Case studies

We ran two real supplement catalogues through the engine before selling it to anyone — 13,933 products in total. Most of the copy in both had to be rewritten, nearly all of it for the same reason: the text made health claims that the EU register does not permit in those words.

Both write-ups follow the same shape: what the scan found, what the rewrite did, what it refused to touch, and what a second pass through the same engine said afterwards — including the parts that did not come out clean.

Catalogue #1 — a 4,613-product supplement shop

76% of the catalogue had its copy rewritten (3,516 of 4,613 products) — almost all of it because the text carried health claims. Prohibited disease wording fell 685 → 96 findings, and unauthorised health claims 11,364 → 5,732.

Catalogue #2 — 9,320 products, twice the size

65% of the catalogue had its copy rewritten (6,017 of 9,320 products). Unauthorised health claims fell 22,773 → 13,094 findings and prohibited wording 435 → 209, with 3,230 products moving from flagged to clear.

The catalogues are anonymous and product names are masked. The numbers are not.