A translated health claim is a new claim — and usually an illegal one

Regulation (EC) No 1924/2006 does not ask whether a health claim is true. It asks whether it is on the register, worded the way the register words it. That is a strange rule until you have to work with it, and then it becomes a very practical one: compliance is a string comparison, not a judgement call.

Which leads somewhere most shops discover too late. The register is not written in English and then translated by whoever needs it. It is adopted in every official EU language at once, and each language version is the authorised text for that language. There is no master version to translate from.

The same claim, four times, from the register

Here is one authorised claim as the regulation itself writes it. These are not our translations — they are copied from the language versions of Regulation (EU) No 432/2012:

Now take the English sentence and run it through any competent translator, human or machine, into German. You will very likely get something close to “Vitamin C unterstützt die normale Funktion des Immunsystems”. It is correct German. It means exactly the same thing. It is not the authorised wording, and publishing it is an unauthorised health claim.

A faithful translation of an approved claim is not an approved claim. It is a new sentence that happens to mean the same thing — and meaning is not what the register authorises.

Why this scales badly

A single product page might carry three or four health claims. A supplement catalogue of a few thousand products carries tens of thousands. Translate that catalogue into one new market and every one of those claims is re-created as a sentence that is not on the register. Translate it into ten markets and the problem is multiplied by ten, in languages nobody at your office reads.

This is what makes machine translation and shop translation plugins actively risky for supplements specifically. They are good at their job. Their job is to produce natural target-language prose, and natural prose is exactly what the register does not want.

What has to happen instead

The claim must not be translated at all. It has to be recognised in the source, matched to its row in the register, and replaced with that row’s wording in the target language, character for character. Everything around it — the selling copy, the headings, the product descriptions — is translated normally.

Three other things have to survive untouched, and they are the ones a prose translator has no reason to protect: the numbers, the mandatory label wordings, and the HTML. A dose that shifts from 1,500 mg to 1.500 mg because a decimal separator was localised is a labelling error. A nutrition table whose structure was rebuilt is a page that no longer imports.

What that looks like measured

We built this for our own group’s stores before selling it to anyone. The numbers here are counted from the delivered files, not estimated: 91,560 rows across 20 EU languages, of which 96% came back ready to publish as delivered.

Two properties hold on every delivered row, and they hold by construction rather than by luck. The HTML tag stream is identical to the source, and every numeric value is unchanged. A row where either differed is never delivered as translated — it comes back in the original language with the reason in its own column.

That is why the remaining 4% matter more than the headline figure. They are the rows the system refused to guess about — mostly health claims with no matching wording in the register in that language, which is a fact about the register and not about the translation.

If you are about to open a new EU market with a supplement catalogue, the question to ask your translation supplier is narrow and easy to check: where does the target-language wording of an authorised health claim come from? If the answer is anything other than “the register”, you are buying a new set of violations in a language you cannot proofread.