Translate the customer’s decision, not just the words of the prompt.
Choose markets deliberately
Start with the regions where you serve customers and have a clear commercial objective. Record language, market, product availability, currency, and important local alternatives. One global average can conceal a visibility gap or an answer that recommends an unavailable product.
Use customer evidence from the relevant market. The terms buyers use, evaluation criteria, and trusted publications may differ. Do not assume your home-market question set is representative after a literal translation.
Review meaning with a native speaker
Ask someone familiar with the category to review each translated question. Preserve the underlying buying task and note where local terminology changes the meaning. Keep matched question IDs across languages while retaining the exact wording used in each test.
Maintain a second set of genuinely local questions. A market-specific regulatory or compatibility concern may have no useful equivalent elsewhere. Report this set separately rather than forcing an artificial cross-market comparison.
Validate names and facts
Document local brand names, product names, and approved aliases for mention detection. Review ambiguous terms manually. A word that uniquely identifies your product in one language may be a common noun in another.
Check whether local pages explain supported markets, commercial terms, delivery, and service availability accurately. Record contradictions between translated pages and the authoritative product information. Route the fix to the owner of the fact, not only the translator.
Compare within clear boundaries
Report performance by question group, market, language, and surface. State whether the test controls location or merely asks about a region in the text. Those approaches are different, and the report should not imply stronger control than your method provides.
Use repeated observations and example answers to investigate differences. A stronger score in one language may reflect question composition or naming rules. Inspect the evidence before attributing it to localization success or a platform preference.
