Brand Mentions vs. Backlinks: The Ahrefs 75K Brand Study

Why brand mentions are now decisive for AI visibility

The logic of discoverability is shifting. In traditional SEO, the link graph decided for years: whoever had more and better backlinks ranked higher. In AI search systems such as ChatGPT Search, Perplexity, Google AI Overviews and Google AI Mode, that equation no longer holds. A large-scale analysis by Ahrefs provides the clearest evidence for this so far.

Ahrefs says it analyzed 75,000 brands to understand which factors influence brand mentions in AI Overviews. The result is uncomfortable for practitioners: the strongest correlation with AI visibility comes from brand mentions on the open web, not from backlinks. A second analysis across three AI systems showed that YouTube mentions have the strongest correlation with AI visibility at about 0.737, stronger than any other factor tested.

For B2B companies in the DACH region, this is not an academic detail. Use of AI search is growing fast. In Germany, according to Bitkom, 50 percent of internet users aged 16 and over already use AI chats instead of traditional search engines at least occasionally, and 12 percent do so mostly or exclusively. In Austria, a survey by Reppublika and Handelsverband found that around 70 percent of consumers use an AI service, even if only rarely. If you don't show up in these answers, you disappear from a growing part of the customer journey.

The Ahrefs 75K study in detail: methodology and findings

The study is based on Spearman rank correlations between traditional search and brand signals on one side and the frequency of brand mentions in Google AI Overviews on the other. It looked at domains with a Domain Rating above 40, meaning established brands rather than niche websites. The ranking of the correlations clearly favors off-site signals: branded web mentions correlate at 0.664, branded anchors at 0.527 and branded search volume at 0.392, while the number of backlinks reaches only 0.218.

This produces the central finding: brand mentions correlate about three times as strongly with AI visibility as backlinks. A follow-up study by Ahrefs across ChatGPT, AI Mode and AI Overviews, again with 75,000 brands, confirms the picture and adds detail. There, the correlation for branded web mentions is 0.709 for AI Mode, 0.664 for AI Overviews as in the first study, and 0.656 for ChatGPT. Backlinks and URL Rating show only very weak relationships across all systems.

One methodological caveat matters: correlation is not causation. Strong brands get mentioned a lot and appear often in AI answers, and both can share a common cause, such as awareness and relevance. The study does not prove that one additional mention mechanically produces one additional AI mention. It does show very clearly that the traditional backlink signal loses explanatory power in the AI world and that brand signals take its place. For prioritizing resources, that is the information that matters.

Why LLMs understand brands from context rather than links

The 3:1 gap between brand mentions and backlinks can be explained by how large language models work. Traditional search engines have historically assessed authority through the link graph: a link is a countable recommendation from one domain to another. Language models work differently. They learn from text and derive meaning from word context and co-occurrence, that is, from the contexts in which a brand name appears together with certain topics, categories and attributes.

For an LLM, the statement "Vendor X leads in solution Y for industry Z" is a strong signal even if it contains no link. This is exactly where the break with traditional link building lies: mentions without a link, known as unlinked mentions, were largely worthless in traditional SEO. For AI visibility they are a core signal, because the model picks up the brand name and its semantic context regardless of a hyperlink.

This leads to a strategic shift. Instead of rating domains by link potential, what counts is how often, in what context and how consistently a brand is mentioned on the web together with its core topics. Co-citation, meaning being named alongside relevant terms and competitors, becomes more important than any single dofollow link. That changes which measures build visibility in the first place.

YouTube as an underrated GEO channel

The most surprising finding of the follow-up study concerns video. Across all three AI systems, YouTube mentions are the strongest single signal for AI visibility at around 0.737, ahead of the already strong branded web mentions. That is plausible because Google and its AI systems can process YouTube content directly, and transcripts form a rich, topically dense text source in which brands are named in plain text.

For B2B in the DACH region, this does not mean building an entertainment channel. It means systematically anchoring expertise in video:

• Expert formats: Short, focused explainer videos on specialist questions in which the company and product are named explicitly in the spoken word, not just in the title.
• Webinar recordings: Repurpose existing B2B webinars as YouTube videos with a clean transcript.
• Product and application videos: Use cases and implementations that clearly name the industry, problem and solution, so the co-occurrence becomes recognizable to the model.
• Consistent naming: Spell out brand, product and topic terms in every video, since AI systems primarily evaluate the transcript text.

This turns YouTube from a reach channel into a visibility signal for AI answers. The methodological caveat still applies: correlation is no guarantee. But no other channel shows a comparably strong relationship in the follow-up study.

Earned media and digital PR instead of traditional link building

If brand mentions are the decisive signal, the leverage moves from your own domain to credible third-party sources. According to the accompanying press release, a Muck Rack analysis of more than one million AI prompts found that 85.5 percent of AI citations come from earned media, not from brand websites. Your own website is necessary, but far from sufficient to get cited in AI answers.

This calls for a different PR logic than classic link building. The primary goal is a context-rich mention in sources that AI systems trust, rather than a placed link:

• Trade media and industry portals: Articles, interviews and expert commentary in relevant B2B publications that name the brand together with its core topics.
• Accept unlinked mentions: Mentions without a link are no longer a failure. They are a valid signal worth actively seeking out.
• Consistent entity description: Communicate the same clear positioning across all sources so the model builds a stable understanding of the brand.
• Digital PR instead of outreach for links: Place topics and data that get cited, instead of trading links.

For Austrian and German B2B brands, this concretely means being present in the relevant German-language trade media, since AI systems draw on regional sources for regional queries.

Front-loading and answer islands: on-page tactics from GEO research

Earned media builds the brand signal externally. On-page work decides whether your own content is citable at all. The academic basis comes from the Princeton study on Generative Engine Optimization, which shows that GEO can increase visibility in generative answers by up to 40 percent. Content that a language model can extract easily works best.

Two principles are central. Front-loading means putting the answer first: every section opens with a clear, self-contained statement before reasoning and context follow. Answer islands are self-contained text blocks that a model can cite as a whole without needing the rest of the article.

In practice, statistics with a source, concise definitions and well-phrased, fact-dense sentences increase the likelihood of being cited. GEO research identifies exactly these elements as levers: numbers, supported claims and clear language. For your work, that means phrasing every key fact so it fits into an AI answer without rewording.

AI referral traffic and conversion: why the ROI works despite low volume

A common objection to GEO is that AI search brings hardly any measurable traffic. The volume is indeed small, but the value per visit is high. Ahrefs reports for its own website that 0.5 percent of traffic from AI search generated 12.1 percent of all signups and that AI visitors convert about 23 times more often than traditional organic search traffic. A broader analysis by Semrush points in the same direction: the average AI search visitor is 4.4 times as valuable as a traditional organic visit, measured by conversion rate.

The explanation lies in the funnel position. Someone who follows an AI answer has already refined their question and is closer to a decision. For B2B with high contract values, that shifts the math considerably. Even a few highly qualified visits justify investment if the conversion rate is correspondingly high.

Volume is also likely to grow. According to Semrush, over the course of 2025 AI Overviews were triggered for 6.49 percent of queries in January, 24.61 percent in July and 15.69 percent in November, and Semrush expects that AI search could send more visitors to websites than traditional search by early 2028. If you invest in the brand signal early, you build a lead in a growing channel.

Share of Model: measuring the new metric correctly

Traditional rankings fall short for AI visibility. The relevant measure is Share of Model, sometimes also called Share of Citation: how often, how prominently and how positively a brand appears in AI answers, relative to the competition. It is a bundle of metrics rather than a single value.

• Mention frequency: For what share of a defined set of relevant prompts is the brand named at all?
• Share of voice in AI answers: The share of your own mentions compared with the competitors named.
• Cited sources: Which domains are linked as evidence, and is your own brand among them?
• Tone: Is the brand presented neutrally, positively or unfavorably in comparison?
• Platform coverage: In which systems does the brand appear: ChatGPT, Perplexity, AI Overviews or AI Mode?

The scale of the effects shows why this measurement pays off. In the first Ahrefs study, brands in the top quartile of web mentions received around 169 AI Overview mentions on average, more than ten times as many as the next quartile with about 14 mentions. The two bottom quartiles barely appear at all. AI visibility is therefore extremely unevenly distributed, which makes measuring your own position and the gap to competitors all the more important.

Platform differences: not every system weights brands equally

The follow-up study shows that ChatGPT, AI Mode and AI Overviews weight brand signals to different degrees. Across several metrics, including branded web mentions, branded anchors and branded search volume, AI Mode consistently shows stronger correlations with classic brand signals than ChatGPT or AI Overviews. Ahrefs therefore describes AI Mode as a kind of consensus machine that prefers to recommend brands that many people already know and search for.

For prioritization, the takeaway is clear. In systems that rely heavily on established brand signals, awareness and consistent mentions pay off most. Systems with weaker brand coupling tend to leave more room for well-structured, topically strong content, including from lesser-known providers. A sensible GEO strategy therefore follows both tracks in parallel: strengthen the brand signal through earned media and video while building citable on-page content. Which platform matters most for a specific business depends on where your target audience actually searches, and that should be part of the measurement.

Common mistakes

• Keeping the budget primarily in backlinks: The first Ahrefs study shows that backlinks, at 0.218, correlate only weakly with visibility in AI Overviews. Pure link building targets the wrong signal.
• Ignoring unlinked mentions: If you dismiss mentions without a link as worthless, you overlook the central brand signal for language models.
• Ruling out video entirely: YouTube is the strongest single predictor in the follow-up study. Dismissing it as unsuitable for B2B gives away the biggest lever measured.
• Optimizing only your own domain: At 85.5 percent, most AI citations come from earned media. Your own website alone is not enough.
• Measuring GEO by traffic volume: AI traffic is small but high-converting. If you only count sessions, you systematically underestimate the ROI.
• Treating all platforms the same: AI Mode, ChatGPT and AI Overviews weight brand signals differently. A one-size-fits-all tactic gives up the benefit of prioritizing.
• Selling correlation as causation: The numbers show relationships, not guarantees. Serious GEO work communicates this openly.

llms.txt reality check

In GEO circles, the llms.txt file is often touted as a technical solution. It is a proposed standard meant to show AI systems, in a structured Markdown file, which content on a website is especially relevant. The sober finding: llms.txt is a proposal, not a binding standard broadly supported by the major AI providers. The file can offer structure and orientation, but it does not replace a brand signal.

The Ahrefs data make the priorities clear. If brand mentions correlate about three times more strongly with AI visibility than backlinks and most citations come from earned media, then the lever is building mentions, video presence and citable content, not a technical file. Technical GEO files are good housekeeping at best, not the deciding factor. If your resources are limited, invest them in the brand signal first.

Metrics and measurement at a glance

Reliable GEO measurement combines external signals with concrete observation of visibility. A monthly cadence makes sense:

• Brand mention volume and context: How mentions on the web develop, including unlinked ones, and in which topical context they appear.
• Share of Model: Your share of mentions across a fixed set of relevant prompts, broken down by ChatGPT, Perplexity, AI Overviews and AI Mode.
• Cited sources: Which domains AI systems use as evidence, and whether your own brand or your earned media sources are among them.
• AI referral traffic and conversion: Capture AI domains in GA4 via referrers and source channels, and report their conversion rate separately from organic traffic to make the high value per visit visible.
• Video presence: Reach and mentions of your topics on YouTube, since this is the strongest signal measured.

A realistic time horizon is several months. AI systems update their knowledge base with a delay, and brand signals build up slowly through consistent mentions.

Going further

The Ahrefs 75K study shifts the basic assumption of discoverability. For AI search systems, the brand understanding that language models derive from mentions, context and video counts for far more than the traditional link graph. For B2B brands in the DACH region, this results in a clear sequence: first strengthen the brand signal through earned media, trade media and YouTube, then build citable on-page content with front-loading, and finally measure Share of Model and AI referral conversion. Technical GEO files and pure link building remain secondary. If you invest early in a channel where AI visibility is so unevenly distributed, you secure disproportionate opportunities before competitors catch up.