Multilingual SEO

Multilingual SEO is five decisions: one URL per language, telling Google which is which, letting people choose, translating the page rather than the template, and researching each language separately. The first two get reviewed. The third, how a site routes visitors between languages, almost never does — so on 27 September 2026 I checked what 110 international brands actually do.

Multilingual is not multi-regional

A multilingual site serves the same audience in several languages. A multi-regional site serves several countries, sometimes in one language. Most international sites are both, and the confusion between the two causes a surprising share of problems: an en-GB and an en-US version are regional variants, not translations, and need different handling from a German one.

One thing worth knowing before anything else: Google works out a page’s language from what it can read. In its own words, “Google uses the visible content of your page to determine its language. We don’t use any code-level language information such as lang attributes, or the URL.” A German URL with an English body is an English page.

The five decisions

1. One URL per language version

   Every language needs its own address that anyone can open directly. Subfolders, subdomains and country domains all work; URL parameters are the one structure Google calls “not recommended”. The trade-offs are in subdomain vs subfolder vs ccTLD.

2. Tell Google which version is which

   That is what hreflang does: it links the versions of a page and says who each is for. It is also where most technical errors sit. The hreflang guide covers the rules and the mistakes, and the hreflang checker tests a live page.

3. Let people choose — do not decide for them

   Google’s advice is blunt: “Avoid automatically redirecting users from one language version of a site to a different language version of a site.” The reason is that such redirects “could prevent users (and search engines) from viewing all the versions of your site.” Link the versions to each other instead. This is the decision the data below is about.

4. Translate the page, not the template

   Google warns that translating only the navigation and footer while the main content stays in one language “can create a bad user experience”. At the other extreme, its spam policies count translation among the “automated transformations” used to turn scraped content into pages, and treat generating many pages “without adding value for users” as scaled content abuse. Machine translation of your own pages is not banned; mass-producing pages nobody reviewed is where the risk sits. Content parity is how you measure the gap.

5. Research each language separately

   The translated keyword is often not the one people search. International keyword research explains why, and SEO localisation vs translation covers what else changes per market.

What 110 international brands do with language routing

Google’s crawler, by its own documentation, “sends HTTP requests without setting Accept-Language in the request header”, and its default IP addresses “appear to be based in the USA”. A site that picks the language from the browser or the IP address therefore shows Googlebot one version — whatever its default is — unless every version also lives at its own reachable URL.

So I requested the home page of each of the 110 brands from my hreflang survey three times from Kraków: with no Accept-Language header, as Googlebot sends, then as a German browser and as a Spanish one. Redirects were followed one hop at a time. Seventy-two answered all three requests; the rest refused an identified crawler (24 answered 403) or timed out.

Nearly half adapt the home page to the visitor in one way or another, which on its face contradicts Google’s advice. The detail is more interesting than the headline.

The mature pattern is to adapt the entry point, never the language pages. For the 21 sites that routed Poland to Polish, I derived each German URL from the Polish one and requested it the same way, from Poland with no language header. All 21 served German; twenty stayed on the exact URL and Stripe normalised /de-de to /de. Of the twelve sites where a German browser landed on a distinct German URL, ten kept that page when a Spanish browser asked for it. The root is a router; every version stays reachable by anyone, crawler included.

Two redirect the language pages themselves. Zoom and Spotify send a visitor without a German browser away from zoom.com/de and open.spotify.com/intl-de — yet both serve those URLs in German to a request identifying as Googlebot. They built an exception for the crawler. A team that copies the redirect without the exception bounces Googlebot, which sends no language header, off every page that is not in the default language.

Three switch language at the same URL. Dropbox, Wise and Coursera return a different language at their home page depending on the browser. Dropbox and Wise also publish every language at its own URL and point x-default at the adaptive home page, which is the arrangement x-default exists for. Coursera’s home page declares no German alternate at all, so whatever German it shows a German browser, Google has no URL to index it under.

One more, for whoever runs the CDN: only one of the 72 sent Vary: Accept-Language. Without it, a cache is entitled to serve the German version it stored to the next visitor, whatever their language — a problem for sites that switch content at the same URL.

If you adapt at all, adapt like this

• Only the bare root. A visitor to / can be routed; a visitor to /de/ asked for German and gets German.
• Make the root your x-default, and list every language URL in hreflang.
• Prefer a banner to a redirect. “This page is also available in German” respects the person who wanted English and costs nothing in crawling.
• Remember the choice. Someone who switched language once should not be switched back on the next page.
• Test it the way Googlebot arrives: no Accept-Language, from an IP outside your market, against a language URL rather than the home page.

Sources

• Google Search Central, Managing multi-regional and multilingual sites and How Google crawls locale-adaptive pages, both last updated 10 December 2025.
• Google Search Central, Spam policies for Google web search, last updated 28 August 2026: scaled content abuse.
• Home pages of 110 international brands, each requested three times on 27 September 2026 from Kraków with an identified user agent: no Accept-Language, German and Spanish. 72 answered all three. Follow-up requests to derived German URLs the same day, and one request each as Googlebot’s user agent to the two German URLs that redirected.

Questions