How I Actually Do Keyword Research for International SEO

“International SEO” may be the most misunderstood phrase in marketing. Translate the homepage, bolt on hreflang, wait for traffic. That is the playbook most teams run. It almost never works.

I have watched companies spend six figures on multi-language launches that produced single-digit weekly clicks for an entire year. The pattern is identical every time: they optimised for keywords that look correct in a translator and have zero pull in the market they actually care about.

The good news is that the real work is not complicated—just specific. Below is the workflow I run when expanding a site into a new country, in the order I run it, with the calls I would actually make. If you came here looking for a polite, balanced overview, this is not that. I have opinions, and the opinions are what saves you from burning a year.

1. Stop thinking in languages. Think in country × language pairs.

If you remember nothing else from this article, remember this: in SEO, “language” is a useless unit of work.

The most expensive mistake I see is “we will do German content for the DACH region.” DACH is three countries with three economies, three retail landscapes, and three sets of default search results. Roboter-Rasenmäher trends differently in Germany than in Austria. The top ten domains for Auto kaufen in Switzerland barely overlap with the ones in Germany. Treat those as a single market and your budget evaporates.

The same trap exists everywhere:

• Mexican Spanish uses computadora; Castilian Spanish uses ordenador. Same device, two completely different keywords.
• Brazilian Portuguese says celular; European Portuguese says telemóvel. One word picks the wrong country.
• British shoppers search trainers, Americans search sneakers, Australians alternate between both depending on the brand.
• Quebec French diverges from Parisian French in tech vocabulary so severely that some queries return entirely different SERPs—even though the headline language is “French.”

Your real unit of work is country × language. Switzerland-de is a separate plan, separate volume data, and separate priority list from Germany-de—even if you reuse 70% of the underlying copy. Lump them together and Google's local intent layer will quietly punish you.

2. Mine seed keywords properly. Most lists I see are garbage.

Most seed keyword lists I review are a copy of the company's product feature page. That is not where keywords come from. That is where keywords end—after a marketing team has already sanitised the language.

A useful seed list pulls from the places where real customers describe their own problem, before they have learned the company's vocabulary:

• Your own site search log. What people type into the search box when they cannot find something. This is gold, and roughly 90% of teams never open it.
• Support tickets and chat transcripts. Read the first 50 chat openings. Customers describe a problem in everyday language—not the categorised product name the marketing team uses.
• Reddit and niche forums. Search your category and read the titles. Forums are full of natural-language phrasing that no translation engine will ever produce.
• YouTube comments under competitor explainer videos. People ask questions using the exact words they would type into Google.
• Competitor H1s and breadcrumbs—not just page titles. Breadcrumbs reveal the competitor's category taxonomy, which is roughly a map of how the market thinks about the space.

Aim for 20–40 seeds in your source language. Anything narrower will give you a narrow keyword list and you will miss the long tail that makes international SEO worth doing in the first place. Anything wider and you will dilute attention.

One contrarian position: do not start by translating your homepage. Start with a 20-minute call with one real customer in the target country. That single call gives you better phrasing than a week of competitor scraping. Most teams skip the call because it feels slower than running a translation script. They are wrong. The script feels productive and produces nothing.

3. Translation is a draft. Treat it as a hypothesis.

The “robot lawn mower / mähroboter” example is now cited everywhere and I am tired of seeing it. So here are three less famous ones I keep encountering:

• “Cheap flights” → French. Direct translation: vols pas chers. Real high-volume term: vols low cost. Yes, English. The English loanword owns the French SERP.
• “Notebook” in Germany. Germans say Notebook far more often than Laptop for the exact same device. If your German page title says “Laptop kaufen,” you are handing traffic to whoever wrote “Notebook kaufen.”
• “Mobile phone” in Brazil → celular, not telefone móvel. In Portugal the same product is telemóvel. One word can multiply traffic 5–10x in one of the two markets and tank it in the other.

The pattern is always the same. Machine translation gives you a grammatically correct word that local users do not actually type. Real users use loanwords, slang, brand-as-category, or shortened forms. None of that comes out of a translation engine.

Three checks I run on every translated seed before it earns a place in the priority list:

• Native sanity read. One fluent local reads the list aloud. Anything they pause on—“we would never say that”—becomes a kill candidate. Five minutes of work; saves months of writing.
• In-country autocomplete. Open Google in incognito with the target country region set. Type each seed and watch the dropdown. Autocomplete is Google telling you, free of charge, what people most commonly type next in that market.
• First-page commercial intent check. If the first page is dominated by Wikipedia, government sites, or pure encyclopaedic content, you picked the wrong word—even if it is technically correct. There is no buyer there.

4. Validate with data—and read it sceptically

Once your translated list survives the native pass, push it into Ahrefs, SEMrush, or your data source of choice. Three numbers per market are non-negotiable:

• Country-specific monthly search volume. Not global. The global aggregate hides the cliff between markets.
• Keyword Difficulty (KD). Useful as a directional signal. Useless as a hard rule, especially for sites under DR 40.
• CPC. A surprisingly underused intent indicator.

A few things I learned the painful way:

• Stop trusting KD alone. A KD of 35 looks doable on paper—but if the top 10 results are all DR 80+ retailers with a decade of backlinks, you are not realistically beating them in 12 months. Always look at competitor DR distribution before committing budget.
• Watch the “EU” or “Worldwide” volume trap. Some tools default to aggregate volume, which lumps in markets where you do not operate. A keyword that looks like 50,000 monthly searches collapses to 4,000 when you filter to Germany only.
• Treat CPC as a signal, not a target. A €0.20 CPC on your main term means advertisers are not bidding it up. Either the term is too informational or the conversion economics are bad. You want to know that before you write the brief.

What I cut: anything with country-level volume of zero, KD over 60 if my site is below DR 30, or top-10 SERPs that are 100% YouTube, Wikipedia, or Reddit (commercial intent absent).

What I promote: the surprise winners. Keywords the translator produced that I would have ignored on instinct, but the data shows 8,000 monthly searches with a KD of 18 and SERPs dominated by mid-sized affiliates. Those are the unglamorous compound-traffic plays that pay rent six months from now.

5. Intent is local. Read the SERP before you write a word.

Intent labels (informational, commercial, transactional) are SEO 101. The problem is most teams apply the labels they learned from the English market, then copy-paste them onto translated keywords—where the actual SERP may look completely different.

Real example: “best CRM.”

• US SERP: dominated by listicles from G2, Capterra, Forbes Advisor. Pure comparison content. You write a comparison article.
• DE SERP: the listicles get pushed down. The first page often shows direct vendor pages (HubSpot, Pipedrive, Salesforce DE). Intent is more product-direct, less editorial.
• JP SERP: a frequent mix of company-run owned-media articles and review aggregator sites that simply do not exist in Western markets.

Same query, three completely different winning formats. If you wrote a US-style listicle for the German SERP, you would lose to the vendor landing page every single time, no matter how good the listicle is.

How I read a SERP fast, in this exact order:

• Top three ads. Tells you whether advertisers value the term commercially.
• Top three organic. What is the dominant format—product page, listicle, blog post?
• SERP features. People Also Ask, video carousel, Shopping, image pack. They tell you what Google thinks users want and how much space is left for you.
• Domain mix. Big retailers, small affiliates, news sites, or vendor pages? The mix decides whether you can realistically squeeze in.

Decide page format from the SERP. Not from the content calendar your CMO printed out in January.

6. Cluster aggressively. Prioritise honestly.

Three hundred keywords for one market is not a content plan. It is anxiety in spreadsheet form.

Cluster keywords by topic and intent until you have 20–40 page concepts. Each page owns one primary keyword (the most-searched, intent-matched term) and 3–7 supporting long-tail terms. The supporting terms live on the same page—as H2s, FAQ blocks, body paragraphs. They do not become separate pages.

A note on cannibalisation: publish two pages targeting near-identical clusters and you split your own ranking strength. One primary keyword per cluster, full stop.

Prioritisation is where most teams cheat. They sort by volume descending and attack from the top. Wrong. The correct order is:

• First, filter to keywords where your DR can realistically compete with the top 10—or where you can outflank with content depth no one else has produced.
• Then sort the survivors by volume × commercial intent.
• Finally, pick clusters where you already have adjacent topical authority. New territory takes longer than people promise.

7. Brief, ship, audit. Nothing else counts.

Keyword research only matters if it ships. The deliverable is a brief another human can act on, not an Excel file in a shared drive. Each brief should answer:

• Target country, language, and audience persona.
• Primary keyword + 3–7 supporting long-tail terms.
• Search intent and the page format that fits the local SERP.
• Suggested H1 and H2 outline. Do not make the writer invent a structure from scratch.
• Internal links from existing pages with topical relevance.
• Success metric: ranking position, organic clicks, or both—and the threshold that means “this worked.”

Hand the brief to a writer who lives in the target country, or at minimum to a writer who works with a native reviewer in a tight loop. Machine-translating one English brief into five languages and assuming all five will rank is exactly how thin-content empires get built.

Audit cadence: 60 days after shipping the first batch, then quarterly after that. International SERPs reshuffle faster than English-market SERPs because competitive density is lower—your competitors are also still figuring it out. If a page is on page two at the 60-day mark, the right call is usually “improve,” not “abandon.” Most teams abandon too early because they have no audit ritual at all.

8. The mistakes I keep watching teams make

• Treating one language as one market. German content for Germany is not German content for Switzerland.
• Optimising for volume without ever opening the local SERP.
• Reusing meta titles across countries because “they are all in German.” A Berliner and a Viennese viewer click on different titles.
• Hreflang implemented wrong—every language variant pointing to a single canonical, which collapses the multi-market signal back into one.
• Failing to localise images, prices, dates, and units. A page that ranks but feels foreign to local visitors does not convert.
• Serving B2B content to a B2C audience (or vice versa) because the source page was written for a different audience and nobody re-thought it.
• No post-launch audit. The site quietly underperforms for nine months and eventually someone asks, “why is this not working?”
• Letting the marketing team approve translations. Marketing knows the product. It does not know how a 32-year-old Tokyo office worker actually searches.

9. The honest tool stack

You do not need a stack. You need three things that work.

• A volume + difficulty source. I use Ahrefs. SEMrush is a real alternative. Ubersuggest is not.
• A SERP-checking habit. Google's own search, region set correctly, in incognito. That is enough. Tooling around this is overrated.
• A native reviewer per market. The single highest-ROI hire on an international SEO project. The next-cheapest version is a freelancer reviewed on Upwork or Malt for less than the cost of one Ahrefs seat.

A multilingual keyword explorer—like the one on this site—compresses steps three through five (translation, country-level volume, intent labelling) into a single search. Use it for the first-pass filter so you are not pinballing between a translator and a keyword tool. Then verify the survivors in your main data source.

Your 7-step checklist

• Define country × language pairs, not language plans.
• Mine 20–40 seed keywords from search logs, support tickets, forums, and one real customer call.
• Translate as a draft, then run native, autocomplete, and SERP checks before anything moves forward.
• Pull country-specific volume, KD, and CPC. Cut the dead. Promote the surprises.
• Read the target-country SERP and let it decide your page format.
• Cluster into 20–40 page concepts. One primary keyword per page. Prioritise where you can actually win.
• Brief, ship, and audit at 60 days. Then quarterly. Forever.

International SEO is not hard. It is just specific. Most teams lose because they want the work to be language-agnostic, when the work is country-specific. The teams that win pick fewer markets, do them properly, and re-audit faster than the competition.

If you want to compress the seed → translation → validation loop into one tool, the keyword explorer on this site is built for exactly that workflow. Drop in a seed, pick a country, and read what people actually type. Then come back here and brief the page.