In 2018, Glenn Gabe documented a case that confused international SEOs. A client sent UK and Indian hreflang URLs to a US canonical, which is technically wrong, yet Google kept showing the right country page in search. Search Console said those URLs were not indexed. John Mueller confirmed it wasn’t a fluke.
Eight years later, Google gave the behavior a label. In August 2026, Gary Illyes replied to a LinkedIn question and called these alternates “alternate names.” The explanation separates discovery, canonicalization, indexing, scoring and serving. A URL can sit outside the index but still be chosen at the SERP because Google knows its relationship to an indexed page.
Hreflang is a cross-reference, not a ranking signal
The practical takeaway is summed up by Search Engine Journal: “hreflang routes rather than ranks.” It does not make a page score higher. It tells Google that a UK URL is the British equivalent of a US URL, or that an Australian page serves English speakers in Australia. Google uses that mapping after a ranking has been determined.
If a US page already ranks in the UK, adding a correct UK alternate won’t give the UK page fresh ranking power. It only changes which URL gets displayed for the result Google already has.
The 38-market duplicate problem
One global shoe brand shows the scale of the issue. It ran around 98 market sites, about 38 in English and nearly identical. Of those, 22 even presented the same dollar-sign currency. To the organization these were separate market websites; to Google they were essentially one resource with dozens of addresses.
Google can cluster near-duplicates, choose a canonical version, and keep enough relationship data to present the right local URL. That is the “magic trick” from 2018: the alternate URL never needed its own ranking score because the serving layer already knew it was the right alternate name.
AI raises the bar for international content
Semantic retrieval changes the question. Microsoft’s Web IQ infrastructure uses multilingual models that place equivalent information close together, even across languages. If 20 pages contain the same product descriptions, FAQs and specs, there’s little reason to retrieve one over another. The unit that matters is the information, not the number of URLs.
Hreflang can declare a relationship, but it cannot create local relevance. Translating text and swapping currency, phone numbers or spelling doesn’t make a market page valuable.
What to do
- Treat hreflang as a routing table, not a growth lever.
- Before launching another market site, find what buyers there genuinely need to know that others don’t.
- Pull local teams into SEO content decisions.
Those differences might include regulations, certifications, warranties, delivery options, pricing models, competitor sets, integrations, use cases and local customer evidence. That is the material a machine can actually retrieve.
The technical annotation establishes the relationship; local knowledge establishes the difference. As retrieval gets better at recognizing equivalent content, the sites that win will be the ones where every market page has a reason to exist.
Source: Search Engine Journal
