Localisation QA against what each market actually searches

The expensive failure in localisation is the accurate translation. A German page built around Terminplanung reads perfectly to a native speaker and never meets the buyer who types Terminbuchung; a French page selling stockage cloud misses the audience searching stockage en ligne. Our working threshold is simple: any page whose main term draws under 20 percent of the in-market volume of the leading alternative gets flagged, whatever the linguist thought of the copy. Machine output plus post-editing has largely solved grammar and register. It has not solved which word the market pays with. The check, in the order we run it: Head term first. Every page's H1 and title term is put against Google Keyword Planner set to that country and that language, against the country database in Ahrefs, and then against the live SERP, which overrules both. If the top ten results for your target term are comparison articles and forum threads, a product page will not rank there no matter how the term tests, and the recommendation becomes a different page type rather than a different word. Second, intent shape. A single source page frequently answers two separate intents once it crosses a border, and the correct action is to split it, not to polish it. The reverse happens too: two English pages that map onto one Spanish query and should be merged before either is translated again. Third, the market layer that translation never touches. Register and address form, which is a commercial decision and not a style one: Sie in German B2B, tú in Spanish DTC, vous in French almost regardless of sector. Prices displayed gross with a shipping statement for German consumers, per the Preisangabenverordnung. The trust markers a buyer looks for and does not find, such as an Impressum in Germany, Bizum at Spanish checkout, iDEAL in the Netherlands. Date order, units, phone and postcode formats, and German compound nouns that push a translated title past the pixel width Google will render. When the answer is that you should stop: Sometimes the finding is that the demand is not there. The category is bought through resellers in that country, or through a marketplace, and organic search for it barely exists. We report that with the volume figures behind it and recommend you stop translating into that market. Two clients have cut a language from the roadmap on this basis. It is not the outcome anyone commissions, and it is usually the most valuable page in the document. What comes back to you: A per-page verdict sheet with one of five decisions against each URL: keep, replace the head term, split, merge, or do not publish. Every verdict carries its evidence in the adjacent cell, so your country manager can argue with the data rather than with us. Rewritten titles, H1s and meta descriptions for every page we flagged, written in the target language by the reviewer who flagged it. A per-market term glossary in a format your TMS accepts, so the same error stops being reintroduced by the next batch. Where the finding is structural, a short note to whoever owns the template. Where we stop: We do not translate body copy at scale and we do not run your TMS. We do not give legal sign-off. Regulated claims in health, finance and food are flagged with the rule that concerns them and sent to your counsel, who decides. Technical hreflang and domain architecture belong to the other service on this profile and are quoted separately; if both are wrong, fix the architecture first, because intent-correct pages in a broken cluster still will not be served. Do not hire us if: You want a proofreading pass, or a score to use against your current translation vendor in a commercial dispute. You need every market covered next week. Or your localisation decisions ultimately rest with a local country manager whose preference cannot be overruled by evidence, in which case the report gets read, admired and shelved, and we would rather not be paid for that. Working languages and cadence: We review English, Spanish and French ourselves. For German, Italian, Dutch, Portuguese and Polish we bring a native reviewer we have worked with before and name them in the deliverable. Ongoing publishing programmes are better served by the monthly arrangement than by repeated snapshots, because the glossary only earns its cost when it reaches the writers before the next batch, not after it.