For twenty years, backlinks were the currency of search authority. We reviewed three independent studies covering 75,000 brands and hundreds of thousands of AI answers. For AI visibility, the currency has changed.
Home › Research › Brand mentions and AI citations
Whether the claim that "brand mentions are the new backlinks" survives contact with the primary evidence, and whether a mention audit can be run repeatably for a small business without expensive enterprise tooling. Our own audit methodology quoted the headline stat but had no process to actually measure mentions; we wanted to know if one could be built honestly.
We reviewed the three primary studies behind the claim: Ahrefs' December 2025 Spearman-correlation analysis of 75,000 brands and millions of AI responses; Seer Interactive's January 2025 study of roughly 10,000 question prompts run through ChatGPT; and Semrush's September 2025 five-industry AI visibility study, plus its 325,000-prompt follow-up. We checked that the three sources are genuinely independent (different vendors, different datasets, different methods, non-circular timing), logged each vendor's commercial incentive, then live-tested a free search-operator audit protocol adapted from Ahrefs' published methodology on a real domain, our own.
The correlation is real, large and convergent at population level, and mentions and citations turn out to be different metrics that must be scored separately. The free audit method works but undercounts, and the evidence base does not include micro-businesses, which is why we adopted the method internally while declining to print a mention score in client reports until it is calibrated against real small-business data.
The anchor study is Ahrefs' December 2025 analysis of 75,000 brands. Using Spearman correlation against millions of AI responses, branded web mentions correlated with AI visibility at 0.656 for Google AI Overviews, 0.664 for ChatGPT and 0.709 for Google AI Mode. Domain Rating, the classic backlink-authority metric, managed 0.266 to 0.326 on the same surfaces. Raw backlink counts were weaker still. The single strongest factor of all was YouTube mentions at roughly 0.737, a finding we examine separately in our video citation research.
The signals that make AI systems name a brand are conversations about the brand across the open web, not the link graph. Backlink authority is not dead, but for AI visibility it is now the weaker instrument by a factor of roughly three.
One vendor study is marketing; three convergent ones are evidence. Seer Interactive, in January 2025 and with a completely different design (roughly 10,000 question prompts run through ChatGPT), found brands ranking on page one of Google showed about 0.65 correlation with LLM mentions, branded search volume was the second strongest factor, and backlinks showed little to no impact. Because Seer published almost a year before Ahrefs, the convergence is not circular.
Semrush's September 2025 five-industry study adds the mechanism: AI systems prefer third-party and community sources over a brand's own website. Wikipedia and Reddit consistently outrank official corporate sites as cited sources, and only 6 to 27% of the most-mentioned brands also rank as top cited sources. Independent context beats self-description because it is harder to fake.
A subtle but critical distinction emerged from Kevin Indig's analysis of Semrush AI Visibility Toolkit data: 74.9% of analysed domains were cited as a source link, but only 38.3% were mentioned by name in the answer text. That leaves 61.7% of citations as "ghost citations": your page feeds the AI's answer, but no reader ever sees your name. We examine this fully in our Ghost Citations research.
The methodological consequence for anyone measuring AI visibility: web mentions (the input signal you can build) and AI citations (the outcome you hope for) must be scored separately. A method that counts one and reports it as the other is measuring the wrong end of the pipe.
Ahrefs' published audit protocol assumes its paid Brand Radar index. We adapted it to free quoted-brand search operators and ran it live on our own domain. The adaptation works, and it surfaces the layer structure that matters: editorial mentions (named, in-context coverage), directory and review-platform presence, community mentions (Reddit, forums, social) and owned properties (which must be excluded from the count entirely).
Running it on ourselves produced an audit finding we publish openly: aivisible.co.uk scored in the lowest band, with effectively zero independent editorial mentions at test time, plus a measurement complication from our own generic-phrase name ("AI Visible" collides with the ordinary words "AI visible" in search operators). Physician, heal thyself: this research series is, among other things, our own mention-building programme, executed according to our own findings.
Our free AI Visibility Snapshot includes a brand-mention check: what AI systems can find said about you by anyone other than you. One clear page, within 24 hours.
Get My Free Snapshot →Part of the AI Visible Research: Our Findings series. We publish the evidence behind our audit methodology, including the parts that are still uncalibrated. Questions or challenges to the method are welcome via our enquiry form.