Learn → Module 01
What is food, really?
Food is something that was alive recently, transformed only by methods a home cook recognizes — everything else is an industrial formulation.
13 min read
What is food, really?
TL;DR. The question sounds silly until you walk into a 33,000-SKU supermarket and try to answer it. Three thinkers land on the same line. Michael Pollan calls most of the supermarket "edible foodlike substances" and offers a great-grandmother test. Carlos Monteiro published a testable category in 2009: NOVA Group 4 ultra-processed food, defined by industrial ingredients no home cook owns. Robert Lustig boils the test down to one pair. Real food has fiber. Ultra-processed food has added sugar without it. Kevin Hall's 2019 NIH trial showed that processing itself drives overeating, not the nutrient label. The food matrix (physical structure) does most of the work. This is a sorting tool, not a moral category.
What you'll learn
- Why "food" needs a working definition before any other diet question.
- The 4 NOVA groups, and why the line between Group 3 and Group 4 is the one that matters.
- How Pollan, Monteiro, and Lustig land on the same boundary from different starting points.
- What the 2019 Hall NIH trial proved and what it didn't.
- Why the "food matrix" often matters more than the nutrients on the label.
- Where the framework gets uncomfortable: NOVA purism, fortified foods, food deserts, edge cases.
The trouble with "food"
The American supermarket sells 30,000 different products. Big stores carry 60,000. About 17,000 new products launch every year. Most are built from cheap pieces of corn, soy, wheat, and seed oil. The shelves carry 5 decades of nutrition fads: low-fat, low-cholesterol, high-fiber, low-sodium, low-carb, high-protein, gluten-free, plant-based. Almost none of those changes turned the product back into food.
Take one product. A national grocer sells Soft & Smooth Whole Grain White Bread. The name is a contradiction. The product is real. The legal cover comes from a 1973 FDA decision. Before then, the 1938 Food, Drug and Cosmetic Act required the word "imitation" on any imitation food. In 1973, after heavy lobbying, the FDA swapped that rule for a "not nutritionally inferior" standard. Match the named nutrients of the original and you can drop the word. That one change opened the door to the modern processed-foods aisle. White bread fortified with a token amount of whole grain is now legal to sell as if it were whole-grain bread.
This is the question Pollan opens In Defense of Food with. How did we get to a place where normal adults need a 700-page nutrition manual to answer "what should I eat"? His guess is that we lost the meaning of food itself. Once you reduce food to its nutrients, you can build any nutrient profile from any ingredients. Margarine is the model. First it was low-cholesterol, then low-trans-fat, then "made with olive oil." Same product, different fad. The label moves. The product underneath barely changes. The way out isn't a new nutrient. It's a working meaning of food.
Three converging answers
Three writers came at the question from 3 angles: journalism, epidemiology, biochemistry. All three landed on the same line.
Pollan: edible foodlike substance
Pollan's rule is informal and useful at the shelf. Real food is something your great-grandmother would know. It came from a plant, an animal, or a fungus. It has few ingredients, all pronounceable. If you can picture each ingredient growing, being raised, or being butchered, you're holding food. If you can't, you're holding what Pollan calls (borrowing the food industry's own term) an "edible foodlike substance."
His shelf rules follow from that. Skip products with more than 5 ingredients. Skip ingredients a third-grader can't pronounce. Skip any health claim on the package. Real food has no lobby. A head of broccoli can't pay for an FDA petition. Only industrial formulas need to advertise their virtues. A loud health claim is a sign of processing, not health.
The 7-word version: Eat food. Not too much. Mostly plants. The hardest word in 2026 is the first.
Monteiro: NOVA Group 4
In 2009, Brazilian researcher Carlos Monteiro and his São Paulo team made the question testable. They were trying to explain a Brazilian puzzle. From 1974 to 2005, household buys of oil, sugar, and salt dropped sharply while obesity climbed. The numbers didn't fit a calorie or macronutrient story. What changed was the share of meals coming from factories instead of home kitchens.
NOVA skips calories and nutrient grams. It sorts food by what was done to it. Four groups.
Group 1 — Unprocessed and minimally processed. Fresh and frozen produce, plain rice, dry beans, eggs, raw nuts, milk, plain yogurt, fresh fish and meat. Cleaning, drying, freezing, pasteurizing, grinding, and bagging don't count as processing here.
Group 2 — Culinary ingredients. Olive oil, butter, salt, sugar, honey, vinegar. Things you cook with, not eat alone. In Monteiro's data, a bag of sugar on the counter is a sign of home cooking. Home cooking tracks with lower obesity rates.
Group 3 — Processed foods. Group 1 plus Group 2, combined with methods a home cook knows. Canned beans. Smoked fish. Real cheese. Salted nuts. Sourdough from flour, water, salt, and a starter.
Group 4 — Ultra-processed foods. Industrial formulas built from things pulled out of food: modified starches, refined-bleached-deodorized seed oils, glucose syrups, protein isolates, maltodextrin. Then they're rebuilt with things no kitchen has: emulsifiers (polysorbate 80, carboxymethylcellulose, DATEM, mono- and diglycerides), gums (xanthan, guar, locust bean), low-calorie sweeteners, color stabilizers, and "natural flavorings." Group 4 is also defined by purpose. Shelf-stable. Hyperpalatable. Brand-able. Built to be eaten in volume.
The line that matters is between Group 3 and Group 4. That is the line that shows up, dose by dose, across roughly 250 studies Chris van Tulleken sums up in Ultra-Processed People. Every 10 percent rise in the UPF share of a diet tracks with about a 12 percent rise in obesity risk, a 10 percent rise in breast cancer risk, and a 25 percent rise in dementia risk. The links hold after adjusting for total calories, sugar, salt, fat, fiber, smoking, income, and overall diet pattern. NOVA-4 predicts the harm. Nutrients don't.
Fernanda Rauber, on Monteiro's team, puts it sharpest: Most UPF is not food. It is an industrially produced edible substance.
Lustig: fiber present, sugar absent
Robert Lustig is the UCSF pediatric endocrinologist behind the 2009 "Sugar: The Bitter Truth" lecture. He wanted a test a clinician could run in 10 seconds. His rule from Metabolical: real food has fiber and does not have added sugar. Ultra-processed food has the opposite.
The biochemistry is direct. Fiber slows down glucose absorption. It feeds the gut microbes. It makes short-chain fatty acids that signal "you're full." It forms a physical scaffold that slows digestion. Strip the fiber and a slow carb turns into a glucose flood. Add sugar on top (especially fructose, which Lustig argues acts in the liver more like ethanol than glucose) and you fuel the fire. The combo drives de novo lipogenesis, hepatic insulin resistance, and the cluster Lustig groups under metabolic syndrome: type 2 diabetes, fatty liver, high blood pressure, several cancers, and dementia.
The Lustig, Monteiro, and Pollan rules pick out almost the same products. A whole apple has fiber and no added sugar. Food by all 3 tests. Apple juice from concentrate, sweetened with apple-juice concentrate, has no fiber and added sugar. Not food by any of them, even with a Nutrition Facts panel a parent might glance at and approve.
The Hall trial: this is causal, not correlational
Studies that watch big groups of people can show two things move together. They can't prove one caused the other. For decades, food-industry-funded scientists used that gap to argue the UPF-disease link might come from income, education, or some other hidden factor. The argument fell apart in 2019.
Kevin Hall is an NIH senior investigator. He started out as a NOVA skeptic. He ran a controlled-feeding study on the metabolic ward at the NIH Clinical Center. Twenty adults lived on-site for 4 weeks. For 2 weeks they ate an 80-percent-ultra-processed diet. For the other 2 weeks, an 80-percent-minimally-processed diet. Both diets matched on total calories offered, sugar, salt, fat, carbs, fiber, and protein. People could eat as much as they wanted. The order of the diets was random and balanced. The study ran in Cell Metabolism.
On the UPF diet, people ate about 500 extra calories a day and gained about a pound a week. On the matched whole-food diet, they lost weight. The macronutrients were the same. The only difference was processing. Hall set out to refute Monteiro and produced the strongest piece of causal evidence in the field.
The trial has limits. Twenty people. Two weeks per arm. A metabolic ward isn't real life. The exact mechanism (softness, calorie density, eating speed, flavor-nutrient mismatch, gut microbiome effects) is still being worked out. But the effect is big, in the predicted direction, and came from a skeptic. If you remember 1 nutrition study from this era, this is it.
Why "the food matrix" matters
Anthony Fardet is a French food scientist at Université Clermont Auvergne. He named what the Nutrition Facts panel keeps missing: the food matrix. Food is more than the sum of its nutrients. Physical structure (fiber scaffolding, intact cell walls, embedded fats, water binding) controls how fast molecules reach the blood, how the gut responds, and how the brain reads fullness.
A 1977 study compared whole apples, apple purée, and apple juice. Same sugars and fiber on paper. Three different blood-glucose curves and 3 different fullness scores. Whole apple: slow rise, lasting fullness. Purée: faster rise. Juice: spike, crash, hunger an hour later. Same fruit, 3 matrices, 3 different outcomes inside the body.
The pattern holds up. Steel-cut oats, rolled oats, and instant oatmeal share the same macronutrients but hit blood sugar at sharply different speeds. Whole almonds and almond flour share a label and your gut takes up fat from them at different rates. Sourdough and sliced supermarket bread can match on the label and split inside a person. Label-thinking counts ingredients and ignores how the food is built.
NOVA Group 4 is, in practice, a category whose matrix has been built for speed. Fast to eat. Fast to absorb. Fast to leave the gut and trigger hunger again. That is what the Hall study isolated. The matrix, not the nutrients, drove the 500 calories.
What this means for "ultra-processed" as a category
"Junk food" is a moral label. It mixes Twinkies and frozen broccoli depending on who's talking. NOVA Group 4 is not that. It is a processing label, based on ingredient lists and how food is made, with predictable behavior at the edges.
Frozen broccoli is Group 1. Dried pasta is Group 1. Canned tomatoes are Group 1. Plain Greek yogurt is Group 1. Real cheddar, smoked salmon, and bakery sourdough are Group 3. A "lower-sugar" sweetened cereal with less sodium and added fiber is still Group 4. Tweaking the recipe doesn't move a product between groups. The substances and processes that defined it are still there.
The label also tracks marketing. Almost every product with a front-of-pack claim ("supports immunity," "good source of fiber," "made with real fruit," "heart-healthy whole grains") is Group 4. Real food has nothing to prove on the front of the box.
The nutritionism trap
Pollan's word for the operating system of the last 50 years is nutritionism. The idea that food equals the sum of its nutrients. Four hidden axioms: nutrients are what matter, nutrients are invisible so experts must read them for you, the point of eating is body health, and nutrients split into heroes and villains. Once an idea is dominant, it turns into weather. You don't see it while it rules.
Nutritionism has a birthday. In 1977, the McGovern Senate Committee drafted Dietary Goals telling Americans to eat less meat. The beef and dairy lobbies pushed back hard. McGovern, a senator from cattle-ranching South Dakota, blinked. The line was rewritten as choose meats, poultry, and fish that will reduce saturated fat intake. With that swap, U.S. diet advice stopped naming foods and started naming nutrients. Nutrients have no voters. The food industry, which can rebuild any product around any nutrient fad, got a permanent license to invent new kinds of processed product.
Every fad since (low-fat, low-cholesterol, low-trans-fat, low-sodium, low-carb, high-protein, high-fiber, plant-based) plays by nutritionism's rules. Each one filled the shelves with a new wave of reformulated products. None reversed the chronic-disease trend. The fix isn't a better nutrient. It's leaving the frame.
Three honest nuances
None of this should turn into religion. Three places where it gets uncomfortable.
Some ultra-processed food is genuinely useful. Fortified infant formula for non-breastfed babies. Vitamin-fortified flour that prevents neural tube defects. Oral rehydration solutions. Shelf-stable disaster-relief foods. All NOVA-4. The category predicts harm at the population level. It isn't a verdict on any single product. Don't be a NOVA purist.
Access and income matter. The framework should help people, not gatekeep them. Tens of millions of Americans live in places where Group 4 is the only food within walking distance. Calling that personal failure is the wrong move. NOVA works best as a policy tool: warning labels, marketing limits on kids' products, school-food rules, sugary-drink taxes. Not as a test at the checkout.
The boundary is real, but a few products are debatable. Plain Greek yogurt and fruit-on-the-bottom yogurt sit on opposite sides of a real line. Minimally processed deli meats, some plant-based milks, and certain artisan items rebuilt for shelf life sit closer to the edge. The arguments at the edges don't sink the framework. They're the friction every useful category produces.
Frequently asked questions
If NOVA is testable, why don't dietary guidelines use it?
Brazil's 2015 dietary guidelines use it openly. Several Latin American countries followed. U.S. and U.K. guidelines do not, even though FAO and PAHO adopted NOVA. The structural answer is industry capture. U.S. guidelines come from a committee with heavy food-industry funding and then get softened by an interagency process. Marion Nestle documents this in What to Eat Now (2025). Naming "ultra-processed food" as the problem points at most of the U.S. food economy.
What's the difference between "processed" and "ultra-processed"?
"Processed" is a casual word covering everything from a cube of cheese to a Dorito. NOVA splits it. Group 3 uses methods and ingredients a home kitchen has: salt, fermentation, smoke, oil, heat, time. Group 4 uses extracted fractions and substances no kitchen has. The harm signal sits on Group 4.
Is bread NOVA-4?
Depends. A 4-ingredient bakery loaf (flour, water, salt, yeast or starter) is Group 3. A supermarket sliced loaf with mono- and diglycerides, DATEM, sodium stearoyl lactylate, soy lecithin, calcium propionate, and added sugars is Group 4. Same shelf, opposite groups.
Are plant milks always NOVA-4?
Almost always. Typical lists include canola or sunflower oil, gellan gum, dipotassium phosphate, plus added sugars and flavors. A few brands sell minimally processed versions (water, nuts, salt). Plain dairy milk is Group 1.
How is the Hall trial different from previous obesity research?
Most prior work was observational (correlation, with hidden factors) or focused on single nutrients (low-fat vs. low-carb, low-glycemic vs. high-glycemic). Hall matched the nutrients and varied the processing. He isolated processing as a causal variable. No prior controlled-feeding study had done that.
Doesn't the food-matrix concept just mean "whole foods are better"?
Yes, with a reason. "Whole foods are better" is a slogan. The food-matrix research explains why. Fiber scaffolding, intact cell walls, water binding, and embedded fats produce slower absorption, longer fullness, and different microbiome signals. Mechanism turns a rule of thumb into a claim you can test.
Is nutritionism the same as nutrition science?
No. Nutrition science is the study of how food affects bodies. Nutritionism is an ideology that treats food as nothing but its nutrients and uses that to license endless reformulation. Pollan's critique aims at the ideology, not the science. Good research (the Hall trial included) is taking nutritionism apart from inside.
Is a Beyond Meat or Impossible Burger NOVA-4?
Yes. The ingredient list (pea or soy protein isolate, refined coconut and canola oil, methylcellulose, natural flavors, modified food starch) is textbook Group 4. "Plant-based" doesn't mean "real food." A bowl of lentils is plant-based. A Beyond Burger is a plant-derived industrial formulation.
Sources
- Pollan, M. In Defense of Food: An Eater's Manifesto. Penguin Press, 2008. Nutritionism, the 1973 imitation-rule repeal, the great-grandmother test, the seven-word manifesto.
- Pollan, M. Food Rules: An Eater's Manual. Penguin, 2009. "If it came from a plant, eat it; if it was made in a plant, don't."
- Monteiro, C. A., Levy, R. B., Claro, R. M., Castro, I. R. R., & Cannon, G. (2010). "A new classification of foods based on the extent and purpose of their processing." Cadernos de Saúde Pública, 26(11), 2039–2049. doi: 10.1590/S0102-311X2010001100005. (Earlier formulation: Monteiro 2009, Public Health Nutrition, doi: 10.1017/S1368980009005291.)
- Hall, K. D., Ayuketah, A., Brychta, R., et al. (2019). "Ultra-processed diets cause excess calorie intake and weight gain: an inpatient randomized controlled trial of ad libitum food intake." Cell Metabolism, 30(1), 67–77.e3. doi: 10.1016/j.cmet.2019.05.008.
- van Tulleken, C. Ultra-Processed People. W. W. Norton, 2023. NOVA-4 mechanisms, the Hall trial reception, the additive vocabulary, the Rauber line.
- Lustig, R. H. Metabolical: The Lure and the Lies of Processed Food, Nutrition, and Modern Medicine. Harper Wave, 2021. "Protect the liver, feed the gut"; fiber-present, sugar-absent as the operational test.
- Fardet, A. (2016). "Minimally processed foods are more satiating and less hyperglycemic than ultra-processed foods: a preliminary study with 98 ready-to-eat foods." Food & Function, 7(5), 2338–2346. The food-matrix concept.
- Schatzker, M. The Dorito Effect. Simon & Schuster, 2015. Why flavor became something engineered onto bland industrial substrate, and what that does to appetite.
- Nestle, M. What to Eat Now. Farrar, Straus and Giroux, 2025. Supermarket politics, the U.S. guidelines process, why NOVA hasn't made it into U.S. policy.