What Correlates With Being Named by an Assistant
In large brand samples, being described in third-party text tracks AI visibility more closely than link accumulation does, though the published coefficients disagree across platforms and dates.
Several large studies have now asked which observable properties of a brand move together with how often an assistant names it. The consistent result is that measures of being written about outrank measures of being linked to. The result is also weaker, more platform-specific and more contested than the headline coefficients suggest, and the publishers who produced it say so themselves. This report sets out what has actually been measured, where the measurements disagree, and what the disagreement implies for how an organisation reads any of it.
The coefficients
The most-cited figures come from Ahrefs, a vendor selling tools that measure brand visibility, which is worth stating before the numbers are. In May 2025 it examined 75,000 brands and computed Spearman rank correlations against brand mentions in Google AI Overviews. It reported that web mentions correlate at 0.664, much more strongly than backlinks at 0.218, with branded anchors 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. The domain sample was filtered to Domain Rating above 40 with highest-volume keywords at 800 or more monthly searches.
The publisher’s own qualification belongs in the same breath as the figures. Ahrefs states plainly that correlation is not causation, and that all the factors studied revealed moderate to very weak correlations on the Spearman scale. None of these numbers describes a mechanism, and none of them establishes that mentions produce visibility rather than both following from something else.
The distributional cut of the same dataset is easier to interpret than the coefficients. Splitting the 75,000 brands into quartiles by web mentions, Ahrefs found top-quartile brands recorded 169 AI mentions on average against 14 for the bottom quartile, and that 26% of brands have zero mentions in AI Overviews. A quarter of a large brand sample is absent from the surface entirely, which is a different kind of finding from a rank correlation and a more consequential one.
Where the studies disagree
Ahrefs extended the work across platforms in December 2025, again over 75,000 brands and again using Spearman coefficients. That study reports that YouTube mentions correlate more strongly with AI visibility than any other factor tested, at approximately 0.737 for ChatGPT, 0.740 for Google AI Mode and 0.712 for AI Overviews, with branded web mentions at 0.664, 0.709 and 0.656 and backlinks between roughly 0.23 and 0.32.
That ChatGPT figure sits awkwardly against the same publisher’s earlier work. In July 2025 Ahrefs measured branded web mentions against AI visibility on the top 50 websites mentioned per platform in its own Brand Radar product, across approximately 76.7 million AI Overviews, 957,000 ChatGPT prompts and 953,500 Perplexity prompts. It found a strong correlation on Google AI Overviews at 0.65, against 0.30 on Perplexity and 0.15 on ChatGPT — that is, effectively no relationship on the platform where the December study reports one of the strongest.
The two studies differ in date, in sample size and in construction: the July analysis rests on 50 domains per platform, the December one on 75,000 brands. Neither has been reconciled by its publisher. The honest reading is that the platform-level breakdown is unsettled, and that a coefficient quoted for ChatGPT without its study date and sample is not a usable number.
An independent study reaches the same conclusion about links without quantifying it. Seer Interactive, working from more than 300,000 finance and SaaS keywords narrowed through nearly 600,000 People Also Ask questions to 10,000 questions queried through OpenAI’s GPT-4o API, reported in January 2025 that backlinks’ impact was weak or even neutral against expectation, while Google page-one ranking correlated at roughly 0.65 with brand mentions in the model’s answers and Bing at roughly 0.5 to 0.6. Seer publishes no numeric coefficient for backlinks, and none should be attached to it.
Citation and naming are separate events
A finding that complicates every correlation above concerns what “visibility” is being counted. Semrush, in a June 2026 study conducted with Kevin Indig, examined 3,981 domain appearances across 115 prompts in fourteen countries and four surfaces, and found that 61.7% of domain appearances were citations in which the brand name never appeared, against 13.2% both cited and mentioned and 25.1% mentioned without any citation. Supplying the source of an answer and being named in it are largely different outcomes, and a quarter of namings come with no link at all.
For scale, Semrush also measured how often any brand is named. Across one million varied non-branded queries put to five systems in December 2025, it found that AI models include brand mentions in 26% to 39% of responses — ChatGPT 26.07%, ChatGPT Search 39.36%, Google AI Overview 36.93%, Gemini 31.14% and Perplexity 30.55%. No category breakdown or confidence intervals were published.
What this means for an organisation
The practical content of this body of work is narrower than its coefficients imply, and it is mostly a warning about measurement. Every figure here is a rank correlation on a sample the publisher chose, most of the publishers sell products that benefit from the finding, and the strongest single result — mentions over links — is stated by its own author as moderate to very weak. It supports a direction, not a target.
The direction is nonetheless consistent across four independent samples: what third parties write about an organisation tracks its presence in machine answers more closely than what points at its domain does. That is a statement about the composition of the record rather than about a tactic. An organisation described substantively in independent text, including in video, is in a different position from one that is merely linked, and the difference is visible in every study here even where the magnitudes disagree.
Two cautions follow for anyone acting on this. First, the platform-specific figures contradict each other and should not be used to allocate effort between assistants. Second, the Semrush ghost-citation result means a visibility count is only as meaningful as its definition: an organisation counting citations and an organisation counting namings are counting largely different things, since the two coincided in only 13.2% of the domain appearances Semrush observed.