Open Food Facts Is Enough To Count The Seven Certified US Food Dyes Product By Product, And The Label Wording Turns Out To Be Regular Enough That A Regex Does It

I wanted a per-product count of the seven FD&C dyes still certified for use in US food, and it turns out the Open Food Facts bulk export is enough on its own. Posting the method and the numbers because I could not find this cut published anywhere.

Source: the Open Food Facts full CSV export, https://world.openfoodfacts.org/data. I used the 2026-08-31 snapshot, 1.275 GB gzipped, 4,535,553 rows. Filter to countries_tags containing en:united-states and require a non-empty ingredients_text, which leaves 444,943 products. Then match the ingredient text against the US number form, the common chemical name and the E number for each dye.

The part I did not expect is that the naming is regular. I went in braced for the inulin problem, where a third of the products containing an ingredient never print its common name and you spend the afternoon chasing synonyms. It is not there. Of the 6,259 products matching any Red 3 form, 23 do not use the plain “Red 3” wording, which is 0.4%. Red 40 is 96 of 28,591, or 0.3%. Yellow 5 is the worst at 0.8%, because tartrazine still turns up. A regex on the number form catches essentially all of it, which is not how ingredient text usually behaves.

There is a regulatory reason for that. 21 CFR 101.22(k)(1) requires a colour additive subject to certification to be declared by the name given in the applicable regulation, and it explicitly allows dropping the “FD&C” prefix and the “No.” So “Red 40” is not shorthand, it is the compliant form. Only colours not subject to certification may be declared under (k)(2) as “Artificial Color”, “Artificial Color Added” or “Color Added”.

That flips something I had assumed going in. 16,768 products use one of those generic phrases, and 10,571 of them also name a certified dye, so in those the generic phrase is covering something else in the same product. 3,914 use a generic phrase and name no colourant at all, and by (k)(2) whatever those are, they are not one of the seven.

Where the label does go dark is (k)(3), which says colouring added to butter, cheese and ice cream need not be declared at all. Most producers declare anyway: 27.5% of the 12,080 cheese records name a natural colourant, mostly annatto. But the exemption is real and it is invisible from the ingredient list.

On timing, FDA revoked the Red 3 authorization on 2025-01-15 under the Delaney Clause and gave food manufacturers until 2027-01-15 to reformulate, so this snapshot sits four months out from the deadline. The 6,256 products span 1,073 distinct first-brands, so it is not one catalogue duplicated.

Limitations, and the second one is large.

• Open Food Facts is crowd contributed. This is a sample of what people scanned, not the US shelf, and popular products are over-represented.
• Records go stale. Only 17.3% of the Red 3 rows were modified after the revocation date, against 43.0% of all rows. Restrict to rows touched since then and Red 3 falls from 1.41% to 0.57%. I am quoting both because neither is the honest single number: the full count includes labels that have already changed, the restricted count drops products nobody has rescanned.
• A string match cannot see inside “natural flavors” or a sub-ingredient it does not parse, so every figure here is a floor.
• I hand-read 12 random Red 3 matches looking for false positives and found none. They are all “fd&c red #3”, “red 3 lake” and similar.

The gap I cannot close from this export is retail availability. Does anyone know a source that would let me weight this by what is actually stocked, rather than by whoever happened to scan it?

Disclosure per rule 1: I write an iOS app called Snack Check that reads these dyes off a barcode, which is why the parser was already sitting there. I am not linking it, and the export above is the only link in this post.