Nobody has ever had to build willpower against a raw potato. The foods people describe fighting are a narrow and consistent set, and they are the ones that were designed the most carefully.
The short answer
You crave junk food because most of the food supply is optimised to be craved, and there is now a numeric definition of what that optimisation looks like. In a 2019 paper in Obesity, Fazzino, Rohde and Sullivan translated the loose phrase hyper-palatable into thresholds, then applied those thresholds to the Food and Nutrient Database for Dietary Studies. 4,795 of the 7,757 foods in it qualified. That is 62 percent.
What hyper-palatable means numerically
The definition came out as three clusters, and reading them is more informative than any amount of general commentary about processed food. Cluster one is fat and sodium: more than 25 percent of energy from fat, and at least 0.30 percent sodium by weight. Cluster two is fat and simple sugars: more than 20 percent of energy from fat and more than 20 percent from sugar. Cluster three is carbohydrate and sodium: more than 40 percent of energy from carbohydrate and at least 0.20 percent sodium by weight.
What those combinations have in common is that they are rare in nature. Foods that are simultaneously fatty and sweet are almost entirely a manufacturing achievement. The pairings are the product.
62 percent of the database
The 62 percent figure is the part worth carrying, because it relocates the problem. This is not a shelf of obviously indulgent items sitting apart from normal food. The paper specifically noted that the criteria captured foods nobody thinks of that way, including items labelled reduced or low fat, and vegetables cooked in creams, sauces or fats.
So the ordinary experience of walking through a supermarket and finding most things appealing is not a failure of discipline. It is an accurate response to what is there.
What the inpatient trial showed
The strongest causal evidence comes from a study that removed almost every confounder by removing the outside world. A 2019 trial in Cell Metabolism admitted 20 weight-stable adults to the NIH Clinical Center and randomised them to two weeks of ultra-processed food followed by two weeks of unprocessed food, or the reverse. Meals were matched for presented calories, energy density, macronutrients, sugar, sodium and fibre. Participants were told to eat as much or as little as they wanted.
They ate substantially more during the ultra-processed weeks. Weight change tracked intake closely, with participants gaining 0.9 kg during the ultra-processed phase and losing 0.9 kg during the unprocessed phase. Same people, four weeks apart, matched nutrients.
Twenty participants is small and it was a single centre, so this is one strong study rather than a settled literature. But the design is about as clean as feeding research gets, and the result points the same way as the palatability work.
The part of the 62 percent that surprised the researchers
The headline number is striking, but the composition is what makes it useful. The criteria did not simply re-identify the foods everyone already suspected. They pulled in items marketed as the sensible choice.
Reduced-fat products qualified, because removing fat from a formulation frequently means adding sugar or sodium to keep it edible, and the carbohydrate-and-sodium cluster catches exactly that. Vegetables prepared in cream, sauce or fat qualified, which covers a great deal of what gets served as a vegetable side. The definition, in other words, cuts across the categories shoppers use.
That is why the personal-discipline framing keeps failing in practice. A person avoiding the obvious offenders is still eating from a supply where most of the options meet the same criteria, and then concludes from continued cravings that something is wrong with them specifically.
Why willpower is the wrong frame
If the food is engineered and the supply is saturated, then treating each individual craving as a test of character is both inaccurate and expensive. Inaccurate because the craving is the designed output. Expensive because the guilt that follows a lost test makes the next episode heavier, and that pattern is documented rather than motivational.
The more useful frame is environmental. What is in the house, what the default lunch looks like, whether the day contains a real meal before the evening. Those are decisions made once and then not re-litigated at nine in the evening while tired. If the fight itself has started to feel like dread, food anxiety is its own problem and needs different handling.
What this does and does not say about food addiction
It is tempting to jump from these findings to addiction language, and worth resisting. The evidence supports the claim that certain foods reliably drive higher intake and are built to. It does not settle whether that constitutes addiction in the clinical sense, which remains genuinely contested among researchers.
The practical difference matters. Addiction framing pushes toward total abstinence, and the craving research suggests strict restriction of a specific food tends to raise craving for that food in the short run. A person who eats these foods sometimes, deliberately, tends to think about them less than a person who has sworn off them. What a healthy relationship with food looks like covers where that line sits.
What to do with an environment you did not design
You cannot fix the supermarket. You can make the two or three upstream decisions that decide how loud the evening gets, and you can stop scoring yourself on an unwinnable test.
It is also worth separating this from anything medical. If your appetite changed sharply after starting or stopping a medication, that is a question for your prescriber, not something to reason about from a food-supply statistic. The environment argument here is about groceries and defaults, which is the part that is genuinely yours to arrange.
OffRamp is built around that idea. There is a protein floor, set at 1.6 grams per kilogram of goal weight and rounded to the nearest 5 grams, because a day that hit its floor produces fewer of these evenings. There is Regain Radar, which reads a 7-day rolling average against your baseline and only speaks up when that average holds at or above baseline times 1.02 for seven consecutive days. And there is no food in the app that comes back marked as bad, on any day, ever. Given what the 62 percent figure implies about the aisle, an app that flagged hyper-palatable food would spend its life shouting, and would be useless within a week.
