International keyword research: a worked example

International keyword research is finding the words people in each market type into a search engine, in their own language, for the thing your page offers. It is a separate piece of research for every language and country you target. Translating your English keyword list gives you a starting set of candidates, and the data then often picks a different phrase. One example from the figures below: in Spain, the textbook translation of "translate website", traducir sitio web, had 20 searches a month in DataForSEO's data, while traducir página web had 1,300.

Why a translated keyword list falls short#

A translator chooses the most correct rendering of a phrase. A searcher types the shortest thing that gets results, often without articles or accents, sometimes in English. Those two habits pull apart, and the gap differs by market.

Three things cause most of it. Languages have several everyday words for one English term, and they are not used equally. Searchers borrow English words, so the "translated" term can lose to the loanword. And grammar that is required in a sentence is dropped in a query, so a well-formed phrase can show almost no demand while its clipped form has plenty.

You cannot see any of this from a dictionary. You see it by pulling volumes for every plausible variant in the target country and language, then reading the results page for the ones that look worth a page.

A worked example: website translation in four markets (DataForSEO, 2026-10-07)#

We ran this research for our own topic. For each market we listed the phrases a native speaker might use to look for a way to translate a website, then pulled DataForSEO's Labs keyword overview for that country and language on 2026-10-07. The volumes are DataForSEO's average monthly Google searches, which its documentation says are based on the Google Ads API. Rows marked "grouped" came back with exactly the same numbers as the row above them, because the tool treats them as variants of one keyword.

Two of the Spanish candidates (web multidioma and página web multilingüe) and one Dutch one (vertaalplugin wordpress) returned no data at all, which usually means too few searches to report. The four calls cost about five US cents together.

What the German rows show#

German has at least four words for a website in daily use: Website, Webseite, Internetseite and Homepage. The first two are close, and the other two are smaller but far from zero. A page that says only Website everywhere is fine for the head term and misses the rest, so a German page can use Webseite in body copy where it reads naturally.

The verb matters too. Website übersetzen lassen means "have a website translated", and someone typing it is looking for a service to hire. It is a different page from a software page, even though both are "website translation" in English.

What the Spanish rows show#

This market makes the case for research on its own. Sitio web is a correct translation of "website", and almost nobody in Spain searches with it. Searchers in Spain write página web or just web. A Spanish landing page whose title is built on sitio web is correct and close to invisible for this query.

What the French rows show#

The grammatical phrase traduire un site web had 40 searches, and the clipped traduire site web had 480. French searchers also use site internet alongside site web. The page should read like proper French, which means the article stays in the copy, while the title can carry the clipped phrase if it still reads well to a person.

What the Dutch rows show#

The Dutch head term contains the English word "website", and DataForSEO's language detection flagged the head term, the row grouped with it and meertalige website as English. That flag is a property of the tool and says nothing about the searchers, who are in the Netherlands searching in Dutch. Expect this in any market where English loanwords are common, and do not throw a row away because of the flag.

How to read the numbers a keyword tool gives you#

Every keyword tool reports estimates, and a few habits keep you from misreading them.

Do not add grouped rows together. When two phrases return identical figures, the tool is reporting one keyword twice. Website übersetzen and website übersetzung are 1,900 searches between them, and 3,800 would be wrong. Google's Keyword Planner help describes the same behaviour from the source: average monthly searches count a keyword "and its close variants", and the figures are rounded, so volumes for several locations "might not add up as you'd expect".

Treat small numbers as "some". The smallest figure DataForSEO returned in this research was 10, for wordpress übersetzen, and a figure that low stands for a handful of searches. Rank candidates against each other, and avoid quoting them to a client as exact demand.

Check the country and the language together. A German keyword pulled for Austria or Switzerland is a separate measurement from the same keyword in Germany, and a Spanish keyword in Mexico is a separate market from Spain. The tool needs both settings, and so does your notes file.

Read the trend before you commit. Nearly every row in our table had a lower yearly figure than the year before. A falling query can still be worth a page, but you should know it is falling.

A method you can repeat for each market#

1. Start from the page and the job it does, such as a product page, a guide or a pricing page. List what a buyer in that market would call that thing.
2. Write candidates with someone who speaks the language natively. Include synonyms, loanwords, the form with and without articles and accents, and the verb forms that signal hiring someone.
3. Pull volumes for every candidate with the right country and language set, and keep the raw export with the date. Google Trends is a useful cross-check between regions, as long as you remember its help page describes the numbers as normalized and scaled from 0 to 100, which makes them relative interest and not a count of searches.
4. Search the top two or three candidates from that country and look at what ranks. If the results are agencies, the query wants a service. If they are tools, it wants software. If they are dictionaries, nobody with your intent is typing it.
5. Pick one primary phrase per page per language, plus two or three supporting ones that read naturally in the copy.
6. Put the primary phrase where search engines and readers look first: the title, the main heading, the meta description and the URL slug. Our article on SEO localization goes through each of those fields.
7. Record the choice so every translated page uses the same word. A glossary entry does this; see how to build a translation glossary.
8. Measure by country once the pages are live. Search Console's Performance report has a country dimension that groups data "by the country where the search originated", so you can read the queries each market used to find each page. Revisit the list when those differ from the ones you picked.

Same language, different countries#

Spanish, French, German, Portuguese and English are each spoken in several countries, and the queries differ between them. Research each country you sell to as its own market. If two countries return nearly the same phrases, one page per language can serve both. If they diverge, that is an argument for regional versions with their own URLs and hreflang values, which our international SEO guide covers. Our language guides note the regional variants for Spanish and German.

Putting the research to work in LangRelay#

LangRelay translates the pages your WordPress site renders, so the research ends up in three places.

The glossary carries the terms with one right answer. On paid plans you can give a term its translation per language, and LangRelay's machine translation follows the glossary, so a product term your research settled on stays the same across pages. A term can also be kept as written, which is how you stop a brand or product name from being translated. Where you want variety, such as Website in the title and Webseite in body copy, edit the line itself rather than adding a glossary entry.

Titles and meta descriptions are translated with the page, and you can review and edit them on the Visual editor's SEO tab; to do that before pages go live, choose Ask me first. That is where a primary phrase gets into the title in its clipped search form when the machine output chose the textbook one.

The address of each page is set by hand per page and language in LangRelay today, so the slug carries the term you picked. Automatic slug translation is on our roadmap. By default a translation goes live as soon as it is ready, and free checks hold a page with untranslated text, the wrong language or a changed price. To approve every page yourself, and check that the chosen phrase reads naturally in the sentence it ended up in, choose Ask me first. If you would like to see this on your own site, book a demo.

Questions people ask about international keyword research#

Do I need separate keyword research for each language?#

Yes, and for each country where the language is spoken differently. The Spanish example above shows why: traducir página web had more than sixty times the searches of traducir sitio web.

Can I use machine translation to find keywords?#

Use it to generate candidates, then let the volumes decide. A correct translation can still be the phrase nobody searches, as traducir sitio web was in Spain.

What if a market shows no volume at all?#

Check spelling variants and the English loanword before concluding there is no demand. If every variant is empty, the page can still serve visitors who arrive from elsewhere on your site, and you write it for them rather than for search.