How GetFoodFacts is made

USDA FoodData Central, the release of April 30, 2026, is the source of every nutrient value. It holds two kinds of record, and every page says which kind it shows.

Packaged products (the Global Branded Food Products Database) are the label values that manufacturers and data providers submit to USDA: the amounts per 100 grams, the serving size and household measure, the barcode (UPC), the ingredient statement, the manufacturer's category and the dates of the record. USDA publishes them; it does not test the products. GetFoodFacts shows 398,909 of them with a full page: one per barcode, and only records that name a brand or owner, carry energy and at least five of the seven core label values (energy, protein, fat, carbohydrate, fiber, sugars, sodium), declare a serving between 0 and 500 grams or milliliters in grams or milliliters, and hold together (no macronutrient above 100 grams per 100 grams, sugars and fiber within the carbohydrate, energy within reach of what the protein, fat and carbohydrate supply). Where a label that passes still declares more saturated fat than total fat, or more added sugars than total sugars, its page shows the label as submitted and says above the panel's source that the two figures cannot both be true.

Reference foods (SR Legacy, Foundation Foods and the Food and Nutrient Database for Dietary Studies) are USDA's own values from laboratory analysis and from the foods people report eating in national dietary surveys, with the household portions USDA publishes for each food (a cup, a medium fruit, a tablespoon). There are 13,694 of them.

Food prices are the Bureau of Labor Statistics' U.S. city average prices per unit and its Consumer Price Index series for food, fetched from the BLS API each month.

Every product and food page gives its FoodData Central ID and links the record on fdc.nal.usda.gov, where USDA shows its current values per 100 grams (for most products the same as the release file; see the check below). The per-serving figures on our Nutrition Facts panel are computed from those values and the declared serving size, and rounded the way FDA requires labels to round (calories to the nearest 5 or 10, fat to the nearest half gram under 5 grams, sodium to the nearest 5 milligrams up to 140). The release file can hold a record USDA has corrected since, so at each release every product that appears in a ranking's top 100, and every product page on Google's surface, is read again from the FoodData Central API: where USDA has corrected the record, the corrected values are used, the rankings are rebuilt, and the product's page says so with the date read; a record USDA has withdrawn leaves the rankings. (The "label values" in USDA's download are for most products calculated back from the per-100 g values, so they cannot confirm them.) Every other product page shows the release's values. The package in your hand remains the authority.

The database is rebuilt by a scripted pipeline. It stops if a rebuild would drop rows that it cannot load again, and it flags any source file older than its publisher's schedule, so a stale release is visible before it is published. At every release a script draws products and reference foods at random and compares their values with the FoodData Central API; a mismatch stops the release.

Duplicate submissions are folded: when USDA holds several records for one barcode, the page shows the newest one that meets the conditions above and lists how many there are; the others point to it. Servings above 500 grams or milliliters are package weights, not portions, so those products are compared per 100 grams only. Placeholder brand names ("NOT A BRANDED ITEM", "N/A") are replaced by the company that submitted the record, or read "brand not named" when that is a placeholder too, and one brand is one brand whatever its capitalization. Identical products under several barcodes (the same owner, name, serving and label values) are folded into one page the same way.

Categories and aisles. A product's category is the one its manufacturer assigned in USDA's database; spelling variants of the same category are merged. The aisles that rank products by a nutrient add a name filter where the category is broader than the aisle (protein bars are bars whose name says protein), and leave out forms not eaten as sold: powders, mixes, concentrates, bases, drops. Aisles are numbered by size; aisle 1 holds the most products.

Rankings. An aisle is ranked per 100 grams as each manufacturer submitted the values, except that single-serve aisles (yogurt, protein bars, protein shakes) rank their "most" lists by the amount in one serving as sold, and say so, one product per UPC and one per product sold under several UPCs with the same label; the top 100 are read again from USDA's live record at each release and USDA's corrections applied. Left out of every ranking: placeholders (zero or near-zero energy in an aisle whose median has 50 calories or more); mixes, powders and concentrates (over 3 or 4 times the median energy of an aisle of drinks, soups or milks, or served in under a third of the aisle's typical serving, or in 15 grams or less where the typical serving is 45 grams or more); records the publisher's review names, each with its reason, and values above a ceiling taken from USDA's reference food for the aisle, printed with its reference on the rankings it applies to (bacon, hummus, jerky, canned beans, rice). A "most" ranking also leaves out values above 1.25 times the aisle's 99th percentile and labels whose protein, fat and carbohydrate supply over a quarter more energy than declared. A "least" ranking admits only products that declare all seven core values, so an undeclared value never passes as zero, and leaves out: a zero fewer than 5% of the aisle declare unless the name claims it ("sugar free", "no salt added"); a zero from a serving under 15 grams where the aisle's median is well above zero (the label's rounding); a label that contradicts itself (calories under half, or for carbohydrate and sugar rankings over a quarter above, what its protein, fat and carbohydrate supply; sugars above carbohydrate); a declared serving over three times the aisle's typical one (a whole package entered as the serving); calories under 0.6 times the aisle's 5th percentile, or sodium in a savory aisle under half of it, unless the name claims it; a record that cannot be what its name says (under a quarter of the median energy of an aisle of foods, salt among the first seven ingredients with under 100 milligrams of sodium per 100 grams, a sugar among the first three with under 1 gram of sugars). Each ranking states how many products each rule left out; ties are listed by fewer calories ("most") or more protein ("least").

Percentiles. "Less sugar than 72% of yogurts" means that 72% of the products in its category have more sugar per 100 grams. Categories with fewer than 50 products get no percentile.

FDA claim thresholds. We count how many products meet a claim's threshold per labeled serving: low sodium is 140 milligrams or less, low calorie 40 calories or less, sugar free under half a gram, high in protein 10 grams or more (20% of the Daily Value), high in fiber 5.6 grams or more. FDA sets these per reference amount customarily consumed; most labeled servings follow it since the 2016 label rule, but not every product's does, and a count is not a claim the manufacturer makes.

Typical portions. The nutrient lists rank USDA's reference foods by what a portion people eat contains. The portion is USDA's NLEA serving where it lists one (the serving FDA's reference amounts define); otherwise the everyday household measure USDA lists (a cup, 3 ounces, a medium fruit, never a pound, a bag or a package) closest to FDA's reference amount customarily consumed for that kind of food (21 CFR 101.12: 30 grams for cheese and nuts, 85 grams for meat and fish, 140 grams for fruit), or, for a kind without one, to a 150-calorie serving. A food with no measure near that amount has no typical portion. One food appears once per list, in its best form.

Everyday foods. The portion ranking leaves out what is not eaten as a portion: seasonings, sauces, syrups, powders, dried and concentrated forms, raw meat and fish, uncooked grains and dry beans, restaurant and fast-food items, game and organ meats other than liver and heart. They stay in the per-100-gram view: nobody eats a cup of dried thyme.

The analyses are computed from the same records at each release and state their populations, rules and limits next to their findings; their tables are downloadable.

Net carbs are total carbohydrate minus dietary fiber. FDA does not define the term; we use it only because readers search for it, and say so where it appears.

GetFoodFacts is built with AI tools, which write the code, the page templates and the explanatory text of the guides and studies. The AI does not decide any number: every figure is the result of a database query over the files above, the sentences on product and food pages are assembled from each record's own values, and each study states its method and its limits next to its finding. We check the data with automated tests against the source files rather than by reading every page, which is why each page shows where its facts come from and how to verify them.

Every packaged product that is its barcode's record is in the A to Z product directory, with its brand, its barcode and its calories, sugars, sodium and protein per 100 grams. A product gets a page of its own when it can be placed against its category: a full label (calories, protein, fat, carbohydrate, sugars, sodium and an ingredient statement) in a category of 50 or more products, on USDA's current record for that product; a product readers already reach from a search engine keeps its page too. Every other product's address leads to its line in the directory, and so does a brand with fewer than five products or a maker whose page would only repeat one brand's. An older USDA re-sample of a reference food leads to the newest record of the same description.

Before each release, figures on every page type (products, the directory, reference foods, common foods, aisles, rankings, lists, brands, makers, prices and the research pages) are checked against the USDA FoodData Central and BLS files they cite, and the pages are read in separate review passes against Google’s and Bing’s published quality guidelines, for claims a manufacturer could dispute, and for any figure that reads differently on two pages. The studio runs these checks with AI tools and records each one; the last full review was for the release of October 8, 2026. GetFoodFacts states what labels and USDA analyses report; it is not dietary or medical advice.

A packaged product's record is what its manufacturer submitted, often years ago: most records in the current release were last modified between 2017 and 2023, and recipes change. Missing values mean the record did not declare them, not that the food has none. Reference foods are averages of samples, and a food you buy will differ. Nothing here measures taste, price at your store, how a food affects your health, or whether it suits a medical diet.

The database is refreshed when USDA publishes a new FoodData Central release (about twice a year) and the prices monthly. To report an error, write to hello@getfoodfacts.com; the corrections page explains what happens next.