GreenCheck

Where GreenCheck's food data comes from

Updated July 15, 20261 min read
The short answer

Product names, ingredient lists, allergen tags and nutrition facts come from OpenFoodFacts, an open database licensed under the ODbL. The 1-to-10 rating for each ingredient comes from an AI analysis pass. The source links attached to each ingredient are added by our servers from a fixed list of authoritative domains, never generated by the model.

There are three separate sources behind a GreenCheck grade, and they are worth telling apart — because they fail in different ways.

1. Product facts: OpenFoodFacts

Product names, brands, ingredient lists, allergen tags and nutrition data come from OpenFoodFacts, a collaborative, open-data project.

OpenFoodFacts data is made available under the Open Database License (ODbL), and its product photographs under CC BY-SA 3.0. We are grateful for it — a project like this simply isn't possible without it.

Because anyone can contribute to OpenFoodFacts, its data is uneven. That's the trade for having an open food database at all, and it's why every product page shows the date its ingredients were last checked.

2. Ingredient ratings: an AI pass

Each ingredient is rated 1 (safest) to 10 (worst) by an AI analysis pass, which also writes the plain-English explanation you see next to it. That rating feeds the product score described in how we grade food.

The links under each ingredient are not produced by the model. They are attached server-side from a fixed list of authoritative domains — PubMed, the NIH, the FDA and similar — and checked for reachability before being cached.

This split is deliberate. A model asked for a citation will happily invent one that reads perfectly and leads nowhere. On a page about what a child eats, that is not an acceptable failure mode.

What this means for you

Common questions

Can OpenFoodFacts data be wrong?
Yes, and sometimes it badly is — anyone can edit it. While building our catalog we found a bar whose entire ingredient list read "gjdwf" and a baby food whose nutrition panel had been scanned into the ingredients field. We screen for this and hand-check catalog products, but it is exactly why the label in your hand beats any database.
Does the AI make up the source links?
No. This is deliberate. Language models can produce citations that look plausible and do not exist, so citations are never left to the model — our servers attach them from a fixed list of authoritative domains, and dead links are stripped before anything is cached.
Why does the app sometimes not know my product?
Because OpenFoodFacts does not have it. Coverage is strongest for widely-sold packaged goods and thinnest for regional or store brands. When a barcode isn't found, you can photograph the ingredients panel instead.

GreenCheck provides information to help you decide; it is not medical advice. Always read the product label and consult a professional for allergy and medical decisions.