Google wants AI search to cite more websites. Brands now need to prove which ones are theirs


GA Team | GA Agency GA Team | November 19, 2026| 4 minutes

Google wants AI search to cite more websites. Brands now need to prove which ones are theirs

Two pieces of search news landed on the same day in November and they are more interesting if read together.

On 12th November, Google rolled out Gemini 4 Hydrogen across Search and described it as “the biggest model upgrade to Search yet”. One of the promised improvements is a broader range of links in AI answers, including sites users “previously might not have discovered”. The same day, a brand-protection firm published a study showing that AI answer engines already struggle to tell a brand’s official website from an impersonator or an abandoned domain.

Coverage has mostly treated these as separate stories, one about product capability and one about consumer safety. For brands, they describe the same shift: as AI search draws on a wider set of sources, being cited is no longer the only question. What matters is whether the right version of the brand is cited, in every market where it trades.

What changed in Google Search

Google says Gemini 4 Hydrogen is now available globally in Google Search for all users, three different interesting takeaways: the first is query fan-out. Google’s documentation describes it as issuing multiple related searches across subtopics and data sources to build a response. Google says Hydrogen improves this by “conducting simultaneous queries across all the relevant subtopics”, which connects users to “a broader range of helpful links than before”.

The second is format. Google says users will see “more images, structured responses, comparison tables and custom tools”.

The third is language. Google says responses are more nuanced “across all supported AI Mode languages”. The announcement does not say which languages those are.

The post offers no data on how much wider the range of linked sites has become. For now, the claim is just a statement of intent.

What the shopping study found

The second story comes from Podqi, an intellectual property protection firm, which shared its findings with The Verge. In September 2026, Podqi put the same shopping questions about 151 consumer brands to ChatGPT, Gemini and Perplexity and collected 6,795 answers.

Podqi classed a site as risky if it appeared to impersonate a brand, offered suspected counterfeits or was dead. Of the 4,437 domains it reviewed, 718 met that definition. Of those, 351 were fake and 367 were dead.

Across all answers, 9% included a risky site. The rate was 15% for Perplexity, 8% for ChatGPT and 4% for Gemini. When a risky site appeared, the answer recommended it 57% of the time, mentioned it without comment 34% of the time and warned against it only 9% of the time.

These are vendor figures and should be read that way. Podqi wrote its own prompts rather than using real user conversations, so the results show how the engines behaved on these questions, not how often real shoppers meet fake sites. To note: the tests took place before Hydrogen launched and Gemini recorded the lowest rate of the three engines.

Why this belongs on the agenda

Being represented correctly is a different thing than being cited 

Most discussion of AI search has focused on presence, meaning whether a brand appears in the answer at all. The Podqi data raises a second question: when a risky site appeared, the engine did not just list it, in 57% of those answers, it recommended it. 

So a brand can be named in an answer and still lose the customer to an address it doesn’t own. In AI search, how accurately a brand is represented needs measuring alongside whether it appears.

Retired domains are now a visibility risk

Just over half of the risky domains in the study, 367 of 718, were dead rather than fraudulent. Being dead does not mean a domain is scamming anyone. But a lapsed address can be re-registered by someone else and Google already recognises the pattern. Its spam policies define expired domain abuse as an expired domain that is “purchased and repurposed primarily to manipulate search rankings”.

A brand that has closed a regional site or let a campaign domain lapse has left behind an address that answer engines may still put in front of shoppers. A domain portfolio is now part of search strategy, not just IT housekeeping.

The problem multiplies across markets

The riskiest prompt in the study asked about a brand’s regional websites. On that prompt, 16% of answers showed a risky site, compared with 6 – 7% for most of the other questions. That is the question an international shopper is most likely to ask.

Google’s language claim is unquantified and the study publishes no breakdown by market. So there is no published basis for assuming that an engine’s accuracy in one market carries over to the next. For brands trading across several markets, accuracy has to be checked market by market.

What to do about it

The first step is to establish which domains AI engines currently name for the brand in each market and language. The regional-site and “where to buy” questions that produced the highest risk rates in the study are the best place to start.

The second is to treat the domain portfolio as a search asset. Retired regional sites, campaign domains and legacy addresses should be catalogued and kept under the brand’s control. 

The third is to make the brand’s official identity unambiguous in the data Google already reads. Google says its AI features are rooted in its core Search ranking and quality systems and need no special markup. Its documentation also says that Organisation structured data on a home page helps it disambiguate an organisation in search results. For a brand with several regional sites, keeping that consistent is as much a coordination task across markets as a technical one.

As AI search widens the set of sources it draws on, the brands it represents most accurately will be those whose official presence is easiest to verify.

If you need support with this, that’s a conversation worth having with our team.

Subscribe to our newsletter to stay updated

"*" indicates required fields


Related articles