# Brand Mentions on Third-Party Sites: Why Mentions Carry More Weight Than Backlinks > In an Ahrefs study of 75,000 brands, branded web mentions correlate with being named in Google's AI Overviews at r = 0.664 — backlinks at only 0.218. What counts is the mention itself, not the link. Ahrefs points out explicitly that correlation does not prove causation and that all measured values are moderate to weak on the Spearman scale. Source: https://www.codaai.ai/en/blog/brand-mentions-third-party-sites-ai/ Published: 2026-09-17 Updated: 2026-09-22 May be quoted with attribution and a link to the source. --- A supplier of medical technology has invested in link building for years. Its Domain Rating is respectable, and so are its rankings. Even so, the company does not appear in ChatGPT or in Google AI Overviews when a buyer asks for suppliers in its category. A considerably smaller competitor does appear — one that two trade publications write about regularly. The contradiction is only an apparent one. It disappears as soon as you know what these systems respond to.
0.664 rank correlation between branded web mentions and being named in Google's AI Overviews — the strongest single factor measured Ahrefs, 75,000 brands, 26.05.2025
0.218 rank correlation for backlinks — roughly a third of the value for mentions same study
26% of the brands examined had zero mentions in AI Overviews same study
169 : 14 median mentions: top quarter by web mentions against the next quarter same study
**Figure 1: The three strongest factors lie outside your own website** Branded web mentions correlate with being named in Google's AI Overviews at 0.664, backlinks at only 0.218; in between are brand name in anchor text, branded search volume and Domain Rating. - Branded web mentions: 0.664 - Brand name in anchor text: 0.527 - Branded search volume: 0.392 - Domain Rating: 0.326 - Backlinks: 0.218 Source: Ahrefs, 75,000 brands, 26.05.2025 – Spearman rank correlation with being named in Google's AI Overviews This building block is the first in the second visibility tier — **[Tier 2 · Recommended](https://www.codaai.ai/en/digital-visibility/#sources)**. It describes the step from “The AI can read us” to “The AI names us because others name us”. ## What Ahrefs measured In May 2025, Louise Linehan and data scientist Xibeijia Guan [examined 75,000 brands](https://ahrefs.com/blog/ai-overview-brand-correlation/) to see which factors are related to their being named in Google's AI Overviews. They used Spearman rank correlation; higher values indicate a stronger relationship. The full ranking: | Factor | Correlation | |---|---| | Branded web mentions | **0.664** | | Brand name in anchor text | 0.527 | | Branded search volume | 0.392 | | Domain Rating | 0.326 | | Referring domains | 0.295 | | Branded traffic | 0.274 | | Backlinks | **0.218** | | Ad traffic | 0.216 | | Ad cost | 0.215 | | URL Rating | 0.18 | | Number of pages | 0.17 | Two things stand out. First: **the three strongest factors all lie outside your own website.** Mentions, anchor texts, branded search volume — none of these can be produced on your own site. Second: paid visibility contributes little. Ad traffic and ad cost come in at 0.216 and 0.215, in the bottom third. ### Why the mention counts without a link The reason lies in how the models work. Ryan Law, head of content at Ahrefs, puts it this way in the study: language models derive their understanding of a brand's authority from the words on the page — from how often certain terms occur, how different terms occur together and the context in which they are used. For a classic search algorithm, a link is a vote that can be counted. For a language model, a sentence such as “Supplier X builds test rigs for drivetrains” is a **statement about an entity**, and whether a hyperlink is attached to the company name changes nothing about that statement. That is why the unlinked mention has an effect in AI search, while in classic SEO it remained largely without effect. ### The figure that shows the order of magnitude The correlation on its own is abstract. It becomes more concrete when the brands are split into quarters. Ahrefs sorted the brands into four groups by how often they are mentioned on the web and counted how often each was named in AI Overviews. The top quarter had a median of **169 mentions**. The quarter below it had **14**. The bottom two quarters had zero to three. So the jump is not at the top end but in the middle — and it is steep. Ahrefs states the consequence bluntly: brands in the bottom half by web mentions are practically invisible to these systems. A second figure from the same study fits this: **26 per cent of the 75,000 brands had no mention at all** in AI Overviews. Roughly a quarter of established brands — the sample included only domains with a Domain Rating above 40 — simply do not appear in this layer of answers. ## What the figure does not say Discipline is needed here, because this figure is currently being quoted a great deal, and often wrongly. **Correlation is not causation.** Ahrefs writes this into the text itself, and it is not a formality. It is just as plausible that well-known brands are both mentioned more often and appear more often in AI answers, without one causing the other. Anyone who buys 500 mentions does not automatically buy mentions in AI Overviews. **The values are not high.** Ahrefs adds that all the factors examined show “moderate to very weak” relationships on the Spearman scale. 0.664 is the highest value in the field — but it is not a strong relationship in the strict statistical sense. **The sample consists of large brands.** The study selected domains with a Domain Rating above 40 and, for each domain, looked at the keyword with the highest search volume, at least 800 searches a month. A mid-sized special-purpose machinery manufacturer with search volume in the hundreds is not represented in this dataset. Applying the finding to such a company is a reasoned assumption, not a measurement. **Only Google was measured.** For ChatGPT there is an independent study by Seer Interactive that shows the same ranking, but at a considerably lower level: Domain Rating 0.25, backlinks 0.10. The direction repeats; the values are not transferable. We state these limitations for a simple reason: a figure passed on unchecked is exactly the kind of evidence that is currently harming this field. ## Four routes to mentions that carry weight The finding points to a direction of work, not a shortcut. Four approaches, sorted by effort. ### 1. Check the mentions that already exist Almost every company is mentioned more often than it thinks — only often incorrectly. An old company name, an outdated product name, a category assignment from before the last repositioning. For a language model, this is not a cosmetic flaw. If the same company appears on five pages under three names and in two categories, the evidence falls apart. Instead of one brand with five pieces of evidence, the system sees three weakly supported entries. **Brand consistency is therefore not a corporate design issue but a visibility issue.** The first step is accordingly unspectacular: search for the company name in quotation marks, go through the hits on other domains and have the incorrect ones corrected. It costs little and has an immediate effect on the underlying data. ### 2. Contribute expertise where people are already writing The second-cheapest route runs through the trade media in your sector, which constantly need material. What they need is not a press release but a verifiable statement from someone responsible for the subject — on a standard, a process, a market development. The side effect is exactly what was measured: the company name is in the text, in the right context, next to the right technical terms. Whether the editors add a link is secondary for this building block. ### 3. Produce your own data that others write about The most effective and most demanding route: measure something that does not yet exist. A survey in your own sector generates mentions for years because it can be quoted — and it generates them in exactly the form language models respond to, namely as a statement with an author. This is not theory but the experience behind our own [study on AI visibility among mid-sized companies in Germany](https://www.codaai.ai/en/study/): 449 companies, 3,592 buyer questions, 7,184 documented AI answers. As a building block, this route sits in the third tier because it needs lead time. ### 4. What we deliberately do not do There is a quick route to lots of mentions, and we do not take it. Planting mentions in forums, appearing in communities under someone else's name or buying mentions in bulk produces numbers without substance — and in forums, regularly a ban. A finding from another study fits here: for niche B2B purchase prompts, Reddit accounted for just **1.4 per cent** of citations ([Overthink Group, July 2026](https://overthinkgroup.com/b2b-ai-citation-stats-2026-q2/)). In this segment, the effort needed for a forum presence bears no relation to that share. What we consider worthwhile instead is set out under [“Six things we deliberately do not offer”](https://www.codaai.ai/en/digital-visibility/#not-included). ## Not every mention carries the same weight The study counts mentions without weighting them. In practice this difference is considerable, and it explains why two companies with a similar number of hits are not equally visible. **Context plays a part.** “Supplier X has won an award” and “Supplier X builds test rigs for drivetrains in line with ISO 1940-1” are the same to a count. For a language model, only the second sentence is usable, because only it connects the company name with terms someone would ask about. Ahrefs gives exactly this as its explanation: models derive their understanding from terms occurring together. **The source plays a part.** A mention in a trade publication that is itself cited has a different effect from one in a directory nobody reads. This cannot be quantified precisely, but it follows from the same mechanism that favours pages on strong domains. **Spread plays a part.** Thirty mentions on three domains are weaker than thirty on fifteen. A brand that is written about independently in many places is a more stable finding for a system than one that appears three times in the same publication. **Tone plays a part.** A mention in a list of insolvencies counts just as much in a frequency measurement as one in a list of top performers. That is a well-known blind spot of such correlation studies and one reason not to treat the figure as a target metric. **Figure 2: The same mention – two effects** A mention such as “Supplier X has won an award” connects the name with no term people ask about; “Supplier X builds test rigs for drivetrains in line with ISO 1940-1” connects name, category and standard. - Counts, but carries nothing: “Supplier X has won an award.” - Carries a statement: “Supplier X builds test rigs for drivetrains in line with ISO 1940-1.” Source: Own illustration based on Ahrefs (2025): models derive authority from terms occurring together ### How to determine your starting point yourself For an initial assessment you do not need a tool. Four steps, a good hour: 1. **Count.** Search for the company name in quotation marks, plus the common spellings — with and without the legal form, with and without a hyphen. Note the hits on other domains. 2. **Sort.** Trade media, directories, comparison sites, forums, competitor sites. The distribution tells you more than the total. 3. **Check.** Are the company name, the category and the description of services correct in every hit? The most common finding is not a missing mention but outdated information. 4. **Compare.** Do the same for two competitors. Only the gap makes your own figure readable — 40 mentions are a lot or a little, depending on where the others stand. What this manual count does not give you is the connection to actually being named in AI answers. Whether a system names you for a specific buyer question depends on more than the number of mentions — and is a measurement in its own right. ## How this building block connects to the others Brand mentions are the first building block of [Tier 2](https://www.codaai.ai/en/digital-visibility/#sources) and the foundation for the ones that follow. The building block right next to it — [comparison articles and vendor lists](https://www.codaai.ai/en/blog/comparison-articles-vendor-lists-ai/) — is essentially a special case: a mention at the point where someone is about to make a decision. How both interact with readability and citable content when a buyer asks ChatGPT for suppliers is shown in the overview [Getting recommended by ChatGPT as a supplier](https://www.codaai.ai/en/blog/get-recommended-by-chatgpt-as-a-supplier/). The order still matters. Mentions have an effect when an AI system can check the evidence it finds. If your website is not accessible to these systems' crawlers, the confirmation is missing — and the mention stands without backing. That is why [Tier 1 · Found](https://www.codaai.ai/en/digital-visibility/#readable) comes first, even though Tier 2 carries the more interesting figure. The terms behind this are explained in the [GEO glossary](https://www.codaai.ai/en/knowledge/geo-glossary/), including [entity](https://www.codaai.ai/en/knowledge/geo-glossary/entity/), [brand mentions](https://www.codaai.ai/en/knowledge/geo-glossary/brand-mentions/) and [grounding](https://www.codaai.ai/en/knowledge/geo-glossary/grounding/). ## The honest conclusion The sentence that sticks from this study is not ours but Ahrefs': visibility begets visibility. Those who are already being discussed continue to be discussed — and those who do not appear do not appear in the AI answer either. That is uncomfortable, because it suggests no shortcut. But it is also the message a mid-sized company can use: the lever is not a technical trick but being visible for your expertise — and putting the mentions that already exist in order. Where your company stands today can be measured. That is exactly what the Digital Visibility Audit does: it checks against real buyer questions whether ChatGPT and Google AI Overviews name you — and which sources they draw on instead. Where ChatGPT finds its evidence in the first place is explained in [why Bing matters](https://www.codaai.ai/en/blog/chatgpt-seo-perplexity-visibility/); how to measure your own mentions in the [analysis of 7,184 AI answers](https://www.codaai.ai/en/blog/ai-visibility-chatgpt-recommendation-practice/). Another source companies can fill themselves is [YouTube](https://www.codaai.ai/en/blog/youtube-ai-visibility-b2b/).