Home / Editorial process
Editorial process
Last updated: November 20, 2026.
How a label becomes data
Every label figure on a product page comes from the brand’s own label image. The transcription is done by AI: two separate AI transcription passes read the same image, and neither sees the other’s work or any figure we already hold. A product publishes only when the two passes agree on every value shown; where they disagree, a third pass has to agree with one of them, and a figure that is still in doubt is left blank rather than estimated.
For a few brands the figures are instead parsed from the brand’s own machine-readable nutrition table, and the product page says so. Those are the only two ways a product counts as verified.
What people decide
People decide which brands, products and categories the site covers, the scoring weights, every reference dose and the published method. Those decisions are dated under “Method changes” on how we rate.
What runs on its own
Prices are read from each brand’s own store and buyer ratings from the brand’s own product page, both on a daily schedule, and every page states the date its price and rating were read. Before any release, automated checks confirm, among other things, that a page’s claims match its own data.
What we do not do
We do not test products by hand, taste them or send them to a laboratory, and we claim no dietetic, medical or laboratory credentials. Nothing on the site is a health assessment; the scores describe what a label states and what a dose costs.
Corrections
If a figure is wrong, email [email protected] with the page and the label you are looking at. Every correction to a figure we published is dated on the corrections page.