Ahrefs recently ran the numbers most agencies have been afraid to ask about: across roughly 75,000 brands, backlinks and ad spend correlated with AI visibility at about 0.22. Branded web mentions correlated at 0.66, about three times as strongly. That's not a small gap. It's a different metric doing most of the work.
It's worth understanding why, because the reason changes what belongs in a reporting deck.
Backlinks made sense when a link was a vote
A search engine's ranking algorithm counted links as endorsements: the more, and the more authoritative the source, the more a page was trusted to rank. That system rewarded volume and authority of linking domains, which is exactly what agencies got good at producing and reporting on for two decades.
An LLM isn't counting votes. It's looking for corroboration
When a generative model decides whether to name a brand in an answer, it isn't tallying inbound links to a URL. It's pattern-matching across everything it has read: does the web, independently and consistently, describe this brand the same specific way? A brand can have thin backlink profiles and still get named constantly in AI answers, if enough independent sources describe it accurately and specifically. A brand can have an enormous backlink profile and still get skipped, if nothing on the web makes a clear, checkable claim about who it is and what it's good at.
Why mentions beat links as the metric
A mention doesn't need a hyperlink attached to count. It needs to be a specific, verifiable claim, ideally with a source, a number, or a date attached, published somewhere the model has read. That's a much lower bar to hit technically, and a much higher bar to hit consistently across many independent sources, which is exactly the kind of distributed work a PR or partnerships agency already runs.
One honest caveat
Correlation isn't causation, and the Ahrefs data describes a relationship, not a guaranteed lever. The figures are Spearman correlations, drawn specifically from Google's AI Overviews. A brand doesn't win an AI answer by chasing mention count as a vanity metric any more than the old game was won by chasing raw backlink count. The actual work is the same as it's always been: say something specific and true about the brand, get it published in enough independently credible places that the pattern becomes undeniable, and keep the story consistent everywhere it appears.
What to put in the next reporting deck
Branded mention volume and consistency across a rolling window, not just backlink count and domain authority. Whether the description of the brand is the same, in substance, across the sources that mention it. And whether AI answers, checked directly across ChatGPT, Gemini, Claude, and Grok, are actually naming the brand for the buyer questions that matter, not just whether the brand ranks on a results page nobody but a bot is reading anymore.
