aim for the share, not perfection. No study identifies a magic number: the two 2025 dose-response meta-analyses both model the relationship as a straight line, and the one cohort that has run a spline analysis found a threshold for cardiovascular and heart-disease death — below roughly 2.3 to 2.4 servings a day the association disappeared — though it never looked at mortality overall.[49] So the honest reason to name a target is practicality, not biology — a share you can hold beats one you abandon. Under 10–20% of calories from ultra-processed food is a realistic destination from the 50–60% most Western adults start at, and the first half of that journey carries the categories with the strongest signals: sugary drinks and processed meat. Going lower is not pointless, just harder — treat it as optional rather than as a standard you have failed.
What's overrated and what's worth knowing
Overrated:
- "Healthy" UPF marketing — granola bars, protein cookies, sugar-free diet versions. These remain ultra-processed formulations with cleaner-looking front labels, and adding protein does not fix the overeating: in 21 adults living for 54 hours in a sealed metabolic chamber, a high-protein ultra-processed diet shifted energy balance favourably but did not prevent overeating.[50]
- Certainty in either direction. "It's just saturated fat and sugar" and "processing itself is the poison" both outrun the evidence: energy density and eating speed explain much of what the feeding trials show, and nothing yet isolates processing from them.
Worth knowing:
- The US regulatory picture. The 2025–2030 US Dietary Guidelines call to "limit highly processed foods, added sugars, and refined carbohydrates" — and notably use "highly processed" rather than the NOVA term "ultra-processed".[51] The FDA revoked the authorisation for the dye FD&C Red No. 3 in January 2025 — though food manufacturers have until January 2027 to reformulate, so it is still lawfully in food today.[52] A proposed rule would put a "Low", "Med" or "High" rating for saturated fat, sodium and added sugars on the front of most packages — filed in January 2025, still not final.[53] And the FDA and USDA asked the public for evidence on how to define ultra-processed food in July 2025, closed comments that October, and sent a first federal definition for final review in August 2026.
- The evidence graders have not followed the governments. The committees whose job is to formally grade the evidence have so far declined to issue UPF-specific guidance — even as Brazil's national guidelines have told people to avoid ultra-processed foods since 2014, six other countries name ultra-processing in theirs,[54] and the World Health Organization has convened a group to write a guideline that does not yet exist. The 2025 Dietary Guidelines Advisory Committee used the term "ultra-processed" throughout, and graded the evidence linking those diets to greater body fat and obesity risk as only "limited" — fifteen observational studies, with the one controlled feeding trial excluded as too short. The substitution to "highly processed" noted above was made downstream, by HHS and USDA, when they wrote the final Guidelines. And the UK's Scientific Advisory Committee on Nutrition looked again in April 2025 and reached the same verdict it reached in 2023: the associations are concerning, but it remains unclear whether the harm comes from processing itself or from the fact that most of these foods are high in calories, saturated fat, salt and free sugars — so its advice stays nutrient-based, and it will revisit the question at its 2026 horizon scan.[55]
- The American Heart Association has taken a position, and it is a hedged one. Its 2025 science advisory concludes that mechanisms "remain unclear" and that most ultra-processed foods "overlap with foods high in saturated fat, added sugars, and sodium, which are already targets for cardiometabolic risk reduction" — so it does not endorse the claim that processing harms independently of nutrients. It nonetheless concludes that "evidence supports food policies that limit UPF intake while avoiding unintended consequences", which is the clearest statement any major cardiovascular body has made on the question.[56]
- Phytochemical loss is a plausible part of the harm. A 2024 analysis of 1,946 food items found minimally processed foods carry roughly ten times the antioxidant capacity of ultra-processed ones. But ordinary processed (group 3) foods scored no better than ultra-processed ones, so this is a whole-food-versus-everything-else gap rather than a gradient across the NOVA scale — and the study measured what is in foods, not what anyone ate, so the step from here to "a chronic deficit in protective polyphenols" is an inference.[57]
Further reading
- Lane MM et al. Ultra-processed food exposure and adverse health outcomes: umbrella review of epidemiological meta-analyses. BMJ 2024.[58]
- Fang Z et al. Association of ultra-processed food consumption with all cause and cause specific mortality — population based cohort study. BMJ 2024.[59]
- Hall KD et al. Ultra-Processed Diets Cause Excess Calorie Intake and Weight Gain: An Inpatient Randomized Controlled Trial of Ad Libitum Food Intake. Cell Metab 2019.[60]
- Liang S et al. Ultra-processed foods and risk of all-cause mortality: an updated systematic review and dose-response meta-analysis of prospective cohort studies. Syst Rev 2025.[61]
- Nilson EAF et al. Premature Mortality Attributable to Ultraprocessed Food Consumption in 8 Countries. Am J Prev Med 2025;68:1091–1099.[62]
- Monteiro CA et al. Ultra-processed foods and human health: the main thesis and the evidence (Lancet Series, paper 1). Lancet 2025.[63]
- Forde CG et al. Eating rate has sustained effects on energy intake from ultraprocessed diets: a 2-week ad libitum dietary randomized controlled crossover trial. Am J Clin Nutr 2026.[64]
- Dicken SJ et al. Ultraprocessed or minimally processed diets following healthy dietary guidelines on weight and cardiometabolic health: a randomized, crossover trial (with three Matters Arising critiques and reply). Nat Med 2025.[65]
- Koh A et al. Microbially Produced Imidazole Propionate Impairs Insulin Signaling through mTORC1. Cell 2018.[66]
- UK SACN — statement on processed foods and health, 2023, and rapid evidence update, April 2025.[67]
- Cardoso BR et al. Association between ultra-processed food intake and biological ageing in US adults: findings from NHANES 2003–2010. Age and Ageing 2024.[68]
- Hojjati Kermani MA et al. The association of ultra-processed food intake on age-related muscle conditions: a systematic review and dose-response meta-analysis with meta-regression. J Health Popul Nutr 2025.[69]
- Rondinella D et al. The Detrimental Impact of Ultra-Processed Foods on the Human Gut Microbiome and Gut Barrier. Nutrients 2025.[70]
- Smith M et al. — ultra-processed food exposure and cognitive outcomes: a systematic review of observational studies. BMJ Nutrition, Prevention & Health 2026.[71]
- Vadiveloo MK et al. Ultraprocessed Foods and Their Association With Cardiometabolic Health: Evidence, Gaps, and Opportunities. Circulation 2025 (AHA science advisory).[72]
- Harvard T.H. Chan — Processed Foods and Health (general-audience overview, last updated 2023).[73]