The honest answer is that mentions correlate strongly with AI visibility, but nobody has proven they cause it. Ahrefs studied 75,000 brands and found branded web mentions had the strongest link to AI Overview visibility of any factor they measured. That is a signal worth acting on, not a law.
This question comes up in almost every kickoff call we run now. A founder has spent three years buying links. Then a competitor with a weaker backlink profile starts showing up inside ChatGPT and Google AI Overviews, and the founder wants to know what changed.
We think the honest version of this debate is more useful than the confident version. So here is what the data says, what it does not say, and how we would spend a budget in 2026.
An unlinked brand mention is any time your company name appears in text on someone else's site without a clickable link back to you. A podcast show notes page that names you. A Reddit thread. A roundup post. A press quote. The words are there. The hyperlink is not.
Traditional SEO tools mostly ignored these. A mention with no link passes no PageRank, so it did not move rankings in a way anyone could measure. Tools like Ahrefs and Semrush were built to count referring domains, because that was the currency.
Language models do not read the web that way. They train on text and they retrieve text. A sentence that says your product is the best option for mid market teams is useful to a model whether or not it carries an anchor tag. The link was never the part the model needed.
Because a model builds its picture of your brand from language, not from graph structure. When a model decides which three tools to name in an answer, it is drawing on how often and how confidently those names appeared next to the concepts in the question. Links are one signal among many.
Think about how you would answer a question about project management tools from memory. You would name the products you have read about most often in that context. You would not consult a link graph. Retrieval augmented systems work in a related way, pulling passages that match the question.
This is also why entity SEO has become more useful than keyword density. The model is trying to resolve your brand into a known thing with known attributes. Consistent naming across many sources helps it do that.
It proves a correlation, not a cause. In a study of 75,000 brands published by Louise Linehan at Ahrefs, branded web mentions scored a Spearman correlation of 0.664 with AI Overview visibility. Branded anchors scored 0.527 and branded search volume 0.392. Backlinks scored 0.218.
Domain Rating came in at 0.326 and referring domains at 0.295. So the classic link metrics were not zero. They were just well behind the mention metrics. The authors were careful about this. They wrote plainly that correlation does not equal causation, and they noted that all the factors showed moderate to weak correlations on the Spearman scale.
We read that caveat as the most important line in the study. Big brands get more mentions and more AI visibility because they are big brands. Some of the correlation is just size. The finding still shifts where we would place a marginal dollar, and that is enough to change a plan.
Yes, and treating this as a replacement is the common mistake. Links still drive classic organic rankings, and classic rankings still feed AI answers. Most citations in AI answers come from pages that already rank well, so the link work you did is still holding up the floor under everything else.
The better framing is that links and mentions do different jobs. A link tells a crawler this page is worth visiting and trusting. A mention tells a model this name belongs in this conversation. You want both, and a good campaign usually produces both at once.
We wrote more about where link building still earns its keep in our piece on whether backlinks still matter. The short version is that we have not cut link work for any client. We have stopped treating link count as the headline number.
They come from places where people talk, not places where marketers publish. Community threads, podcast transcripts, video descriptions, review sites, industry newsletters, conference recaps, and news coverage. Almost none of these are things you can buy directly, which is why they are hard.
Review platforms such as G2 and Capterra generate a steady stream of your name next to category words. Reddit threads do the same with far more skepticism attached. YouTube descriptions and transcripts put your name into text that models can read even though the content is video.
In our work the pattern is consistent. Studios and software companies that show up in AI answers are usually the ones with a real presence in a community, not the ones with the tidiest content calendar. That is uncomfortable if you were hoping for a mechanical fix.
You publish things other people need to reference. Original data, a clear framework with a name, a tool that solves one annoying problem, or a strong position on a live debate. Each of these gives a writer a reason to type your company name into their own sentence.
Original numbers are the strongest of these. When Ahrefs published a correlation study across 75,000 brands, every SEO blog on the internet had to name Ahrefs to discuss it. That is the mechanic. The research became the citation, and the citation carried the brand.
Being quotable is the second lever. If you say something specific and slightly contrarian, people quote you when they agree and when they do not. Vague, agreeable content gets read and forgotten. We would rather publish one sharp opinion than four safe summaries.
You measure appearance rate in answers, not mention count. Pick twenty questions a buyer would actually ask, run them across ChatGPT, Perplexity, Claude, and Google AI Mode on a schedule, and record how often your brand appears and how often it is cited. The trend line is the metric.
Raw mention volume is a vanity number in the same way raw backlink count was. What you want to know is whether the mentions are landing in the contexts that matter. Two hundred mentions in unrelated threads are worth less than fifteen in the exact category conversation you sell into.
We go deeper on the tracking side in our guide to measuring AI search traffic. The part worth repeating here is that you need a baseline before you change anything, or you will not be able to tell whether the change worked.
Stop paying for links on sites nobody reads. A link from a site with no audience produces no mentions, no traffic, and no conversation. It was a weak tactic before AI search and it is a weaker one now, because the model has no reason to trust text nobody else repeats.
Also stop treating traffic as the only scoreboard. Pew Research Center analysed the browsing data of 900 United States adults across 68,879 Google searches in March 2025. On pages with an AI summary, users clicked a traditional result in 8 percent of visits. Without a summary, they clicked in 15 percent.
Clicks inside the summary itself happened in just 1 percent of visits. Users also ended their session on 26 percent of pages with an AI summary, against 16 percent without one. If you judge your content only by sessions, you will underrate work that is doing its job.
Being nameable will matter more than being rankable. The academic work points the same way. The GEO paper by Aggarwal and colleagues, accepted to KDD 2024, found that content changes aimed at generative engines could boost visibility by up to 40 percent in generative engine responses. Presentation of the text mattered.
So the plan we would run is simple. Keep the technical and link foundations healthy, publish things worth citing, get into the rooms where your category is discussed, and track appearance rate in answers as a first class metric. That is a slower plan than buying links, and it compounds better.
If you want a second opinion on where your brand actually stands in AI answers today, we are happy to run the check and walk you through what we find. You can reach our team at phoenix.studio and we will tell you straight whether the gap is content, mentions, or something in the build.
Tell us where you want to go. We'll tell you how we'd get you there.