Britain has three separate answers to one question — is this product unhealthy? — and they do not agree.

A cross-sectional analysis published on 10 November in BMJ Medicine set them side by side on the same 1,898 processed and ultra-processed items drawn from the UK’s National Diet and Nutrition Survey, then added three new scoring models built directly from the country’s own dietary guidance. The result is less a verdict on the food supply than a measurement of the distance between what the government tells people to eat and what its regulatory tools recognise as a product to discourage.

What was actually compared

The database is the year-12 release of the National Diet and Nutrition Survey, drawn from four non-consecutive 24-hour dietary recalls. Of 3,105 items, the authors excluded 35 alcoholic drinks, 109 outside the Nova classification such as supplements, and 16 with no nutrient match, leaving 2,945: 1,636 ultra-processed, 262 processed, 61 processed culinary ingredients and 986 minimally processed. The last group was not tested, so every figure here concerns the packaged end of the aisle.

Each product then went through six systems. Two were versions of the Food Standards Agency nutrient profiling model: the 2004-05 model written into British advertising and promotions law, and a 2018 revision the paper says was published this year but is not in use. One was the voluntary front-of-pack traffic-light label. The remaining three were built for this study from Eatwell Guide thresholds — saturated fat, free sugar and sodium in the first; those plus total fat in the second; those plus energy density capped at 125 kcal per 100 g in the third.

For the traffic lights and the three new models, a product counted as unhealthy if it exceeded any single threshold. For the two FSA models, it counted if its composite score crossed the “less healthy” cut-off legislation uses.

Six models, six different answers

The traffic-light label identified 48.2% of ultra-processed products (95% confidence interval 45.8 to 50.6) and 46.2% of processed ones as unhealthy. The FSA model in current law flagged 57.6% (55.2 to 60.0) of ultra-processed products and 50.0% of processed. Its unused 2018 revision reached 68.3% (66.1 to 70.6) and 55.0%.

The three guidance-based models ran much higher: 91.5% (90.2 to 92.9), 94.9% (93.9 to 96.0) and 98.7% (98.2 to 99.3) of ultra-processed products, and 91.6%, 93.1% and 95.0% of processed ones. Every gap between the guidance-based models and the three systems in use was significant at p<0.05. Within each family the answers were closer: for processed products, the three guidance-based models did not differ significantly from one another, and neither did the traffic lights and the two FSA models. The strictest of the three flagged between 43% and 46% more products than the traffic lights, which the authors describe as nearly twice as many.

Where the numbers come apart

Part of the explanation sits in individual nutrients. On the traffic-light scheme, 22.7% of ultra-processed items carried a red for sugars, 12.1% for sodium, 28.6% for saturated fat and 22.4% for total fat. Applying the Eatwell thresholds to the same products produced 51.9%, 51.6%, 47.1% and 48.7% respectively.

Part of it sits in what the label never shows. The traffic-light scheme has no colour for energy density, and energy was the single most discriminating criterion in the study: 74.3% of ultra-processed products and 69.1% of processed products exceeded the 125 kcal per 100 g threshold. The paper also notes that the traffic lights grade total sugars rather than the free sugars UK guidance uses, that the low cut-offs were borrowed from European Union nutrition-claim legislation, and that the reasoning behind the red cut-offs is “unclear” in the scheme’s own documentation.

The exemptions, and the offset

The FSA model is narrower than its high-fat-salt-sugar nickname suggests. It applies to 13 categories of packaged goods, and the paper lists savoury pastries such as sausage rolls and chocolate-based nut spreads among the products outside its reach. Its arithmetic is compensatory too: points for protein, fibre and fruit, vegetables and nuts are subtracted from points for energy, saturated fat, sugar and salt. The authors argue that the evidence for letting one offset the other is lacking, and that UK dietary recommendations contain no such instruction.

Both points land on regulation rather than on shelves. New restrictions on advertising less healthy food on television and online came into force on 5 January 2026, and they lean on this score.

What the 91% figure is not

The eye-catching numbers come from models the authors built themselves, and those models were designed to be sensitive. A single exceeded threshold is enough to mark a product; the authors describe their first model as the most conservative of the three and still report more than nine in ten products flagged. Adding energy density pushes the figure towards 99%, and the paper is explicit that this criterion would capture foods almost nobody would call unhealthy — unsalted nuts among them — if applied without care about which items are in scope.

So 91% is not a measurement of how much of the food supply is harmful. It is a measurement of how far the current tools sit from the dietary guidance, on the authors’ reading of both.

Their competing interests bear on the subject. One author receives royalties from a self-published book that discusses ultra-processed food, plus consultancy payments; another receives royalties for a book on the topic and has been paid for broadcasting about it; a third reports honoraria from food and pharmaceutical companies and a stake in a health clinic; a fourth is a staff member of the Pan American Health Organization. The journal reports the paper as externally peer reviewed and not commissioned.

What the study cannot say

Nothing here follows a person. The unit of analysis is a nutrient record for a packaged item, so there are no health outcomes, no diagnoses and no evidence that relabelling a product would change what anyone eats. Minimally processed foods were excluded, which removes the comparison group that would show whether these models can tell a good diet from a bad one.

Two practical caveats cut in opposite directions. Traffic-light labelling is voluntary, so in a real shop fewer products carry it than the analysis assumes, which makes the scheme look more consistent than it may be; and because the FSA model’s 13 categories could not be mapped onto the survey database, the paper says its count of FSA-flagged products may be an overestimate of what regulation actually reaches. Nova itself is contested — the paper cites criticism that it is ambiguous and difficult to implement, and notes that UK government members have argued the category is too broad for policy use. That claim is precisely what the study set out to test.

Where the American rules stand

The US version of this argument is still at the definition stage. The FDA’s ultra-processed foods page, last updated in September 2025, says the agency and the US Department of Agriculture issued a joint request for information on 24 July 2025 to help develop a uniform definition of ultra-processed foods, with the comment period extended to 23 October. The same page states that an estimated 70% of the US food supply is made up of foods commonly considered ultra-processed and that children get more than 60% of their calories from them, while noting that many questions about how and why these foods may harm health remain unanswered.

On labelling, the FDA’s front-of-package page — current as of May 2025 — describes a proposed “Nutrition Info box” that would print saturated fat, sodium and added sugars on the front of most packaged foods as Low, Medium or High. Those are the same three nutrients behind the UK’s strictest guidance-based model. Calories would not appear in the box, though manufacturers could declare them voluntarily. The proposal drew in part on a 2023 agency experiment with nearly 10,000 US adults, in which a black-and-white design carrying percent Daily Values performed best.

What a label does tell you

The Nutrition Facts panel is the more detailed document, and its limits matter. MedlinePlus notes that percent Daily Values are calculated against a 2,000-calorie diet, that they are a general guide rather than a personal target, and that a serving size reflects what people typically eat — “not a recommendation for how much of that food to eat”. Protein, trans fat and total sugars carry no percent Daily Value at all.

WHO’s healthy-diet fact sheet gives the scale those figures are set against: free sugars below 10% of daily energy intake, saturated fat below 10%, salt below 5 g a day. Those are population reference points, not prescriptions for an individual, which is why a threshold model and a front-of-pack label can both be defensible while flagging different products.

The authors decline to propose a policy, writing that redefining unhealthy food has “far reaching consequences” and needs independent evaluation. Their narrower argument: Britain’s three systems are inconsistent with each other and with its own advice.

What labels leave out is a running theme here — we looked at the same gap in nutrient labelling for plant-based alternatives, and at why ultra-processed food research is moving towards trials.

None of this is a diagnosis or a personal dietary instruction. If you are managing a diet-related condition, or want to know how these thresholds apply to you, that conversation belongs with a clinician or a registered dietitian; our medical disclaimer sets out where the site draws its line.