The useful ultra-processed-food question is what replaces it

Food research interpretation · Checked November 20, 2026

The useful question about ultra-processed food is what replaces it. Removing a category name from a shopping basket is not a complete diet. A trial compares actual menus, people and conditions. If we erase the comparator, we erase much of what the experiment can tell us.

Two influential randomized feeding trials make that point unusually well. One found weight gain on an ultra-processed diet. The other found weight loss on both diets, with more loss on the minimally processed one. These results are not a license to pick whichever headline confirms your existing position. They are an invitation to read the diets.

The inpatient trial: more food was eaten

In Hall and colleagues' 2019 NIH experiment, 20 adults lived at the research center and received ultra-processed and unprocessed diets for two weeks each, in randomized order. They could eat as much or as little as they wanted. Meals were designed to match several presented nutritional characteristics, but participants did not consume identical amounts.

Energy intake averaged 508 kcal/day higher on the ultra-processed diet, with a reported standard error of 106. Participants gained about 0.9 kg during that condition and lost about 0.9 kg during the unprocessed condition. This was evidence about the supplied dietary patterns over two weeks, not a calorie penalty attributable to every packaged item.

The free-living trial: both diets followed dietary guidance

In Dicken and colleagues' 2025 UPDATE trial, 55 adults in England were randomized to two eight-week diets, both designed around the UK Eatwell Guide. Participants had BMI from 25 to below 40 and habitually obtained at least half their calories from ultra-processed food. Fifty contributed primary-outcome data for at least one diet.

Estimated weight change was -2.06% on minimally processed food and -1.05% on ultra-processed food. The within-participant difference was -1.01 percentage points, with a 95% confidence interval from -1.87 to -0.14. Both conditions involved weight loss. The minimally processed condition produced more.

Selected trials, not a pooled estimate or an exhaustive review
FeatureNIH 2019UPDATE 2025
SettingInpatientFree-living, food provided
Duration per diet2 weeks8 weeks
Population count20 adults55 randomized; 50 in primary intention-to-treat sample
Observed weight patternGain on UPF, loss on unprocessedLoss on both; greater on minimally processed
What differs beyond the category label?Actual menus and their propertiesActual menus, including energy density

Both studies provided food. Neither establishes the cost or practical burden of reproducing its menus through your own shopping and cooking. Nor does a group result predict an individual's response.

The numbers do not cancel each other out

A difference between two diets is not the same quantity as change from baseline within one diet. In UPDATE, a negative weight change on the ultra-processed menu does not mean processing caused weight loss. It means participants lost weight while following that complete supplied menu under those study conditions.

Likewise, the greater loss on the minimally processed menu does not isolate a single additive or manufacturing step. The trial paper notes that its ultra-processed diet was more energy dense and that potential carryover could not be ruled out. A January 2026 published critique raises attribution concerns including attrition, order effects, recognizability and energy density. We reviewed its publicly accessible summary; the full correspondence was paywalled.

Those limitations do not make randomized feeding results disappear. They constrain the causal story. We can learn from the difference between complete dietary patterns while remaining uncertain about which components produced it.

A percentage-point example, with the prediction removed

Our worksheet applies the two UPDATE percentages to an invented common baseline of 80 kg, solely to explain the units. A 2.06% decrease is 1.648 kg; a 1.05% decrease is 0.840 kg. Their difference is 0.808 kg, corresponding to 1.01 percentage points of that invented baseline.

This is not the trial's observed kilogram difference, an estimate for an 80 kg reader, or a forecast. Individual baselines, model estimates and crossover measurements cannot be reconstructed by multiplying rounded group percentages by a convenient body weight. The calculation is a teaching example that keeps “percent” and “percentage points” separate.

Do not extend the eight-week result into a year by multiplying it by 6.5. That assumes the same rate continues, ignores changing circumstances and turns unobserved follow-up into a claimed outcome. The same problem arises when the NIH intake difference is treated as a permanent daily effect for anyone buying a particular snack.

Ask a better question at the supermarket

A useful comparison names both options. “Replace this breakfast with what?” is answerable in a way that “Is processing bad?” often is not. Identify the actual alternative, its portion, nutrients, price, preparation demands and whether you will eat it consistently. The trials reviewed here do not supply those personal facts.

This is also why a food app should preserve the source of a processing classification. A category label is not a measured clinical outcome, and a barcode match is not a test of an entire diet. If classification is uncertain, say so. Do not invent a numerical health penalty simply because a database has a category field.

Our practical contribution is a comparison template: record the population, comparator, duration, whether food was provided, what was measured and what the study leaves unresolved. Use it to read the next diet headline without reducing the findings to a universal blacklist or permission slip.

Methods, funding and limits

Download the trial map and calculation or JSON with definitions and source links. We selected these two whole-diet trials to explain differing comparators and outcome units. This is not a systematic review of all processing research, a new experiment or personalized dietary advice.

The NIH paper reports NIDDK intramural support and an author's food-industry relationships. UPDATE lists NIHR UCLH Biomedical Research Centre and Rosetrees support, with funders uninvolved in design, conduct or analysis; authors report several commercial relationships in its disclosures. Those disclosures belong beside the findings, not hidden behind a claim of perfect independence.

Publisher disclosure: we make Vora, which combines proactive AI with food, workout and wearable context. Our protein portion reference answers a narrower shopping question using exact label values. Neither a label comparison nor a personal food log reproduces a feeding trial. Corrections: support@askvora.com.

Frequently Asked Questions

Did both diets cause weight loss in the 2025 trial?

Both supplied diets were associated with weight loss during their respective conditions, with greater loss on minimally processed food. That does not isolate processing itself as the cause of the within-diet change.

Can the 2019 calorie difference be assigned to a snack?

No. The 508 kcal/day group difference concerned complete supplied diets in a small inpatient trial. It is not a fixed penalty for an individual packaged food.

Is the 80 kg calculation a personal prediction?

No. It is an invented common baseline used to explain percentage-point arithmetic. It is not the trial's observed kilogram contrast.

Sources & References

  1. Hall et al. 2019 inpatient feeding trial
  2. Dicken et al. 2025 UPDATE trial
  3. Wang and Peng 2026 critique: public summary

All research discussed in this article is summarized in our own words. We link to original sources for full access. This content is for informational purposes and does not constitute medical advice.

NutritionEvidenceResearch interpretation

Experience it yourself

AI-powered workouts, photo nutrition logging, HRV recovery tracking, voice coaching, and more - all free to start.

Related Articles

Nutrition

Your water target may be counting the wrong thing

A total-water reference includes food and other drinks. Our source comparison and accounting worksheet show why a bottle-only log can misread the gap.

Read more
Nutrition

A tested supplement is not a proven benefit

What NSF and USP certification can establish, what FDA status means, and the separate evidence needed for a supplement benefit claim.

Read more
Nutrition

Creatine brain claims need a smaller headline

Selected controlled trials show why an acute sleep-deprivation result, everyday cognitive performance and longevity are different claims.

Read more