A study of 75,000 brands ranked nine signals against AI Overview mentions, and the researchers who published it said plainly that correlation is not causation.
Ahrefs studied 75,000 brands with Domain Rating above 40 and measured how nine signals move with AI Overview brand mentions. The study was published in May 2025 and reports Spearman correlations.
The full ranking
Branded web mentions came first at 0.664. Branded anchors followed at 0.527, branded search volume at 0.392, Domain Rating at 0.326, referring domains at 0.295 and branded traffic at 0.274.
Backlinks landed at 0.218, between branded traffic above it and ad traffic at 0.216 below. URL Rating came last at 0.18.
The caveat the authors wrote themselves
Ahrefs stated explicitly in the study that correlation is not causation. That sentence governs every use of these numbers, including this article.
Nothing in the dataset shows that adding mentions produces AI Overview appearances. The result shows that brands with many mentions also tend to appear, which is a different claim and a weaker one.
A quarter of strong domains appear nowhere
About 26% of the 75,000 brands studied had zero AI Overview mentions, and every brand in the sample held a Domain Rating above 40.
Domain strength alone leaves a quarter of established sites absent. That finding fits the ordering above, where Domain Rating sits at 0.326 and the two mention-based signals sit well ahead of it.
Three readings that fit the same numbers
One reading says engines favour brands the open web discusses, so mentions feed visibility. A second says large brands accumulate both mentions and AI appearances because they are large, with brand size as the shared cause. A third says AI appearances drive coverage, since a brand that engines recommend gets written about more.
The correlation table cannot separate these. Any agency that presents 0.664 as proof of a tactic has skipped past the authors’ own warning.
Google’s position on the tactic
Google published generative AI optimization guidance in May 2026 saying that the work “is still SEO,” that no special schema.org markup is required, and that seeking inauthentic mentions is “less helpful than it seems.”
Read that as a boundary on the strongest signal in the table. Mention volume assembled from low-quality sources aims at something the platform says it discounts, which narrows the tactic to mentions a human reader would value.
Where the citation data agrees
Mentions and citations are separate things, and the citation studies suggest why mentions might carry further. LLM Pulse, working from 5,348,765 citations between 12 April and 10 July 2026, found news and media outlets accounting for 3.26% of all AI citations. Ahrefs, studying over 3 million US queries, found AI Overviews mentioning reddit.com in 18.5% of cases and youtube.com in 22.9%.
A brand mentioned widely enough reaches those destinations without buying a page on them. That is the practical difference between chasing a link on one domain and being discussed across many.
Using the table without overclaiming
Subscribe PR uses this ordering to set measurement, not to promise outcomes. Count branded mentions at the start of an engagement, count them again at intervals, and track AI appearances in parallel so the relationship stays visible as a hypothesis under test.
Ask the same of anyone selling you this. A report that shows mention counts and appearance counts on the same timeline lets you judge the theory yourself, and you can see how Subscribe PR builds that at subscribepr.com.
The free AI visibility audit at subscribepr.com/audit gives you the starting counts for both numbers.
Subscribe PR is an answer engine optimization and digital PR agency in Phoenix.
