Ask an AI assistant for the best value hyaluronic acid serum and it may well recommend The Ordinary. What is interesting is where it sends you. In one documented example, the AI recommended the product but cited a retailer and a national newspaper rather than the brand's own site (Exposure Ninja, "A Complete Guide to AI Search Optimisation for 2026", March 2026).
The brand won the recommendation. Someone else's page supplied the evidence.
That is the shift this article is about, and there is now real data behind it.
Why does what others say about you matter more than your own website?
Because AI answer engines are not following links, they are judging whether you are a recognized, credible entity worth naming. Ahrefs studied 75,000 brands and found that unlinked mentions of a brand across the web correlated with appearing in Google AI Overviews at 0.664, while traditional backlinks correlated at only 0.218 (Ahrefs). Being talked about was roughly three times as predictive as being linked to.
Key takeaways
- Unlinked brand mentions predicted AI visibility about three times as strongly as backlinks in the largest study available.
- The mechanism is entity recognition, not link counting. The system is deciding whether you are real and credible, not tallying votes.
- Your own website still matters and is usually one of your top cited sources. Authority supplements it, it does not replace it.
- Where the mentions live matters. ChatGPT leans encyclopedic, Perplexity leans community driven, and national press is weaker than people assume.
- This is correlation, not proof of causation, and buying fake mentions is a known way to waste money.
What did the Ahrefs study actually find?
Ahrefs took 75,000 brands and measured which signals lined up with a brand showing up in Google's AI Overviews. The headline gap is the one above: 0.664 for branded web mentions against 0.218 for backlinks.
For a business owner, the useful translation is this. Ten years of link building advice was built on a model where a link was a vote and votes accumulated into rank. That model still describes classic search reasonably well. It describes AI answer selection poorly.
A follow up analysis extending the same brand set across ChatGPT, Google AI Mode and AI Overviews reported YouTube mentions as the strongest individual signal, at roughly 0.737. We have only seen that follow up restated in secondary coverage rather than verified end to end, so treat the specific figure as directional rather than settled. The direction it points, though, matches everything else in the data: presence in places other people control beats polish on the page you control.
Why would an unlinked mention matter to a language model?
Because a language model has no link graph. It has a representation of the world built from text.
When independent sources repeatedly discuss a business by name, in context, alongside the category it operates in, the system's underlying model of the world starts to treat that business as a real, known, credible entity worth referencing. A clickable hyperlink adds nothing to that process that the words themselves did not already supply.
Technical breakdowns of how AI Overviews select sources describe an authority filter that behaves closer to a pass or fail gate than a sliding scale, applied before individual passages are ranked (ZipTie.dev). You either read as a credible entity or you do not. Getting slightly better on page does not move a gate.
This is why practitioners increasingly separate two ideas that classic SEO merged. Link authority helps you rank. Entity authority determines whether you get quoted once you are in the retrieval pool (Growthvibe).
Does this mean my website stopped mattering?
No, and anyone selling you that is overreaching.
Two things have to be true for you to be cited. Your page has to be retrievable and readable in pieces, which is ordinary technical and structural work. And you have to clear the credibility gate, which is the authority work. Skip either and the other is wasted.
The agency behind the serum example makes exactly this caveat: your own website is typically one of the top cited sources for your own brand unless something has gone wrong. Their point is not that owned content stopped counting. It is that owned content alone will not convince a system that has never seen anyone else mention you.
There is also a piece of folklore worth correcting here, because it gets repeated as fact. People often cite a 2012 Google patent about "implied links" as proof that Google ranks pages higher for unlinked mentions (SEOptimer). That reading is contested, and Google's own John Mueller has publicly said unlinked mentions are not used as a direct classic ranking signal in the way many assume (Search Engine Journal). The honest case for mentions is the AI citation data, not the patent.
Where do the mentions actually need to be?
Not everywhere. The mix differs sharply by platform, which means a generic public relations push is an expensive way to guess.
A large study of citation patterns found that Wikipedia alone accounted for roughly 7.8 percent of ChatGPT citations and nearly half of its ten most cited sources, while Perplexity leaned on Reddit, which supplied about 6.6 percent of its citations and nearly half of its top ten sources. Google AI Overviews sat more evenly across source types (Profound).
The counterintuitive result is what is missing. In a study of US ChatGPT citations, the Wall Street Journal, the New York Times and Bloomberg did not appear in the top 20 sources at all (5W Public Relations, via PR Newswire). A national newspaper feature is a fine thing to want. It is not the fastest route to being named by an AI assistant.
So the sequence is: find the prompts your buyers actually type, run them, write down which domains appear in the answers, and work on those specific domains. That is a research task before it is a marketing task.
What does honest authority building look like?
Ethan Smith, CEO of the growth agency Graphite, describes three levers that move the needle for his clients: strong landing pages, YouTube content, and genuine participation on Reddit. He notes these can start appearing in AI answers far faster than a traditional link building campaign would take to shift domain authority.
HubSpot co founder Dharmesh Shah explains the Reddit mechanism plainly: when a comment directly answers a question close to the user's prompt and has real upvotes behind it, it has a very high likelihood of being cited. The system is reading crowd validation, not brand copy.
Which leads to the failure mode both of them warn about. Smith describes the biggest mistake he sees as clients concluding that citations are mostly Reddit, then hiring an agency to run fake accounts that farm karma before dropping product mentions into threads. He says he has not seen it work, and points out that Reddit's own spam filtering is a large part of why AI systems trust it in the first place. His alternative is unglamorous: make a real account, say who you are and where you work, and answer questions usefully.
What this evidence does not prove
Three caveats, stated plainly, because the alternative is selling you something.
Correlation is not causation. The Ahrefs study shows that brands with more mentions appear more often in AI answers. It does not prove that manufacturing mentions causes citations. It is entirely plausible that businesses people talk about are also businesses that are genuinely better known and better regarded, and that the AI is picking up on the same underlying reality.
Measurement is unstable. The same prompt asked twice can return different sources. An academic study of the problem found that roughly 7 repeated runs are needed before the measurement error settles to a usable level (arXiv:2604.07585). Anyone reporting a precise authority score from one query pass is overstating their evidence.
Citations are not revenue. Agencies working in this space have publicly noted that brands are earning citations in ChatGPT and AI Overviews without matching growth in conversions or pipeline, and that the gap usually sits in the system around the citation rather than in the citation strategy itself. Being named is step one.
How do you find out where you stand?
Start by looking, not by buying. Ask the questions your customers ask, in a clean logged out session, and write down which businesses and which domains the AI names instead of you.
SignalCheck does that part automatically. It checks the structural side of your site and then runs a live grounded query asking an AI model who it would recommend in your category and city, from a neutral session with no saved memory, and shows the raw answer. The list of who does get named is usually the most useful thing in the report, because it tells you exactly which reputations you are competing against.
Frequently asked questions
Should I stop building backlinks? No. Backlinks still correlate with classic search ranking, and classic ranking is what puts you in the retrieval pool in the first place. The point is that links alone are a weaker lever for AI citation than most budgets assume, and mentions deserve a real share of the spend.
Can I just pay a service to get me mentioned everywhere? You can, and it tends to underperform. Bulk placements on low quality sites do not build entity credibility, and the fake community account version is specifically called out by practitioners as something they have not seen work. The mentions that count are on sources the AI already treats as trustworthy.
Does a Wikipedia page help? It appears to, since Wikipedia is one of the most cited sources for ChatGPT. But Wikipedia has notability rules, and creating a page for a business that does not meet them is a fast way to get it deleted. A cleaner starting point is a Wikidata entry plus structured data linking your verified profiles together.
How long does authority building take? Months, not weeks. Structural fixes on your site can move in 2 to 4 weeks, while consistent citation behavior across multiple AI engines typically takes 4 to 6 months (WebFX). Reputation accumulates at human speed regardless of how fast the software iterates.
If AI mostly cites other sites, why bother with my own content at all? Because your site is the source most likely to be quoted for questions about you specifically, and because retrieval systems still need a crawlable, extractable page to pull from. Authority determines whether you are trusted. Your content determines what gets said.