Download websites that use hreflang

About hreflang

hreflang is not a product, a vendor or a library, but an attribute a page attaches to a link element to declare that another URL carries the same content for a different language or region. Nobody licenses it and nobody supports it. A search engine reads it and decides which version of a page to show a searcher in Lyon rather than in Leeds.

Because the markup only exists where alternates exist, presence is a business signal wearing technical clothes. A site annotates alternates after somebody translated the content, somebody decided which markets deserve their own URL, and somebody kept the mapping current as pages were added and retired. Sites doing that are running an international operation with a budget behind it. Sites that sell abroad from one English page never appear here at all, which is exactly what makes the column a qualifier rather than a description.

As of October 2026 we see the annotation on 7,116,165 sites, 3,458,563 of them in United States, on a domain mix where .com is heavily represented. Country in this particular row is close to the wrong axis, and reading it naively will mislead. A German retailer selling into France and Poland is recorded as a German site. What distinguishes rows here is not where the site is registered but how many alternates it declares and which locales it names.

Search engines have honoured the convention since 2011, and it has stayed stubbornly difficult to implement correctly. Return references have to be reciprocal. Region codes get confused with language codes. Canonical tags contradict the alternates they sit beside, and one mistake quietly removes the wrong page from the wrong market with no error message anywhere. Agencies have built entire audit products on the failure modes, which tells you how common the failures are.

Our detection reads what we can read, and the shortfall is systematic rather than random. The annotation is equally valid in HTTP response headers and in XML sitemaps, and neither placement appears in page markup. Large multilingual estates are precisely the ones that push the mapping into sitemaps, because putting a long list of alternates in the head of every page is a payload nobody wants to serve. The count should be treated as a floor for internationalised sites rather than a census of them, and the sites missing from it skew towards the larger end.

Localisation agencies, translation platforms and cross border payment providers want the same qualification out of the file: prospects that already committed to selling in more than one language, which is a far stronger buying signal than company size. SEO tooling vendors want the opposite end of the quality range, the sites whose alternates are declared badly enough to justify an audit. Trade researchers use the attribute as a cheap proxy for genuine market entry, since publishing a maintained alternate for a country is a heavier commitment than translating one landing page.

For a single domain, a browser extension settles the question in seconds. The population questions do not decompose that way. What share of a national retail sector has any localisation at all, which locale pairs travel together, whether a market gained or lost coverage after a platform migration, which content management systems generate the annotation automatically and which leave it to hand: each needs the whole set at once. The download carries that set with country, suffix and detected platform on the same row, so international reach can be crossed against the stack that produced it.