AI can prepare two versions of a product description, but the store still needs to verify what each version promises. Hostinger’s recent account of AI mistakes is a useful prompt to build a review that checks product facts and French text before publication.
1. Establish the product record
Start with one item and collect approved specifications: product code, materials, dimensions, package contents, variants and care instructions. Record where each value came from. An old marketing page is not necessarily authoritative when the item has changed since it was written.
Give the assistant this limited record and request independent French and English drafts. State that missing information must remain missing. Delivery promises, certifications and compatibility claims should not appear merely because they sound typical for the product category.
Keep technical identifiers and measurements exact. If the product has several sizes, make the selected variant explicit. Replace an open-ended instruction to make the copy more persuasive with a concrete audience and a defined set of facts that the assistant may use.
2. Check claims across languages
Review the source record beside both drafts. Match factual claims rather than sentence structures: natural French and natural English need not follow identical wording. They should, however, describe the same item and preserve the same limitations.
Use a compact review table:
| Item | Passing check |
|---|---|
| Measurements | Values and units match the record |
| Variants | No invented or missing option |
| Care | Instructions have a documented source |
| Promises | Each claim has supporting evidence |
Remove language that implies an unverified property. Pay particular attention to restrictions such as indoor use. A polished translation may still change the meaning by omitting that limitation. Ask a competent reader of each language to resolve ambiguity instead of relying on a second generated draft as proof.
3. Test the saved page
Save source files as UTF-8 and select the matching import encoding. Try the import on a draft first. Inspect French words containing accented letters, such as “été” and “façade,” and check “cœur,” apostrophes and spacing around measurements. If stored text is damaged, reimport from the intact source rather than disguising the problem in the display.
Open both language previews on desktop and mobile. Check the title, lists, selected variant and links. Copy a sentence from the rendered preview into a text editor to catch replacement characters or unexpected breaks that are easy to overlook in a busy layout.
Finish with a human review of both facts and wording. Save the approved product record and review date with the drafts. When a specification changes, that record tells you which statements need updating in each language and helps prevent an old AI-generated claim from surviving unnoticed.
Sources: Hostinger