AI engines recommend brands they can retrieve and corroborate. The brand needs to appear in the source set an engine pulls for that specific question, with a clear passage the model can use. It also needs third-party evidence behind it, including reviews, expert coverage, and mentions on authoritative sites. Correlation studies link authority and brand mention volume to AI visibility far more strongly than content volume.
Google has said its AI Overviews and AI Mode may fan out into multiple searches across subtopics and sources before composing an answer (Google AI features documentation). If your page is not in the candidate pool for those subqueries, nothing about its quality gets a vote. This is why crawl access, indexing, canonicalization, and text that exists outside of JavaScript are the unglamorous first half of the job.
Once you're in the pool, the model has to find a passage it can use as evidence for naming you. When we read the source passages an engine actually pulled on a recommendation prompt, they tend to be short stretches of text where a claim and the proof for that claim sit within a few sentences of each other, close enough that the model can carry both into its answer without stitching them together from different parts of the page. The practical version of this on our own builds is that any assertion we want cited has to travel with its number, so when the supporting figure is sitting in a conclusion at the bottom of a page, we move it up next to the claim it supports.
Two large studies have tested which brand signals track with AI visibility.
| Signal | Correlation with AI visibility | Source |
|---|---|---|
| YouTube mentions | \~0.737 | Ahrefs, 75,000 brands |
| Branded web mentions | 0.66 to 0.71 | Ahrefs |
| Backlink authority | 0.65 | Semrush, 1,000 domains |
| Site page count | \~0.194 | Ahrefs |
Semrush also found that link quality mattered more than raw link volume in its sample. We end up quoting the page-count number in a lot of planning meetings, because "publish more" is the reflex that budget defaults to and 0.194 is a thin foundation to build a year of headcount on.
These are correlations rather than causal proof, and the distinction matters here. A brand with heavy YouTube coverage is usually a well-known brand already, and well-known brands get recommended for reasons that have little to do with the videos themselves. Where the numbers are genuinely useful is in what they argue against, since a 0.194 correlation on page count is weak evidence for the publish-more theory that still drives most content budgets, while the mention and authority figures point toward overall brand presence across the web doing more of the work.
Google's AI optimization guidance explicitly warns against inauthentic mentions and manufactured signals (Google AI optimization guide). Given how differently the engines source their answers, a synthetic mention campaign has to fool several source sets at once and stay fooled through model updates, which is a poor use of a budget.
The durable version is accurate, consistent facts on your own properties combined with genuine third-party corroboration: reviews from real customers, expert commentary that a publication chose to run, comparison coverage you didn't write, and community discussion where your name comes up unprompted. That combination is what the off-site half of our AEO system is built to produce, and it takes longer than a schema sprint because every piece of it requires a third party to agree to say something about you first.
What we watch happen in the citation data is that an engine with no way to verify your claims against a source outside your own domain will quietly route around you and recommend a competitor whose claims it can check. Kevin Barber, our Head of AI Growth, has described that outcome as the whole ballgame: "It literally can be the difference between citation and recommendation versus being totally left out."