Google can serve hreflang URLs it never indexed on their own

In August 2026, Google’s Gary Illyes answered a LinkedIn question about country URLs that appear in local results without being indexed and described those URLs as “alternate names,” according to Search Engine Journal’s account of the exchange. The explanation separates discovery, canonicalization, indexing, scoring and serving into distinct steps, so a URL does not have to pass through all of them on its own to show up in search.

That puts a name on something international SEOs have been reverse-engineering since 2018, when Glenn Gabe, an SEO consultant known for his analyses of Google updates, documented a client that canonicalized its UK and Indian hreflang URLs to the U.S. Version. The setup broke Google’s own guidance, yet the UK and Indian URLs kept surfacing in their country results while Search Console reported them as not indexed. Google’s John Mueller confirmed the behavior in a webmaster hangout: Google could swap in the correct country or language URL at the SERP level even when only one page in the hreflang cluster was indexed.

The reporting consequence is narrow. A not-indexed alternate is not evidence that the market is unserved, because indexing status describes whether Google stores that URL as its own document, not whether it can be shown to someone in Italy or Australia. The better check is the market’s own results, or the Performance report filtered to that country. That fits Mueller’s separate point that Search Console’s indexing data trails its performance data.

Hreflang picks the URL, not the ranking

Hreflang does not earn the ranking it affects. Bill Hunt, president of the international marketing consultancy Bisan Digital and the piece’s author, reports that across tens of thousands of implementations he has never seen the annotation itself raise a page’s position. If a U.S. English page already ranks in the UK, adding a correct UK alternate changes which URL is presented, not the score behind it. Hunt’s model is a cross-reference table, and it lines up with what Illyes described: indexing and scoring land on one cluster member while the serving layer selects the alternate that fits the user.

When the alternates have nothing of their own

The arrangement gets expensive when the alternates carry no information the canonical lacks. A global shoe brand Hunt has worked with ran roughly 98 market websites, about 38 of them in English and close to identical. Of those, 22 shared the same dollar-sign currency symbol. Dozens of URLs held the same product information in the same language, often with the same pricing presentation, separated by domain or directory and a handful of localized details. Hunt’s argument is that Google has little reason to store, process and score 38 near-copies, so it clusters them, picks one representation and keeps enough about the alternatives to serve the right one by country.

Clustering is survivable in blue links, where Google only has to choose one URL to display. AI answers work differently: the system retrieves information, pulls evidence from more than one document and assembles a response, so the unit that matters becomes the information rather than the URL. Motoko Hunt, an international search consultant who also writes for Search Engine Journal, makes the point in a piece Bill Hunt cites that hreflang can declare which market a page is intended for without making that page the best evidence for a searcher in that market. Microsoft’s description of Web IQ, the retrieval and grounding system behind its AI products, says its multilingual models place semantically equivalent information as close together as possible even across languages. For a brand whose English market pages all say the same thing, the likelier outcome is that one of them gets used and the rest are redundant.

What to do

• Confirm serving before debugging indexing. Run the market query from that country, or filter the Performance report by country. If the right URL appears, the not-indexed label describes storage, not visibility.
• Keep discovery clean during rollouts and migrations. Sitemaps, internal links and hreflang with complete return tags are how Google learns which URLs are equivalents. None of it buys ranking.
• Put the recovered time into differences a machine can see: local currency and pricing, stock and shipping that match the market, local regulations and certifications, customer examples from that market, and the phrasing searchers there use. One brand’s translated titles missed the phrase local searchers type, which is the kind of gap that separates two otherwise identical English pages.
• Where a market page cannot carry anything its canonical lacks, ask whether it needs to be a separate page at all. Bill Hunt’s piece stops short of saying so, but 38 English sites competing to be the same document is a case for consolidation.