When it comes to international keyword research, the standard process is to send the English keyword list to a translation vendor, get it back, and hand it to whoever is writing the German pages. What comes back looks like research and doesn’t tell you much about how Germans search. What a person types is a behaviour rather than a word, and behaviours don’t translate. Somebody has to go and observe them.
The method below is slower than translation, and it’s the difference between entering a market and appearing to. It also costs less than you’d expect, because most of the work is a handful of conversations nobody’s got round to having.
Why translating a keyword list fails in international SEO
Three failure modes, all of them common. The first is vocabulary divergence, where the correct translation isn’t the term anyone uses. Germans say Handy for a mobile phone rather than Mobiltelefon. Spaniards search móvil where Mexicans search celular. Quebec French uses courriel where France uses email. A translator producing correct prose hands you the dictionary term, because that’s the job you paid for.
The second is morphology. German compounds where English separates (Projektmanagement-Software against “project management software”), which changes how modifiers attach and collapses several English long-tail phrases into one German head term. Finnish and Turkish inflect heavily enough that a keyword tool reports a dozen surface forms of one concept. Arabic and Hebrew add transliteration and diacritics on top of that.
The third is intent divergence, and it’s the one that costs money. The same translated phrase can sit at a different point in the buying process in a different market, because the competitive set and the buying culture differ. A term returning comparison pages in the US may return regulatory guidance in Germany. You only find that out by looking at the SERP. It’s the failure nobody catches in review, because the translation itself reads perfectly.
Where in-market seed terms actually come from
Interview three to five people who live and sell in the market: local sales reps, support staff, distributors etc. Ask what customers say on calls rather than what the product’s called internally, and record the verbatim phrasing including the wrong and informal terms, because those get searched too.
Then pull the full organic keyword profile of three local competitors, including the local incumbent nobody at head office has heard of. Read the terms you don’t recognize instead of filtering them out. The unfamiliar ones are the point.
Mine your own data for that country. Search Console groups performance data by the country where the search originated, so it already shows what people in that market typed to reach your English pages. That’s the most best sample you own, with one caveat: Google omits some queries from the report to protect user privacy, so what you export is a floor rather than a census.
Read local retailer sites, marketplaces and forums, where the category taxonomy is already written in the language buyers use. Otto and Idealo in Germany, Bol in the Netherlands, Rakuten in Japan. Merchandisers there have been testing category names against real buyers for years, and they’ve published the results.
Only then run the translated list through a keyword tool, as a coverage cross-check rather than as the source. It’s there to catch what you missed.
Last, take the top 20 terms and look at the live SERP for each, in-country, and classify intent from what ranks. If the results are all local regulatory bodies, that term isn’t commercial and no amount of volume makes it so.
International keyword research on search engines that aren’t Google
Across most of Western Europe, Latin America and the Anglophone markets, Google holds enough share that you can research and optimize for it alone without losing much. A few markets don’t work that way, and treating them as though they do produces a plan aimed at a minority of the country’s searches.
Yandex matters in Russia. Naver and Kakao matter in South Korea, where a large share of commercial discovery happens inside blog and café content rather than the open web. Baidu matters in mainland China and arrives with hosting, ICP licensing and content obligations that make it a legal project before it’s an SEO one. Seznam still holds meaningful share in Czechia, and Yahoo Japan runs on Google’s results while keeping its own ecosystem bolted around it.
In each case the data comes from that engine’s own tools. Yandex publishes query statistics through Wordstat, region by region and over time, and the behaviour it shows differs enough that a Google-derived list makes a poor proxy. Treat any market share figure you read for these engines as approximate and check it against the current year before building a plan on it. The numbers move, and the published sources disagree with each other.
How much to trust international search volume data
Search volume is a modelled estimate everywhere, and the size of the error varies enormously by market. In the US, UK and Germany the major tools run deep clickstream panels and the estimates hold up for relative comparison. In Portugal, Greece, Vietnam or Nigeria the panels are thin, and a reported 90 searches a month might be a rounding artifact rather than a measurement. You shouldn’t be setting a page count off a number like that.
In thin-data markets, rank the terms against each other and ignore the absolute figures. Ordering survives bad data. Magnitude doesn’t.
Three corroborating sources help. Google Trends is a normalized sample rather than a volume count, scaled from 0 to 100 against total searches for the geography and period you picked. That makes it good for direction and growth and no use at all for size (low-volume terms just show as zero). Paid search impression data from a small in-market campaign is the most reliable number you can buy, and two weeks of spend costs less than a year of building against a fiction. And your own Search Console impressions for that country are measurement rather than modelling.
Mapping in-market demand to your domain structure
Research that ends in a spreadsheet of terms is half-finished. The output is a URL plan: which page serves which cluster in which market, on the structure you committed to.
That’s where the market work meets the structure decision. Google’s own guidance treats country-code domains, subdomains and subfolders as different geotargeting signals with different costs, and it names one constraint people forget: a ccTLD can only ever target a single country. So if demand in Mexico and Spain clusters around genuinely different terms with different intent, those are two pages rather than one Spanish page with hreflang variants. If the gap is only vocabulary and price expectation, one localized template with market-specific copy will do. Deciding that per cluster rather than per language is the whole exercise.
Expect the map to come out asymmetric. A market may need eight pages where the US site has thirty, because the demand isn’t there, and forcing structural parity across markets produces thin pages in the small ones. Let demand set the page count. The same logic governs how I build a content strategy for any single market.
How to start international keyword research
Book three 30-minute calls with people who sell in the target market and ask them what customers call the thing. That costs you an afternoon, it’s the highest-yield step in the whole method, and on every engagement where I’ve run it the list came back with at least one head term nobody at head office knew existed.
Then filter Search Console by that country and read the queries you’re already getting. Between those two you’ll have a better seed set than any translation vendor will hand you. Budget for the SERP review too. It’s the step that gets cut when a launch date moves, and it’s the one that catches intent divergence before you’ve paid a vendor to render thirty pages nobody there will search for.