International SEO is the practice of making a single organization discoverable and served correctly across multiple countries and languages. It involves deciding which URL represents which market, signalling those relationships to search engines, and producing content that matches how each market actually searches. It combines a domain architecture decision, a set of annotations, and a content problem that is usually underestimated.
Only the first two are technical. The third is where most international programs fail, because a site can be flawlessly annotated and still be answering questions nobody in that market asked.
The domain structure decision
There are three viable structures, and the choice is close to irreversible once traffic accumulates.
- ccTLDs (
example.de,example.fr): The strongest geographic signal and the strongest local trust signal. Also the most expensive: each domain accumulates authority separately, so you are running five link acquisition programs instead of one. - Subdomains (
de.example.com): Easy to point at separate infrastructure or a regional CMS instance. In practice they share authority less generously than subfolders and they buy you very little that a subfolder does not. - Subfolders (
example.com/de/): One domain, one authority profile, one set of technical standards. Every link earned anywhere helps everywhere.
My position (because it is the question every client asks): default to subfolders on one domain, and require a specific reason to deviate. The consolidated authority advantage is substantial and compounding, and running one technical estate rather than five is worth more to most teams than a marginal local trust signal.
There can however be genuine reasons to deviate. Germany, Japan and Russia are markets where a local domain measurably affects user trust, and in some categories that matters more than authority consolidation. Legally distinct entities, separate pricing and inventory systems, or a market where you may need to exit cleanly all argue for a ccTLD. Regulatory data residency sometimes settles it before SEO gets a vote. If you inherit ccTLDs that are already performing, consolidating them is a large migration with real downside and should be justified on more than tidiness.
hreflang and x-default
hreflang annotations tell search engines that a set of URLs are equivalent pages aimed at different languages or regions, so the right one is shown to the right user rather than the wrong one being filtered out as a duplicate. It is a clustering signal, not a ranking signal. It does not make anything rank higher, it makes the correct version appear.
The rules that break implementations, in the order I find them broken:
- Annotations must be reciprocal. If the UK page points at the Australian page, the Australian page must point back, or the relationship between those two pages is ignored.
- Every page in the cluster must include a self-referential annotation. Omitting it is the single most common error I see.
- Values are language, optionally hyphenated with a region (e.g.,
en-GB,en-AU,fr-CA,de-CH). Language codes are ISO 639-1, region codes ISO 3166-1 Alpha 2.en-UKis not valid; the country code isGB. hreflangmust not conflict withrel=canonical. If the German page canonicalizes to the English one, you have told Google the German page should not exist, and the cluster loses.- Annotated URLs must be indexable and return 200. Pointing at a redirect or a
noindexpage invalidates the entry.
x-default names the page to serve when no other annotation matches the user. It is the fallback for a Brazilian visitor to a site that covers the UK, Germany and France. Point it at a dedicated country-selector page if you have one, or at your primary global version (often the main English or US URL) when you need a sensible default, and do not omit it: without it, unmatched users get whatever the algorithm guesses.
On large multi-market estates, put the annotations in XML sitemaps rather than in <head>. Twenty markets means twenty tags on every page, which bloats every response, and sitemaps are easier to generate correctly from a single source of truth than templates edited by five regional teams.
Search Console geo-targeting deprecated
Google retired the International Targeting report and its country-targeting setting in Search Console in 2022. You can no longer tell Google that a generic domain or a subfolder is aimed at Canada.
What remains is inference: hreflang annotations, ccTLD where you have one, server location, language of the content, currency and address signals, and links from within the market. The practical consequence is that hreflang now carries more weight than it used to, and getting it wrong has fewer safety nets. Domain properties in Search Console are still how you segment performance by country in reporting, which is a different thing and still useful.
Translation is not localization
Translation converts a page from one language to another. Localization produces the page that market would have wanted, which frequently is not the same page.
The mechanical failure is keyword lists translated out of English. Demand does not map one-to-one across languages: the highest-volume term for a category in German is often a compound noun with no elegant English equivalent, and the second-highest term in French may be a phrase your translator would never choose. Research demand natively, per market, before anything is written. A translation memory system optimizes for consistency with what you already said, which is precisely the wrong objective here.
A perfectly annotated site full of translated pages ranks for terms nobody in that market searches.
Then there are the non-linguistic differences that decide conversion: local payment methods, local proof and reviews, VAT-inclusive pricing where that is the norm, date and measurement formats, competitors who are unheard of in your home market and dominant there. None of this is hreflang. All of it determines whether the traffic you finally win is worth anything.
Where to start
Validate the cluster before you write anything. Crawl the site, extract every annotation, and check four things: reciprocity, self-reference, 200 status on every target, and no conflict with canonicals. Broken clusters are common and the fix is cheap relative to its effect.
Then take your two largest non-English markets and run native keyword research against your existing translated pages. The gap between what you published and what that market searches for is your international roadmap, and it is usually larger than the technical backlog.