International Keyword Research: Why Translating Your Keyword List Doesn't Work

International keyword research is the process of discovering how people in a specific market actually search for a product or service, in their own language, and then building the target pages around those real queries. It is not the translation of an existing keyword list. The two activities produce different words, different page structures and, eventually, very different traffic.

That distinction sounds obvious written down. It rarely survives contact with a launch deadline. A marketing team exports the English keyword list, sends it to a translator with the rest of the site, receives it back tidy and grammatical, and publishes. Six months later the German or Italian pages sit on page four for terms nobody types, while a competitor with half the content ranks above them.

Why translating a keyword list fails

A keyword is a piece of behaviour, not a piece of vocabulary. It records the exact string a person typed while wanting something. Translation reproduces meaning; it has no way of knowing which of several correct renderings is the one that gets typed.

Take a mundane example. An English SaaS company sells "employee onboarding software". Translate that into Italian faithfully and you get something close to "software di inserimento dei dipendenti" — accurate, professional, and almost entirely absent from Italian search behaviour, where HR buyers are far more likely to search around "onboarding" (the English word, which Italian business vocabulary absorbed intact) or around "gestione risorse umane". A translator working without search data has no reason to prefer one. A localizer working with search data has no reason to hesitate.

There are four recurring reasons the translated list underperforms.

The first is borrowed vocabulary. Every language borrows unevenly. Italian and French professional registers absorb English technology terms at different rates and in different places, so "cloud", "software", "marketing automation" may be the real search term in one market and a foreign intrusion in the next.

The second is register mismatch. Search queries are typed, not spoken, and they skew informal and abbreviated. Translated marketing copy tends to be more formal than the query it should match. A French page that consistently says "solution logicielle de gestion" will never quite meet a user searching "logiciel de gestion".

The third is different mental models. Markets categorise the same product differently. What is sold as "field service management" in one country may be searched as maintenance software, technician planning, or intervention scheduling in another, because the job title that buys it is different.

The fourth is volume asymmetry. A term that carries the whole English strategy may have negligible volume elsewhere, while a related term you never targeted carries the market. Direct translation preserves the structure of the English demand curve, which is the one thing that does not transfer.

What international keyword research actually involves

The workable version of this process is smaller and less glamorous than most agency decks suggest. It has roughly five stages, and it needs a native speaker with commercial judgement at every one of them.

You start from intent rather than from words. Before opening any tool, write down what the buyer in that market is trying to accomplish, who they are internally, and what they would type at each stage: a problem query, a category query, a comparison query, a supplier query. This is the part that translation cannot do, and the part that decides whether the eventual page is worth building.

Then you seed. A native speaker generates candidate phrasings for each intent — not translations, alternatives. Two or three people who actually work in that market will produce a better seed list in twenty minutes than any tool will, because they know which words their colleagues use in email.

Then you expand and validate with data. Volume and difficulty figures from keyword tools are estimates and should be treated as such, but they are excellent at ranking candidates relative to one another within the same market. Comparing an Italian volume against a German volume tells you mostly about market size; comparing two Italian terms against each other tells you which one to target.

Then you read the results page. For each shortlisted term, look at what is actually ranking. If the top ten are all informational guides, a service page will not break in, regardless of how commercial the term feels. If the top ten are marketplaces and directories, the honest answer may be that the term is not winnable and the budget belongs elsewhere. SERP inspection is the cheapest quality control available and the most frequently skipped.

Finally you map terms to pages. One primary intent per URL. Terms that share intent belong on the same page, even when they look different; terms that look similar but resolve to different intents need different pages. This mapping is where cannibalisation is prevented, and doing it before writing is much cheaper than fixing it afterwards.

Translation, localization and research are three different purchases

It helps to be explicit about what you are buying, because these are frequently bundled under the single word "translation" and then priced as if they were the same work.

Most sites need all three at once, applied selectively. The pattern that works is straightforward: translate what must stay exact, localize what has to compete in search, and research the market before deciding which pages exist at all. We have written about how that split plays out page by page in website translation vs website localization.

A worked example

Consider a hypothetical French manufacturer of industrial water treatment equipment expanding into Italy. Their strongest French page targets "traitement des eaux industrielles" and converts well.

The literal Italian equivalent, "trattamento delle acque industriali", exists and is used. But an Italian seed session with a technical native speaker also surfaces "depurazione acque industriali", which is the term Italian plant engineers and municipalities use in tenders and specifications. Both are correct Italian. Only one matches the vocabulary of the people who sign purchase orders.

The research step would also reveal something the French site cannot: that a meaningful share of Italian demand arrives through queries about regulatory compliance and discharge limits rather than through equipment queries. That is not a translation finding. It is a content strategy finding, and it argues for a page the French site does not have.

This example is illustrative rather than a client project, but the shape of it is the single most common finding in market research of this kind: the new market needs a page the home market never needed.

Common mistakes

A handful of errors account for most disappointing multilingual launches.

• 01Running keyword research only in the source language and translating the output, which produces pages optimised for a market you are not entering - Using one keyword list across every Spanish-speaking or French-speaking market, when Spain and Mexico, or France and Quebec, differ in both vocabulary and competition - Optimising a translated page after publication instead of researching before writing, which usually means rewriting the page twice - Ignoring the SERP, and targeting commercially attractive terms that are structurally unwinnable - Letting the same intent spread across three near-identical translated pages that then compete with each other - Treating hreflang as the SEO deliverable — it prevents the wrong page from being served, it does not create demand

Where the technical layer fits

Technical hygiene matters, but it is not a substitute for demand research. Self-referencing canonicals, correct and reciprocal hreflang annotations, a consistent URL strategy per language, and locale-appropriate metadata all help search engines serve the right version of a page. Google's own documentation on managing multi-regional and multilingual sites is the reference worth reading, and it is short.

None of that changes which words the page targets. A perfectly annotated site built on a translated keyword list is a perfectly annotated site ranking for the wrong queries. Get the demand map right first; the technical layer then does its job.

How AI search changes the calculation

Generative answer engines — AI Overviews, ChatGPT, Perplexity, Gemini — add an argument in favour of research-led localization rather than weakening it. These systems synthesise answers from content they can parse and trust, and they operate in the user's language. A page that uses the market's real terminology, answers a question directly and defines its terms cleanly is easier to cite than a translated page whose vocabulary sits slightly to one side of how the topic is discussed locally.

Practically, that means writing clear definitions near the top of a section, keeping comparison content in genuine comparison form, and using the terminology that market's specialists use, including the borrowed English words when those are what people actually say. It is the same discipline that helps human readers, applied a little more deliberately.

Who should do this work

The combination you need is a native speaker of the target market who understands the sector, working with search data and with someone who can decide what the business is willing to compete for. Translators are excellent at accuracy and often have deep sector vocabulary; SEO specialists are excellent at demand data and rarely have the language depth. The research is only reliable when both perspectives are in the room, which is why it usually sits with a localization partner rather than being split between a translation vendor and a domestic SEO agency who never speak to each other.

FAQ

Is international keyword research the same as SEO translation?

No. SEO translation adapts an existing page to be search-friendly in another language. Keyword research happens earlier and decides which pages should exist and what each one should target. Doing the second without the first means optimising pages that may be aimed at the wrong queries.

Can I use machine translation for the keyword list?

Machine translation is useful for generating candidate phrasings quickly, but its output is a starting point, not a list. It cannot tell you which of five plausible renderings people actually type, and it reliably misses borrowed terms and market-specific jargon. Validate every candidate with search data and a native speaker.

How many keywords do I need per market?

Fewer than most lists contain. A focused set covering the main commercial intents plus a handful of informational queries that support them is enough to build a first release. Long lists tend to produce thin pages that split the same intent.

Should each market get its own content, or a translated version of the same site?

It depends on the page. Legal, technical and product-reference content should be consistent across markets. Commercial and editorial content usually performs better when the market's demand shapes it, which sometimes means pages that exist in one language and not another.

Do keyword volumes from tools reflect reality?

They are modelled estimates and should be read as relative signals rather than forecasts. Use them to rank candidates within one market and one language. Use the results page, not the volume figure, to judge whether a term is realistically winnable.

When should the research be done?

Before translation is commissioned. Research after publication is possible but means paying for the same pages twice, and it delays the point at which the market pages start earning traffic.

Planning to enter a new market and unsure whether your existing keyword strategy will transfer? Verbavox runs multilingual SEO and keyword localization with native specialists in each target market, and can align the research with your website localization work so the pages are built once, around the right queries. Tell us the markets and we will tell you what the demand actually looks like.