SEO translation: how to translate pages so they still rank

Why literal translation loses rankings

Search demand is not evenly distributed across languages. A term with 40,000 monthly searches in English may have three plausible equivalents in German, one of which takes 80 percent of the volume while the other two are near zero. A translator with no search data has no way to know which one to pick, so they pick the one that reads best.

The result is a page that competes for a phrase nobody searches. It is not a translation error; it is a market research error that happens to be expressed in language.

• Compound languages such as German and Dutch split demand across compound and separated forms.
• English loanwords often outrank the official native term, especially in tech and marketing.
• Regional variants matter: the Spanish that ranks in Mexico is not the Spanish that ranks in Spain.
• Question phrasing differs: some markets search in full questions, others in two-word fragments.

The eight-step SEO translation workflow

This is the sequence we built Smart GSEO Translator around, and it works whether you run it by hand or automate it.

• 1. Inventory the source page: primary keyword, secondary keywords, the questions it answers, the conversion goal.
• 2. Generate candidate equivalents in the target language, several per intent, including loanwords and colloquial forms.
• 3. Measure each candidate against real search volume, cost per click and competition in that country.
• 4. Pick the winner per intent and record why, so it can be reviewed later.
• 5. Rewrite the page around the winning terms, with the heading structure rebuilt for how that market phrases the question.
• 6. Localise the proof: currencies, units, dates, examples, regulations, names of local competitors.
• 7. Ship the technical layer: hreflang cluster, self-referencing canonical, localised metadata, localised internal links.
• 8. Review after 60 days against impressions, not just rankings, and refine the page where the query mix surprises you.

Keyword equivalents: the part everyone skips

The hard part of step 2 and 3 is that you need both linguistic judgement and volume data at the same time. Keyword tools give you volume for terms you already know. Translators give you terms without volume. The gap between them is where most international SEO programmes leak traffic.

The fix is a two-pass process. First a language pass proposes how a native speaker in that market would phrase each intent, producing several candidates rather than one. Then a data pass measures all of them in the target country and language, so the choice is made on evidence.

Structure your copy for both Google and AI answers

Google still rewards a clear match between query, heading and body. Answer engines add a second requirement: passages must stand alone. When ChatGPT, Perplexity, Gemini or Google's AI Overviews build an answer, they retrieve a passage, not a page, so a paragraph that depends on the sentence before it is far less useful to them.

Writing for both is not a compromise. Answer-first paragraphs, question-shaped headings and explicitly named entities help human readers too, which is why they also lift classic rankings.

• Lead each section with the answer, then explain it.
• Keep sentences under roughly 25 words and one idea per paragraph.
• Name the product, brand, standard or place instead of writing 'it' or 'this'.
• Convert facts, never invent them: figures, dates, units and currencies must be correct for the market.

The technical layer, in one pass

Content work fails quietly when the technical layer lags. Every language version needs a self-referencing canonical, a complete reciprocal hreflang cluster including x-default, a localised title and meta description, and internal links that point at the same-language version of the target page.

One rule prevents most of the damage: never link from a German page to an English page when a German version exists. Cross-language internal links dilute the cluster and confuse crawlers about which version serves which audience.

How to measure whether it worked

Rankings for your chosen keyword are a lagging and narrow signal. Impressions per country in Search Console tell you far more, far sooner: if the page is picking up impressions on a phrase you did not target, your keyword choice was wrong and the market has just told you the right answer.

For AI visibility, track mentions instead of positions. Ask the assistants the questions your page answers, in the target language, and record whether you are cited. Repeat monthly; the trend matters more than any single answer.

Frequently asked questions

Keep reading