What is co-occurrence, and why do mentions beat backlinks for AI?
Last updated: September 14, 2026
The mechanism (why this isn't just SEO with new words)
A search engine reads the web as a graph of links and passes authority along the edges. A language model doesn't. It reads the web as text and learns which words show up next to which other words. So when a model is asked "best CRM for a startup," it reaches for the names that appeared, over and over, near "CRM" and "startup" in what it was trained on and what it retrieves. Whether those mentions carried a hyperlink is close to irrelevant. The proximity is the signal.
That's co-occurrence. Your brand next to your topic, repeatedly, across many independent sources. It is the single most load-bearing idea in getting found by AI, and it is why the old backlink playbook underperforms here.
The data, because this is a claim people resist
Ahrefs ran Spearman correlations between dozens of brand signals and AI visibility across 75,000 brands, on three platforms. The strongest signals are all off-site brand presence. Link authority sits well below.
| Signal | ChatGPT | AI Mode | AI Overviews |
|---|---|---|---|
| YouTube mentions | 0.737 | 0.740 | 0.712 |
| Branded web mentions | 0.664 | 0.709 | 0.656 |
| Branded anchors | 0.511 | 0.628 | 0.527 |
| Branded search volume | 0.352 | 0.466 | 0.392 |
| Domain rating | 0.266 | 0.285 | 0.326 |
Read the bottom row. Domain rating, the number an entire industry spent twenty years and billions of dollars chasing, sits at a third the strength of a plain mention. Ahrefs shows raw backlink count only as a chart with no exact figure published, so we won't quote one, but it sits in the same weak band. The study is here.
Two studies, two methods, one answer
A correlation on its own invites the "so what" objection. What makes this solid is that a second team, asking a different question with a different method, got the same result.
How we got here
Ahrefs, December 2025: asks which brand signals predict AI visibility? Method: correlate signals against visibility across 75,000 brands. Answer: off-site mentions (0.66 to 0.71), not link authority (0.27 to 0.33).
Muck Rack, July 2025 to May 2026: asks where do AI citations actually come from? Method: classify 25 million links cited by ChatGPT, Claude and Gemini across 17 industries. Answer: 84% earned media, 27% journalism specifically, 0.3% paid or advertorial, and the split held between 82% and 89% across three editions, which Muck Rack reads as systematic sourcing behavior rather than noise.
Put together: one study says mentions predict being cited. The other says citations are made of earned mentions. Prediction from one side, composition from the other, and they close the loop on each other. Neither says "co-occurrence is the currency" outright. Both, read together, leave no other explanation standing. The one caveat worth stating: both are vendor studies, Ahrefs sells a brand-monitoring product and Muck Rack sells PR software, so each has a reason to like this result. That is exactly why it matters that they agree using methods that don't share a weakness.
The strongest single signal is one most GEO advice ignores
Look at the top row of that table again. YouTube mentions out-correlate everything, on every platform, at 0.71 to 0.74. Almost all generative-engine-optimization advice is about text: schema, answer-first paragraphs, listicle placements. All of that is real. But the single strongest predictor Ahrefs found is your brand being named in video, and nobody is building for it. If you want an edge that the crowd hasn't crowded yet, it is getting talked about on camera.
How you actually build co-occurrence
You get described, by other sites, near your topic, again and again. Concretely: getting into the roundups and "best X" listicles AI pulls from, getting quoted in articles, showing up in forum threads and Reddit answers, earning press mentions, and, per the row above, getting named in video. The link is a bonus. The mention next to your category is the point. One clean mention on a topically relevant page does more for your AI visibility than a dozen backlinks from off-topic domains.
This also explains a thing that drives people crazy: a competitor with worse SEO getting recommended by ChatGPT while you don't. They aren't winning on links. They got described near the topic more consistently than you did.
The honest disagreement
Does this mean backlinks are dead? The camps split, and it decides where you spend money:
School A: links still rule, mentions are a proxy for size
Correlation isn't causation. Big brands have more mentions AND more links, so mentions may just track being big. Keep building links; they still drive the Google ranking that partly feeds AI Overviews.
School B: mentions are the real currency now
The mechanism backs the data: models read text, not link graphs. A 0.66-versus-0.27 gap is too wide to be a size artifact, and Muck Rack's composition data confirms it from the other direction.
We're School B, and the "proxy for size" objection is the weakest part of School A's case, because the study already contains the size proxy. Domain rating is the classic measure of how big and established a site is, and it sits at 0.27 to 0.33. If size explained the mention correlation, DR would track mentions. It doesn't. The one thing School A gets right: on Google's AI Overviews specifically, ranking still partly carries over, so don't torch your SEO. Build both. But with one dollar and a new brand, spend it on the mention.
Related entries
- Explains Why is my brand invisible to AI even though I rank on Google?
- Explains How consistent are AI recommendations, really?
- Conflicts with Do backlinks still matter for AI visibility?
- Prerequisite Retrieval vs training: which one decides whether AI mentions you?
- Related How do I get mentioned on the third-party sites AI reads, without a backlink?
Changelog
September 14, 2026: Rewritten on primary sources only. Replaced a secondary-sourced backlink figure with Ahrefs' published domain-rating values; added the per-platform table, the Muck Rack cross-read, and the video finding. First published earlier the same day.
Sources (primary): Ahrefs, "Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews", 75,000 brands, Spearman correlations, December 12, 2025; Muck Rack, "What Is AI Reading?", 25M+ links from ChatGPT, Claude and Gemini across 17 industries, editions July 2025 to May 2026. Both are vendor studies; incentives noted above. AI behavior evolves, so this entry is dated and revisited.