Brand Mentions vs Backlinks for AI Search: What the Studies Actually Show

One correlation, many pitch decks

The pitch goes like this. Backlinks are the old currency. Mentions are the new one. AI engines read the whole web, so every time your name appears on someone else's page, you get a little closer to being recommended. Buy our mention package and watch it happen. It sounds modern and it borrows its authority from a genuine piece of research.

The research deserves better than the pitch. Two studies get cited most often, one from Ahrefs and one from Seer Interactive, and both are careful about their limits. Read in full, they support a sensible idea: businesses that are widely discussed tend to show up in AI answers, and link counts alone explain little. They do not support the leap from there to buying mentions by the dozen. This page walks through both, then sorts what is established from what is plausible, the same discipline we apply on the GEO section front page.

The Ahrefs study, all the way to the caveats

In May 2025 Ahrefs compared 75,000 brands against how often each one appeared in Google AI Overviews. It used a Spearman correlation, which asks whether brands that score high on one measure tend to score high on another. Branded web mentions came out on top. Backlinks came out near the bottom. The full ranking, from the Ahrefs article, is below.

The spread between brands was steep. Brands in the top quarter for web mentions averaged 169 AI Overview mentions; the next quarter averaged 14. About 26% of all brands studied had none. Ahrefs said plainly that correlation is not causation and described the correlations as ranging from moderate to very weak, meaning several signals combine. Note the scope too: this was AI Overviews only, not ChatGPT or Perplexity, and the sample was 75,000 brands, so a two-truck plumber on the South Side should expect the pattern, not the exact numbers.

The Seer study: rankings mattered, link counts barely did

Seer Interactive published its study in January 2025. It tracked more than 300,000 keywords in finance and SaaS, distilled People Also Ask questions down to 10,000, put them to OpenAI's GPT-4o, and recorded which brands each answer named. Then it compared those brands against their search performance.

• Page-one Google rankings correlated about 0.65 with brand mentions in the answers.
• Bing rankings correlated about 0.5 to 0.6.
• Backlinks showed a weak or neutral relationship.
• Filtering out forums, social media and aggregators made the correlations noticeably stronger for solution-oriented sites.
• Seer called it phase one, limited to search factors, and noted that matching brand mentions to domains was hard.

Two industries, one model, one point in time. The finding lines up with what engines say about themselves: ChatGPT search draws on third-party search providers, and Google's AI features are grounded in its core ranking systems. A page that ranks gets retrieved; a retrieved page can be quoted. Our GEO vs SEO page covers that overlap.

Correlation, cause, and the third variable behind both

Here is the trap. A brand that people talk about tends to have lots of mentions, lots of searches, plenty of links and a strong presence in AI answers, all at once, because all four flow from the same source: it is well known and does something people discuss. A large regional bank has more of every measure than a two-truck landscaper. Measuring mentions against AI visibility across 75,000 brands will find a strong relationship even if mentions themselves cause nothing.

Cause can run backward too. A brand that keeps appearing in answers may get more searches and more write-ups because of it. None of this means mentions are useless. It means a study of correlations cannot tell you that paying for 50 new mentions will move your number, and neither study claims it can.

Links still count, through a side door

Reading the Ahrefs table as links are dead misses two of its rows. Branded anchors, links whose text is the business name, ranked second at 0.527. That kind of link is a mention and a link at once. And the Seer finding that rankings track AI mentions puts links back in the story indirectly, because links still help pages rank in ordinary search, and ranking pages are what the engines retrieve.

So the practical reading is not mentions instead of links. It is that a link from a page nobody reads, with anchor text nobody would write, does little for AI answers, while a story in the local paper that names you, describes what you do and links to you does several jobs at once. Odysseus, the backlink agent in Athena CMS, looks for exactly that kind of prospect: real sites with a reason to mention the business. Ordinary SEO and solid internal linking remain the floor everything else stands on.

Where honest mentions come from for a local business

For local searches the mix looks different from the brand studies. BrightLocal's November 2024 study of ChatGPT local searches found third-party mentions made up 27% of sources, behind business websites at 58% and ahead of directories at 15%; within those mentions, Wikipedia accounted for 39%, according to the study. Most small businesses will never have a Wikipedia article, so the rest of the list matters more.

A March 2026 Peec AI analysis of 30 million citations ranked Reddit as the most-cited domain across five AI engines. Peec sells AI visibility tracking, so read it as a vendor's data, but it fits the pattern: engines lean on places where people talk. Our post on why Reddit is the new Yelp has the local angle.

A mention plan for a business without a PR department

You do not need an agency press list. You need a short list of places your customers already read, a reason for each one to mention you, and the patience to do it monthly. Start by seeing whether you are named at all.

1. 1

   Measure first

   Run your customers' questions through the free AI visibility check and write down which competitors get named. Then search those competitors' names to see where they are mentioned that you are not.

2. 2

   Fix the record you already own

   One name, one phone, one description across Business Profile, Apple Business, Bing Places and directories, so new mentions attach to the right entity.

3. 3

   List ten places that should know you

   The chamber, two trade associations, your suppliers' locators, the neighborhood paper, a local podcast, a school or charity you already support.

4. 4

   Give each one a reason

   A project worth photographing, data from your own jobs, a sponsorship, an expert answer to a seasonal question. Mentions follow news.

5. 5

   Track what moves

   Recheck the same questions monthly. AI share of voice explains how to count, and how AI chooses businesses explains why changes take time.

Our programs include this work, alongside the site, listings and reviews, and track the answers with AICiteBeam. Programs start at $625 a month, set after a free GEO audit; GEO pricing explains what pushes a quote up or down, and financing is available.