Multilingual keyword research fails for one reason far more often than any other: someone translated the English keyword list and built pages around the output. Translation gives you a correct word. Search gives you the word people actually type, which is frequently not the correct one — and the gap between those two is where six months of localization budget disappears.
Here is how to do it so the pages you build in each language target demand that exists.
Pick the market before the language
Spanish is not a market. Mexico, Spain, Argentina, and US Hispanic audiences are four markets with different vocabulary, different competitors, and different search volumes for the same concept. “Coche” versus “carro” versus “auto” is a real decision with real traffic attached.
Start by choosing the countries you can actually serve — ship to, support in, invoice in — and rank them by opportunity. Then set your research tool’s index to that specific country. This is not a cosmetic setting. Querying a Singapore business against the US index returns a near-empty dataset that makes a healthy market look dead, and I have watched teams abandon a viable region on the strength of that mistake.
Seed from native sources, never from English
Get your starting terms from inside the target market. Competitor sites that already rank there. Local marketplace category names. The autocomplete on the local Google domain. Your own support tickets and sales calls in that language, if you have them.
Then run those native seeds through a multi-seed explorer with the country index set correctly. A single search should return up to 150 ideas with volume, difficulty, CPC, and SERP features — and this is the point where the surprises show up. German compound nouns collapse three English terms into one query. Japanese searchers mix scripts, so the katakana and the Latin-alphabet form of a brand can have wildly different volumes. French searchers drop accents in a way that splits data across variants.
Use English as a hypothesis, not a source
There is one legitimate use for your English list: as a checklist of concepts to look for. Take each English head term, find the native equivalent by looking at what ranks for it in-market, and then research from that native term outward. You are using English to make sure you did not miss a topic, not to generate the keyword.
Machine translation is fine for this bridging step. It is not fine as the final answer. Every translated term needs a volume check in the target index before it earns a page, and a meaningful share will return zero — that is the tool telling you nobody phrases it that way.
Validate with a native speaker in one hour
You do not need a full localization agency to sanity-check a keyword list. You need one fluent person and sixty minutes. Give them your top 40 candidate terms and ask three questions per term:
- Would you type this into Google, or does it read like a textbook?
- If you saw this on a page, would it sound like a local wrote it or like a translation?
- Is there a shorter or slangier way people actually say it?
This single hour catches the errors that cost the most: formal register where searchers use informal, brand-name genericization (“Hoover” problems exist in most languages), and terms that are technically correct but commercially dead. It is the highest-leverage hour in the entire process.
Read the SERP in each market separately
Same keyword concept, different country, completely different competitive picture. In one market page one is dominated by local publishers; in another it is Amazon and two aggregators; in a third the query triggers a shopping carousel that pushes organic below the fold.
Check the SERP for your top ten terms in each target country before committing. Difficulty scores are modeled estimates built from link data and periodic crawls — useful for sorting a list, insufficient for a go/no-go on a market. Look at the actual page-one results and benchmark against the weakest one on it, not the strongest. A local competitor sitting at position nine with eleven referring domains is your real entry cost.
Keep the language and the country signals straight
Research is wasted if the pages cannot be served to the right people. Whatever structure you choose — country subfolders, language subfolders, or separate domains — keep one URL per language-market pair and declare hreflang between them, including a self-referencing tag on each page and an x-default for everyone else.
Then verify it with a crawl. Hreflang errors are the single most common technical failure in multilingual sites, and they are invisible until you look: unreciprocated tags, wrong region codes, tags pointing at redirecting URLs. A full-site crawl that reports the actual URL and the actual tag value is worth more here than any advice, because the failure is always specific.
Track each market on its own
Never average rankings across countries. A blended position number hides a market collapsing while another grows, and by the time the average moves you have lost a quarter. Track each language-market pair as its own project with its own keyword pool and its own trend line.
Expect noise. Positions move two or three places between checks as a matter of course, and international results vary further with the searcher’s location and language settings, so read four-week slopes rather than daily readings. Google Search Console is your ground truth per country and per query — third-party estimates are for competitive comparison, not for reporting on your own site.
A realistic sequence for launching a new language
- Pick one country. Set the research index to it. Resist doing three at once.
- Gather 20 to 40 native seed terms from in-market sources
- Expand, filter to terms with real volume in that index, export the list
- Validate the top 40 with a native speaker in a single session
- Run a content gap against three local competitors to catch what you missed
- Build 10 to 15 pages written natively — not translated — around the survivors
- Set hreflang, crawl to verify, then track that market’s positions separately
Give a new market two full quarters before judging it. Multilingual keyword research done this way is slower to start and far cheaper than the alternative, which is discovering after 200 translated pages that you targeted vocabulary nobody uses. SEO Rocket handles the country-specific indexes, the competitor gap, and the per-market tracking in one place at $50 a month — but the native validation step is human work, and skipping it is the mistake that no tool can catch for you.
