Brand Mentions vs Backlinks: AI Citation Study (Ahrefs 2025)

TL;DR

Brand mentions correlate ~3x stronger with AI citation than backlinks. Ahrefs measured 0.664 correlation for brand web mentions vs 0.218 for referring domains across 75,000 brands.
The top three correlations are all off-site brand signals: web mentions (0.664), brand anchors (0.527), and brand search volume (0.392).
A follow-up study extended the analysis to ChatGPT and Google AI Mode. YouTube mentions showed the single strongest individual signal — ~0.737 correlation.
Per Muck Rack's December 2025 analysis, 94% of AI citations come from non-paid, non-brand-owned sources. Brand-owned content barely registers in the citation set.
The action plan is not "build more links." It is "get talked about in places AI models trust."

The 75K-brand study, summarized

In August 2025, Ahrefs published an analysis of brand visibility factors across 75,000 brands in Google AI Overviews. The methodology: measure brand-level features (mentions, anchors, search volume, traffic, referring domains) and correlate them against whether the brand surfaced in the cited set of an AI Overview block.

The headline finding was clean: brand web mentions show a 0.664 Spearman correlation with AI Overview brand visibility. Referring domains — the classical backlink metric — correlated at 0.218. That is roughly 3x stronger for mentions than for backlinks. The other top correlates — brand anchors (0.527) and brand search volume (0.392) — are also off-site brand signals, not link signals.

A December 2025 follow-up extended the same correlation work to ChatGPT and Google AI Mode. The pattern held and, in places, intensified. YouTube brand mentions correlated at ~0.737 — the single strongest individual signal Ahrefs measured across all platforms.

Methodology and what to take seriously

A few caveats worth surfacing before you re-allocate your budget:

Correlation is not causation. Brands that get mentioned in editorial media also tend to be brands that have invested in PR, product quality, and earned coverage. The mentions are a proxy for "this is a real brand," not the lever itself.
The study measures presence in the cited set, not citation count. A brand that appears in 1 of 10 AI Overviews is treated similarly to one that appears in 1 of 10. The study is good at "is this brand visible at all" and less good at "in what depth."
It is one study. Replication matters. SEJ's commentary flagged that the result is striking but consistent with what practitioners had been seeing anecdotally for 18 months prior.

Even granting those caveats, the gap between 0.664 and 0.218 is large enough that the practical implication holds: if you are doing one or the other, mentions win.

Why mentions matter more than links for LLMs

The intuition is structural. A backlink is one site's editorial choice to point at another site. To an LLM, a link is a hyperlink — a piece of HTML markup. The model does not "follow" the link the way a crawler does; it reads the surrounding text.

A brand mention, by contrast, is a piece of textual evidence sitting inside the corpus the model was trained on (or grounded in). When a model sees "Asana, Notion, and ClickUp" repeatedly in editorial coverage of project management software, it learns that those are the three known entities in that category. The next time someone asks for project management tools, those three are the candidates in the model's head.

Backlinks influence whose pages get crawled and which ones rank. Brand mentions influence which entities the model treats as known and citation-worthy. Different mechanism, different lever.

Where mentions count most

Per the follow-up analysis and broader practitioner reporting, the platforms that move the needle for AI visibility cluster around five surfaces:

YouTube. The single strongest correlate in Ahrefs' data. Transcripts of widely-watched videos make their way into training corpora and grounding sets. Sponsored mentions in established channels show up.
Reddit. Heavily over-represented in the citation sets of ChatGPT and Perplexity. Authentic discussion in subreddits relevant to your category matters more than thread count.
Wikipedia. When a brand has a Wikipedia entry, the model treats it as a confirmed entity. The entry does not need to be glowing — it needs to exist and be neutral.
LinkedIn. Editor-published articles and posts from credentialed practitioners. Heavier weight for B2B categories.
Established trade publications. TechCrunch, The Verge, Search Engine Land, industry-specific magazines. A single substantive article in a trusted outlet outweighs a hundred listicle inclusions.

The pattern: mentions on platforms where editorial judgment or community moderation already filtered for quality.

Action plan

If you are deciding where to spend the next quarter of marketing budget, the data argues for this rough order:

Earn one substantive trade-publication article per quarter. Pitch a real story. A 1,200-word piece with quotes from your team in a category-relevant publication beats a hundred guest posts.
Get on the right podcasts. Particularly podcasts with YouTube distribution. The transcripts and the video metadata both compound.
Be present in Reddit and category-specific communities. Not promotional. Useful. Answer hard questions in your domain under your real account.
Get a Wikipedia entry if you qualify. This is harder than it looks — Wikipedia notability criteria are real. Most B2B brands do not qualify; the ones that do, should.
Then, if there is budget left, do classic link building. It still matters. It just matters less than the above.

The order is the point. Most teams have it inverted.

Limits of the study

Worth keeping in your head:

English-language web. The study is dominated by English-language brands and English-language AI surfaces. Coverage in other languages is thinner and the pattern may differ.
Brand visibility, not revenue. Being cited by AI is not the same as being purchased. The downstream conversion question is unsolved.
The metrics will shift. As AI engines mature, the relative weight of each signal will change. The current 3x gap is a 2025 snapshot, not a permanent law.

That said, the structural argument — that LLMs reason over entities, and mentions build entities faster than links — is unlikely to invert. The numbers may move; the direction probably stays.

FAQ

Should I stop building backlinks?

No. Backlinks still matter for traditional Google rankings, which still drive a meaningful share of traffic and still feed into the candidate sets that AI engines retrieve from. The argument is about marginal budget: if you are choosing between a $10K link campaign and a $10K trade-publication PR push, the PR push wins on AI-citation expected value.

How many mentions do I need before I start seeing AI citations?

There is no clean threshold in the public data. Practitioner reports converge around "consistent mentions across 5+ trusted surfaces over 60–120 days" before AI citation share moves meaningfully. Faster for narrow categories; slower for crowded ones.

Can I buy mentions?

You can buy advertising and sponsored content, and some of those mentions get indexed. But the Ahrefs data and the Muck Rack analysis both point at earned media as the dominant signal. Paid placements that read as paid placements appear to be discounted.

Sources

Want to know which brand mentions are actually working for you in AI surfaces? CiteFlow tracks mention share, citation share, and the off-site authority signals that move AI visibility.

Brand Mentions vs Backlinks: What Actually Drives AI Citation (Ahrefs 2025 Study)