For AI, being mentioned beats being linked: the 0.664 vs 0.218 gap
Your SEO agency says you need more backlinks. Your competitor is the one ChatGPT recommends. Both things can be true, and the reason is a number that came out of the SEO industry's own tools in 2025 — and quietly rewrote where off-page effort should go.
the short answer
For AI visibility, being talked about beats being linked. A 2025 study of 75,000 brands found unlinked web mentions correlate with AI Overview visibility at 0.664, versus 0.218 for backlinks — about three times stronger. AI engines read your brand as an entity built from language, not from the link graph, so earned third-party mentions carry more signal than links. Links are not dead; they are just no longer the highest-impact off-page work for getting cited. Spend the marginal dollar on earned mentions.
key takeaways
- A 2025 study of 75,000 brands found web mentions correlate with AI Overview visibility at 0.664, versus 0.218 for backlinks — roughly 3x stronger.
- The three strongest signals are all off-site: web mentions (0.664), brand anchors (0.527), and brand search volume (0.392).
- YouTube mentions correlate even higher (0.737); AI understands brands as entities, through language, not through the link graph.
- Correlation is not causation — but the direction is consistent across studies: AI cites earned, third-party media over owned content or bought links.
- The budget implication: the marginal dollar buys more AI visibility spent on earned mentions than on link-building.
correlation with AI visibility — 75,000-brand study, 2025
In words, so the bars do not carry it alone: a 2025 analysis of 75,000 brands measured how strongly different signals correlate with showing up in AI answers. The strongest were all off-site — YouTube mentions at 0.737, web mentions at 0.664, brand anchors at 0.527, brand search volume at 0.392 — while backlinks, the classic SEO currency, trailed at 0.218. Mentions out-correlate links by roughly three to one. The brands earning the most web mentions pulled up to ten times more AI-answer mentions than the next tier.
What the gap is actually telling you
The instinct is to read this as one tactic beating another, but the deeper message is about how AI decides what to trust. Google's link graph infers authority from who points at whom; AI engines infer it from how a brand is talked about in language they can read and weigh. Those are different signals, and the study makes the difference measurable: the things that predict AI visibility are overwhelmingly off your own site and mostly not links. Three of the top signals — web mentions, brand anchors, brand search volume — are all earned reputation showing up as language across the web, not engineering you do on your pages.
It is worth being careful here. The people who published this number had no reason to flatter earned media: Ahrefs is an SEO tool, its business is links and rankings, yet measuring 75,000 brands it kept finding the dominant signal was the one its own product does not sell. That is worth trusting precisely because it cuts against the source's incentives. The authors are explicit that correlation is not causation, and we hold the claim at that strength — but the direction has been consistent across independent 2025 studies, and it points the same way every time: toward earned mentions.
The shift, in three parts
Why mentions win, what actually counts as one, and why none of this means links are finished. Open each layer for the part that matters to how you spend.
01 Why mentions beat links for AI
Search engines spent two decades treating the link as the unit of trust: a page voting for another page. AI engines read differently. They build an understanding of your brand as an entity — who you are, what you do, who vouches for you — out of language, by reading how often and in what context you are talked about across the web. A link is a thin signal in that model: it points, but it does not say much. A sentence in a respected publication that names you, describes you, and places you among credible peers is dense with the exact information the model needs. That is why unlinked mentions out-correlate links by roughly three to one: the model is reading meaning, and mentions carry more of it.
02 What actually counts as a mention
Not all mentions are equal, and the ones that move AI visibility are earned, not manufactured. A reference to your brand on a credible third-party source — editorial coverage, an analyst note, a podcast, a community thread where real users compare options — teaches the model that others independently find you worth discussing. Mentions you place on your own properties or pay for directly teach it far less, because the credibility signal is missing. This is the uncomfortable part for teams used to controlling their channels: the highest-value mentions are precisely the ones you cannot fully control, only earn. Breadth and credibility of who is talking about you matter more than the raw count, and a handful of trusted sources can outweigh a flood of low-quality ones.
03 Why this is not "links are dead"
It would be easy to overcorrect into abandoning links, and that would be a mistake. Backlinks still move classic organic rankings, still build domain trust, and still correlate weakly-but-positively with AI citations — they have not stopped working, they have stopped being the best lever for this specific job. The right read is opportunity cost: if your off-page budget is fixed, every dollar on a generic guest-post link is a dollar not spent on the branded mention, review, or feature that carries the stronger signal. Keep the link-building that earns search traffic; just stop treating it as your AI-visibility strategy, because the data says the marginal dollar works harder somewhere else.
Why language beats the link graph
The reason sits in how the two systems were built. A search engine ranks pages, and a link is its vote: page A endorses page B, and enough endorsements lift B up the list. An AI engine generates an answer, and to do that it has to know what is true and who is credible on a topic — which it learns by reading the web as text. When it encounters your brand named repeatedly, in context, on sources it has learned to trust, it forms a confident entity: this company exists, does this, and is regarded this way. A pile of backlinks does not build that picture nearly as well, because a link asserts a connection without describing it. Mentions describe; links merely point.
This also explains a pattern that frustrates a lot of well-optimised companies: they rank, and still go uncited. Research into generative engines has found that even strong pages can be skipped if their authority lives only on their own blog, and that pages improve their odds of being cited when they sit within a web of external references rather than standing alone. Ranking is something you can earn largely through on-site and link work; citation increasingly requires the off-site reputation that only mentions build.
Where this sends your budget
The practical move is to rebalance off-page effort, not to torch the link-building program. If a fixed budget currently buys guest-post links by the dozen, some of that money does more work redirected toward earning mentions: original data worth citing, third-party editorial coverage, podcast features, a real presence in the communities where your buyers compare options. The test for any off-page dollar becomes less "does this get a link" and more "does this get a credible source talking about us." Those are not always the same activity, and the data says the second one is now the stronger lever for being cited by AI.
None of this is a trick, and that is rather the point. Earning mentions is slower and less controllable than buying links, because you cannot manufacture a credible third party vouching for you — you have to be worth vouching for. That is the same discipline that rewarded genuinely useful work in every prior era of search, surfacing again in a new place: the brands that are talked about, clearly and often, on sources the model trusts are the ones it recommends. The agencies still selling links as the answer are selling the 0.218 signal while the 0.664 one sits unworked. The arbitrage will not last forever, but right now it is wide open.
Mentions and links: quick answers
Do backlinks still matter for AI search?
They matter, but less than they do for classic ranking, and far less than mentions. In a 2025 study of 75,000 brands, backlinks correlated with AI Overview visibility at 0.218, while unlinked web mentions correlated at 0.664 — roughly three times stronger. Backlinks still move organic rankings and still contribute to domain trust, so they have not stopped mattering; what changed is the opportunity cost. A dollar spent on a generic link is a dollar not spent on the mention, review, or feature that carries a stronger AI-visibility signal. The honest framing is not "links are dead" but "links are no longer the highest-impact off-page work for getting cited by AI." If your only off-page motion is link-building, you are optimising for the weaker signal.
What is the difference between a brand mention and a backlink?
A backlink is a clickable hyperlink from another site to yours; a brand mention is any reference to your brand in text, whether or not it links anywhere. The distinction matters because AI engines read language, not just the link graph. When a model is deciding what to cite, it is reading how often and in what context your brand is talked about across credible third-party sources — a pattern that mentions capture and bare links do not. A respected publication that names you without linking still teaches the model that you exist, what you do, and that others vouch for you. That is why unlinked mentions out-correlate links: the model is building an understanding of you as an entity, and mentions are the raw material for that, with or without the anchor tag.
How do I earn more brand mentions?
The same way earned media has always worked, now pointed at a new payoff. Publish things genuinely worth talking about — original data, clear strong viewpoints, useful tools — so that journalists, analysts, and communities reference you without being asked. Pursue third-party editorial coverage in credible publications, get featured on podcasts and in roundups, and show up in the communities where your buyers compare options. The point is breadth and credibility of who is talking about you, not volume of placements you control. Mentions you manufacture on your own properties teach the model little; mentions earned on sources it already trusts are what move the needle, which is exactly why this is harder, and more durable, than buying links.
Is this just PR rebranded?
It overlaps heavily with PR, and that is the point — the off-page work that earns AI citations looks more like earned media than like a link-building campaign. But it is PR with a measurement layer and a specific target: not coverage for its own sake, but the mentions and entity signals that AI engines actually read when they decide what to cite. Traditional PR optimises for impressions and sentiment; this optimises for being the entity a model recognises and trusts enough to name in an answer. So it borrows PR’s methods — relationships, newsworthiness, credible third parties — and aims them at a measurable visibility outcome. Calling it "just PR" undersells the targeting; calling it "new" oversells the novelty. It is earned media doing a job the link graph used to do.
A note on sources and certainty
The headline figures — web mentions at 0.664, backlinks at 0.218, the off-site signals topping the list — come from a 2025 analysis of 75,000 brands by an SEO toolmaker measuring what predicts AI-answer visibility. We present them as correlations, not proven causes, because the authors do, and the strength of the claim should match the strength of the evidence: a consistent direction across independent studies, not a single proven mechanism. The exact coefficients will shift as engines and datasets change, and we will date any material update. What has held all year is the ranking of the signals — earned, off-site reputation above links. The AC Group has argued for 27 years that being worth talking about beats gaming the index; this is the rare moment when the index's own measurers published the proof.