Expanding a store into new markets is the point where most technical SEO debt becomes visible. Duplicate the catalogue across six country sites without a plan and you have multiplied every existing problem by six, plus a new one: search engines cannot tell which version belongs to which audience. International ecommerce SEO is mostly the discipline of proving that distinction — and being honest about when a separate version is not justified at all.
The first question is not “how do I set up hreflang.” It is “does this market need its own URLs?” Get that wrong and no amount of annotation saves you.
When a Separate Market Version Is Justified
A distinct URL set earns its keep when the commercial reality genuinely differs. Different currency and pricing. Different shipping options, delivery times and costs. Different stock — half your catalogue may not ship to that country at all. Different tax display conventions, different returns law, different sizing conventions, different regulatory labelling.
Those differences are the substance a separate page is built on. A German shopper needs prices in euros, VAT-inclusive display, EU return rights and delivery from a warehouse that ships there. That is a materially different page, not a translated one.
What does not justify separate URLs is a machine-translated copy of the same catalogue with the same currency, same shipping and no local presence. That is thin duplication at scale, and it tends to cannibalise rather than expand. If the only difference between your .de and .com pages is that a script swapped the words, do not launch it — spend the effort on one strong market instead.
Choose a URL Structure and Live With It
Three viable structures, each with a real trade-off:
- ccTLDs (
example.de,example.fr): the strongest geographic signal and the clearest trust cue for local shoppers. Also the most expensive — separate domains build authority separately, so each one starts from zero. Choose this when you have real local operations and the budget to grow several domains. - Subdirectories (
example.com/de/): everything inherits the authority of one domain, one crawl surface, one set of technical fixes. The pragmatic default for most retailers, and the easiest to unwind if a market underperforms. - Subdomains (
de.example.com): sits between the two, and in practice inherits less than a subdirectory while costing nearly as much operationally as a ccTLD. Choose it when infrastructure or platform constraints force your hand, not because it looks tidier.
Whichever you pick, keep language and country distinct in the path when you serve multiple languages in one country — /ch-de/ and /ch-fr/ rather than a single Swiss version that guesses. Never encode market in a query parameter, and never use a cookie or session state as the only thing separating versions; a crawler will only ever see one of them.
Hreflang: The Three Rules That Break Implementations
Hreflang tells search engines which version to show which audience. It is not a ranking booster and it does not consolidate authority — it is a routing hint, and it fails loudly when done carelessly.
- Return links must be reciprocal. If the UK page points to the German page, the German page must point back to the UK page. One-way annotations are ignored. This is the single most common failure, and it usually appears when a market is added and the existing pages are not updated.
- Every set needs an
x-default. That is the fallback for users whose language and region match nothing in your set. Point it at your global or language-selector page, not at a random market. - Every page in a cluster must list every other version, including itself. Partial sets produce partial behaviour.
Use correct codes: ISO 639-1 for language, ISO 3166-1 Alpha 2 for region, in that order — en-GB, not en-UK, which is not a valid code and is annotated wrongly on an alarming number of stores. Implement in the HTML head or in the XML sitemap; both work, and the sitemap approach scales better across large catalogues because you are not asking a template to render a hundred link tags on every product page.
Hreflang also has to survive product churn. When a SKU is discontinued in one market but not another, the cluster breaks. Build the annotation from live availability data rather than a static map, or accept a monthly cleanup job.
Canonicals and Hreflang Are Not Interchangeable
A frequent and damaging mistake: pointing every market version’s canonical at the primary market. That instructs search engines to drop your local pages entirely, and it overrides the hreflang set you just built. Each market version must self-canonicalise.
Hreflang handles “these are equivalents for different audiences.” Canonical handles “this is the preferred version of duplicate content within one audience.” Use both, keep them consistent, and never let them contradict each other — when they conflict, the canonical usually wins and your international rollout quietly disappears.
The related trap is English variants. en-US, en-GB and en-AU pages with identical copy are near-duplicates by every measure except the annotation. Differentiate them where it matters — spelling, sizing, currency, shipping, seasonal framing (a July page in Australia is a winter page) — so there is a real reason for each to exist.
Do Not Auto-Redirect by IP
Automatically redirecting visitors based on IP geolocation is one of the most common ways stores lock crawlers out of most of their own site. Googlebot crawls predominantly from one region; if every request from that region is forced to your US site, the other markets may never be seen properly.
Serve a suggestion instead. A dismissible banner offering the local version, with a persistent, crawlable link to a market selector, gives users the convenience without hiding pages. Make the selector a real HTML list of links, not a JavaScript dropdown that generates no anchors.
Translation, Localisation and the Quality Floor
Machine translation as a starting point is defensible in 2026; publishing it unreviewed is not. The failure is rarely grammar — it is that translated product copy misses how people in that market actually search. Nobody in France searches for a literal translation of your English category name if the local term is different.
So do keyword research in-market rather than translating your keyword list. Country-specific search indexes matter enormously here, because volume, difficulty and competitive landscape differ by market and querying the wrong index gives you a near-empty or misleading picture. Then run a content gap against local competitors, not your home-market rivals — the retailers ranking in Germany are usually not the ones ranking in the US, and the gap list is completely different.
For product copy at scale, an AI writer with brand-voice support gets you differentiated text rather than translated manufacturer boilerplate, which is the duplication trap arriving in a second language. Human review on the top revenue SKUs is still non-negotiable.
Technical Housekeeping That Scales With Markets
Every market multiplies your technical surface, so the hygiene items matter more, not less:
- Separate XML sitemaps per market, each containing only that market’s canonical, indexable URLs.
- Product schema with the correct
priceCurrencyper market — a euro price marked up in USD is a mismatch between markup and visible content. - Local structured data for physical operations, and market-appropriate contact details, which do more for trust than any annotation.
- Full-site crawls run per market, because a template bug frequently affects only one locale. Crawls that return actual titles, URLs and H1 text let you spot a locale where a variable never populated.
- Search Console properties configured for each market version so you have ground truth per region rather than one blended number.
Measure Each Market on Its Own Terms
The biggest reporting error in international ecommerce SEO is aggregating markets. A rollout that is working in two countries and failing in four looks like modest growth in a blended chart. Segment everything by market from day one.
Give each market its own baseline, its own competitor set, its own head-term basket, and its own weakest-page-one benchmark — the bar in a less contested market is often far lower than at home, which is exactly why the expansion may be worth it. Expect six to twelve months before a new market version reaches steady state, and remember that third-party volume and position figures are modeled estimates that vary in accuracy by country, so lean on Search Console for anything you plan to defend in a meeting.
SEO Rocket handles the research side across country-specific indexes, the competitive gap work and per-market crawls in one workspace at a flat US$50 a month. Hreflang implementation still belongs to your developers — no tool does that for you, and no tool should pretend to.
