← Longevity & Supplement Guides
Health claims on food labels: true sentences, missing paragraphs
Pick up almost any new box in a supermarket and it will talk to you before you have read a single number. Reduced sodium. No trans fat. Organic. Zero sugar. The front of the package has become a small billboard, and it is very good at its job.
This week a team at Cornell University published the largest check I have seen of whether that billboard tells the truth about the whole product. They looked at close to 150,000 foods and drinks launched in the United States over the past decade. About half carried some kind of health-related claim. Only about a third of the launches passed an independent test of overall nutritional quality.
The headline most outlets ran was that “healthy” labels mislead. That is almost right, and the small gap is the interesting part. The claims were not lies. They were true sentences with the rest of the paragraph left out.
What the Cornell team actually measured
The study, by Pei Zhou, Miguel Gómez and colleagues in Nature Food, used a market-research database of new product launches and scored every item with the nutrient-profiling model developed by the UK Food Standards Agency. That model does something a front-of-pack claim never does: it looks at the whole recipe at once. It counts energy, sugar, saturated fat and sodium against the product, and fibre, protein and fruit or vegetable content in its favour.
Then the researchers asked a simple question. Do products that make a claim score better on the whole-recipe test than products that make none? Mostly, not reliably. In some categories, the products with the claim scored worse. The gaps were largest in drinks. According to Cornell’s summary of the work, reduced-sodium sodas made up the difference with sugar, and meals that advertised no trans fat often carried more saturated fat and sodium instead.
This is an observational study of labels, not a trial of what people ate, and nutrient-profiling models have their own arguments built in. But as a map of what the packaging industry chooses to say, it is hard to argue with the size of the sample.
A claim is a true sentence, not a verdict
Here is the part the outrage headlines skip. Most on-package claims are regulated, and they are regulated for accuracy about the one thing they name. If a drink says “reduced sodium,” it must contain less sodium than the reference product. It does. Nobody checked the sugar, because the sugar was never the subject of the sentence.
That is the mechanism the study exposes. Take one ingredient out of a recipe and something usually has to go back in to keep the taste, the texture or the price where the product team wants it. Fat leaves and sugar arrives. Salt leaves and sweetness arrives. The claim describes the subtraction. The addition gets no billboard.
So the useful correction is not “claims are false.” It is “claims are partial.” A true statement about one nutrient is a very different object from a judgment about the food, and the front of the box is designed so that you confuse the two.
Figure
The front answers one question. The back answers the rest.
“Reduced sodium” on a drink
True about salt. In the Cornell data, drinks with sodium claims were among the widest gaps between claim and overall quality, with sugar often filling the space. Check the sugars line.
“No trans fat” on a meal
True about one kind of fat. Foods with this claim often scored worse overall, with more saturated fat and sodium. Check both lines.
“Organic” or “non-GMO”
A statement about how the food was produced. It makes no promise about sugar, salt, fat or energy. Read the panel as if the word were not there.
“Zero sugar”
True about sugar. It says nothing about sodium, saturated fat or the rest of the recipe. Compare the whole panel with an ordinary version of the product.
The halo does the rest of the work
A partial truth would be harmless if we read it as partial. We do not. In a small set of experiments published in 2010 in Judgment and Decision Making, Jonathon Schuldt and Norbert Schwarz showed people cookies described as organic or conventional. Participants judged the organic cookies lower in calories and fine to eat more often, even when the nutrition information was identical.
The strangest finding was the second one. When people evaluated someone with a weight-loss goal, skipping exercise seemed more acceptable if that person had just chosen the organic dessert. A word about farming practice had leaked into a judgment about the treadmill.
These were student samples in a lab, and one set of experiments does not describe a whole population. But they give the Cornell numbers a mechanism. A claim does not need to be false to mislead. It only needs to be the one thing you read.
What happens when the label talks about the whole food
There is a natural experiment on the other side. In 2016 Chile introduced black warning octagons on the front of packaged foods that exceeded set limits for sugar, sodium, saturated fat or calories, together with marketing restrictions and limits on school sales. The label described the whole recipe, not the best ingredient.
A before-and-after study of more than 2,000 households by Lindsey Smith Taillie and colleagues in PLOS Medicine found that purchases of drinks carrying the warnings fell by about 24% after the law took effect. A later follow-up analysis reported that, three years in, households were buying roughly 37% less sugar and 22% less sodium than the pre-law trend predicted.
The caveats are real. The law bundled several policies together, so the label alone cannot take all the credit, and the authors note that their estimates shift with the choice of baseline. Still, the direction is consistent: when the signal on the front describes the whole food, buying moves. In the United States, the FDA’s updated definition of “healthy” moves slowly in the same direction, with limits on added sugar, sodium and saturated fat that manufacturers must meet from 2028.
Our own labels, read the same way
We sell supplements, and our labels make claims too, so this applies to us. An authorised claim such as “vitamin C contributes to the normal function of the immune system” is a true statement about the normal role of one nutrient. It does not tell you whether you need more of it, and it is not a verdict on your diet. It is a sentence, not a paragraph.
That is also why we think about evidence the way we do. Our Agen Evidence Score exists because a single true fact is easy to print and hard to weigh. The same habit that helps with a supplement label helps with a cereal box: ask what the sentence leaves out.
One thing to try this week
Pick one product you buy often that carries a claim on the front. Turn it over. Read four lines on the nutrition panel: energy, sugars, saturated fat and sodium. Then find the ordinary version of the same product, the one without the claim, and read the same four lines per 100 grams, not per serving, so the portions cannot hide anything.
Sometimes the claimed product wins on every line. Good; keep it. Sometimes it wins on the named line and loses on another. That is the trade the Cornell team found, and now you have seen it for yourself. The ingredient list helps too: ingredients are listed by weight, so a sweetener or a fat near the top matters more than a promise in large type.
None of this needs an app or a spreadsheet. It needs about thirty seconds, once, per product. After the first few, you will start to recognise which claims your usual shop uses as decoration. If you want the bigger picture of how processing and recipes shape a food, our guides to ultra-processed food and fibre versus protein trends go further.
The bottom line
The Cornell study does not show that food companies lie on their packaging. It shows something more useful: that the most visible sentence on a box is chosen for what it says, not for what it represents. About half of new products make a health-related claim. Far fewer pass a test of the whole recipe.
Claims are not the enemy. They are just small. Read them as the first line of a longer story, turn the box over for the rest, and let the numbers correct the billboard rather than the other way round.
