Brand mentions beat backlinks for AI citations: the 2026 data

For twenty years, off-page SEO meant one thing: build links. Then researchers started measuring what actually correlates with being cited in AI answers, and the ranking of levers flipped. In a study of 75,000 brands, unlinked brand mentions lined up with AI citations far more tightly than backlinks did. If your off-page playbook is still link-first, the data says it is optimizing for the wrong thing.

This is a big claim, so it is worth being careful about what the study shows and what it doesn't - and then about what to do with it.

What the study found

Ahrefs measured the statistical relationship between various brand properties and how often a brand appeared in AI answers across AI Overviews, ChatGPT and Google AI Mode. The correlations told a consistent story:

• Unlinked brand mentions had the strongest relationship among the classic off-page signals - roughly three times stronger than backlinks.
• Branded anchor text came next - links where the anchor is your brand name, which is really a mention wearing a link.
• Brand search volume - how often people search your name - also outperformed raw link counts.
• Total backlinks were the weakest of the named off-page factors.

A caveat worth stating plainly: correlation is not causation, and the exact numbers vary by how each source reports the study. But the direction is stable across multiple analyses - the things traditional SEO optimized hardest for have a weaker statistical link to AI citations than the things it largely ignored.

Why mentions matter more to a model

The reason is in how models understand your brand. A search engine ranking a page cares about link equity - a vote from one page to another. An answer engine assembling a response cares about entities: does it recognise your brand, know what category you are in, and trust what it knows enough to name you?

Unlinked mentions feed that entity recognition directly. Every time your brand is described accurately in text a model has read - a Reddit thread, a review, a YouTube video, an editorial roundup - it reinforces the model's picture of who you are and what you do. A link is a technical signal between pages; a mention is a semantic signal about the brand. Answer engines run on the second kind.

"A link is a technical signal between pages; a mention is a semantic signal about the brand."

What this does not mean

It does not mean links are dead. Backlinks still help pages get discovered, crawled and ranked, and discovery still feeds AI. The point is narrower and more useful: the marginal dollar of off-page effort buys more AI visibility when spent on earned conversations than on link placements. If you were going to spend on a link roundup, the data suggests spending instead on getting talked about accurately in the places models actually read.

How to earn the mentions AI rewards

1. Show up where models read. Communities like Reddit, review platforms, YouTube and industry roundups are heavily cited by AI engines. Being described there, accurately, is worth more than another link on a low-signal domain.
2. Make your description consistent. A model gains confidence when independent sources say the same clear thing about you. Contradictions - a stale category here, a different positioning there - lower that confidence and cost you citations.
3. Feed the entity, not just the page. Publish the plain facts about your brand, and get third parties to repeat them. Corroboration across sources is what turns a name into an entity a model will name.
4. Measure mentions and citations, not just links. Your link tools will not tell you whether AI names you. Track the mentions and the citations directly, and watch which sources drive them.

How a model actually builds your entity

It helps to picture what happens on the model's side. An entity is not a page or a domain. It is a cluster of facts a model has come to associate with a name: your category, your competitors, who you serve, what you are known for. Models assemble that cluster the same way a person would if they only knew you by reputation - by reading many independent descriptions and keeping the parts that agree.

The mechanics behind this are well studied. Research on building knowledge graphs with large language models describes two steps that matter here. The first is entity recognition: pulling your brand name out of raw text and classifying it as an organization. The second is entity linking: deciding that "Stellarcast," "Stellarcast AEO," and "the Stellarcast platform" all point to the same thing rather than three different companies. Both steps get easier and more reliable when many sources describe you in consistent language. A single mention barely moves the needle. A few dozen that all say the same clear thing turn a name into a resolved, confident entity the model is willing to surface.

This is why an unlinked mention can outperform a link. The link tells the model that page A points to page B. The mention tells the model what you are. Answer engines are in the business of the second question, so the signal that answers it directly carries more weight.

A playbook for earning mentions AI rewards

The earlier list covered the principles. Here is the operational version - the actual channels, in rough order of leverage, and what "good" looks like in each.

1. Original data studies. Publish a number nobody else has. A survey, a benchmark, an analysis of your own anonymized usage data. Original research is the single most citable asset in earned media because journalists and models both reach for a concrete stat. BuzzStream's 2025 digital PR survey reported that campaigns built around original research earned roughly 4.7 times more links than campaigns built around company announcements, and the mention footprint follows the same curve.
2. Digital PR and editorial coverage. Pitch reporters with a genuine angle, not a press release. Editorial mentions in credible outlets are disproportionately trusted. Muck Rack's 2026 analysis reported that earned media accounted for the large majority of AI citations while paid and advertorial content barely registered, which tells you where to point the budget.
3. Reviews and third-party listings. G2, Capterra, Reddit threads, "best tools for X" roundups. These are read heavily by AI engines and they describe you in category terms, which is exactly the language that feeds entity recognition.
4. Podcasts and video. A guest spot gets you described in your own words and, once transcribed, indexed as text. When you appear, name your category and your framework out loud. Do not assume the model will infer it.
5. Communities and expert commentary. Answer real questions in the forums your buyers use, without a link and without a pitch. The goal is an accurate description in a place models read, not a placement.

One rule cuts across all five: say the same thing everywhere. Whenever you write a guest post, join a podcast, or answer a question, refer to your category and your product the same way. Consistency is what lets a model link every one of those mentions back to a single, confident entity.

What this does to agency and marketing budgets

If mentions outperform links, the line items have to move. Most off-page budgets were built for a link-first world: a fixed monthly quota of placements, priced per link, measured by domain rating. That structure quietly optimizes for the weakest signal in the Ahrefs data.

The reallocation looks like this in practice:

• Fund research, not just placements. A single strong data study can seed months of accurate, corroborated mentions across dozens of outlets. That is a better use of a quarter's budget than the same spend spread across low-signal link roundups.
• Reprice the retainer. Paying purely per link rewards volume on low-value domains. Tie at least part of the engagement to earned mentions in outlets and communities that models actually read.
• Reclaim the mentions you already earned. A meaningful share of PR placements ship without a hyperlink on first publish. Even unlinked, they feed the entity - so the reclamation work that used to be about chasing the link is now valuable on its own, and worth staffing.
• Change the report. If the monthly deck still leads with backlink count and domain rating, the incentives never move. Put mention volume, source quality, and actual AI citations at the top.

None of this means firing your link team. It means judging their work by whether your brand gets described accurately in the right rooms, rather than by how many links they shipped.

Reading the correlations honestly

It is worth being disciplined about what the numbers can and cannot carry. The Ahrefs study reported correlations - unlinked mentions around 0.664 against AI citations, backlinks around 0.218, with branded anchor text and brand search volume in between. A correlation near 0.66 is strong for messy real-world marketing data, but it is still a correlation.

The honest reading has two parts. First, direction: across the study, the signals traditional SEO ignored line up more tightly with AI citations than the ones it optimized hardest for. That pattern is stable enough to act on. Second, causation: we cannot say from this data alone that adding mentions causes citations. Big, well-known brands get talked about more and get cited more, so some of the correlation is simply brand size showing up in both columns at once. A confounder like that is real and you should assume it explains part of the effect.

What that means for you is modest and useful. Treat mentions as a lever worth pulling, not a guaranteed formula. Earning accurate, corroborated mentions is a good idea regardless of the exact coefficient, because it is also how you build the brand recognition that sits underneath both numbers. Just do not promise a client that a mention count will convert to citations at a fixed rate. The data does not support that, and claiming it undersells the more durable point: you are building an entity, and that pays off in more ways than one metric can show.

The bottom line

The 2026 data reframes off-page work for the AI era. Links earned your pages a vote; mentions earn your brand a place in the answer. The teams that adapt fastest are the ones that stop counting backlinks as the headline metric and start earning - and measuring - the accurate, corroborated mentions that AI engines actually reward.

Related: How to get cited by ChatGPT, Perplexity & Gemini →