Why this exists.
Because the people paid to warn you about bad food publish once a week, in PDF, in a dialect only lawyers speak, and then act surprised when nobody reads it.
The short version
A company finds Listeria in its cheese. It quietly starts pulling product. Weeks later, the FDA's enforcement report lists it, on a Wednesday, in a table, in a dataset most people have never heard of. The cheese has been in your fridge the whole time. This site reads those tables every morning, rewrites them in English, and tells you where the stuff was sold. That's the whole thing.
Who we're pointing at
The system, mostly, and the people running it. Recalls in this country are overwhelmingly voluntary: started by the company that made the mistake, on the company's timeline, described in the company's words. The agencies that are supposed to be watching have been asked to do more with less for years, and lately the answer to "less" has been "even less." That is an opinion, and it lives on this page and nowhere else on the site.
The delays are not an opinion. We measure them, per notice, from the government's own dates: the day the company started pulling product, and the day the government listed it publicly. The number sits on the home page. If it embarrasses somebody, it should.
Where the data comes from
Every morning we pull recall and outbreak notices from the FDA (openFDA enforcement reports), USDA FSIS (by email, because their website blocks every request from the network this site runs on, which is its own kind of statement), and, as we add them, the CDC, major retailers' recall pages, and regulators in Canada, the UK, and the EU whose products end up on US shelves. Each notice is stored exactly as received. The full list, with timestamps, is on the status page.
What the AI does
A language model reads each notice and pulls out the brand, products, lot codes, hazard, regions, and retailers, then rewrites the summary in plain English. It is told, in no uncertain terms, to extract only what the notice says and never to fill gaps. It rates its own confidence. Anything under 80% waits for a human before it appears here; the count of things waiting is on the home page, because you deserve to know.
What the AI does not do
It never decides whether something is Do not eat, an Outbreak, or a Caution. That label comes from a fixed set of rules based on which kind of notice the source published. Severity comes from the source when the source gives one (FDA Class I, II, III; FSIS recall class); when it doesn't, we say so on the page. The jokes are written by people. The facts are copied from the government.
Caution versus Do not eat
Striped yellow means a regulator flagged something without ordering it off shelves: a public health alert, an import alert, a "do not consume" notice. Red means a recall or an active outbreak investigation. We keep them visually distinct on purpose, and we will never dress up an advisory as a recall to make the list longer.
When we get it wrong
We will. Wrongly saying a product is recalled is the worst thing this site can do, and we treat it that way. Every incident has a report an error link. Corrections are made within 24 hours and logged publicly.
What this site is not
Not a lab. Not a substitute for the linked source. Not medical advice. Not funded by anyone who sells food. If you think you've eaten something on this list and feel unwell, call a doctor, not a website.