AI search will cite your content without ever naming your brand. A June 2026 study by Semrush and growth advisor Kevin Indig, analysing 3,981 brand appearances across ChatGPT, Gemini, Google AI Overviews, and Google AI Mode, found that 74.9% of brand appearances included a citation but only 38.3% included an actual brand mention. In other words, the majority of the time an answer engine draws on your content, it uses you as a source and never says your name. This is the ghost citation problem, and it breaks the assumption underneath most AEO measurement.
Citations and Mentions Are Not the Same Win
For a decade, search success meant a link. The link was the destination, the credit, and the conversion path in one. Answer engines have split those apart. A citation is the model using your page as a source. A mention is the model naming your brand in the answer it gives the user. Only 13.2% of appearances in the Semrush dataset were both cited and mentioned.
The distinction matters because users act on mentions, not citations. When a buyer asks ChatGPT which platform to use and the answer recommends three named brands, those three brands win the consideration set. The dozen sources the model quietly read to assemble that answer, including possibly yours, are invisible to the buyer. You can be the research behind the recommendation and still lose the recommendation.
Why the Two Diverge
The mechanism is straightforward once you see it. Answer engines synthesise. They pull facts from source pages (citations) but they name the entities they have the strongest, most consistent representation of (mentions). A model cites a page because it answered a query. It names a brand because it knows that brand as an entity: a well-defined node with consistent attributes across the web that the model is confident enough to put in front of a user.
That is why a small brand with excellent content can be cited constantly and named rarely, while a brand with weaker content but a strong, consistent entity footprint gets named repeatedly. BrightEdge has documented the corollary: AI engines frequently cite different sources but recommend the same brands. The sources are interchangeable. The named brands are not.
The citation slots are also less stable than they look. Ahrefs, analysing a large sample of AI Overview results, found that only 38% of cited pages rank in the top 10 for the query, down from 76% a year earlier. The source an engine reaches for shifts query to query and month to month. The brand it has learned to name is far more durable. Optimising for the unstable slot while ignoring the durable one is the strategic error the ghost citation data exposes.
What This Looks Like in a Real Query
Consider a buyer who asks an assistant to recommend a platform for a specific job. The model runs the question, reads a dozen or more pages to assemble context, and returns an answer that names three or four products with a sentence on each. The buyer evaluates those named products and ignores everything else, because everything else is invisible to them.
Your page may have been one of the dozen the model read. If your brand was not in the named shortlist, you contributed the research that sold a competitor. The citation counted in your analytics. The recommendation went to whoever the model was confident enough to name. That asymmetry, repeated across thousands of commercial queries a day, is the commercial cost of the ghost citation gap.
The Measurement Shift This Forces
Most brands measuring AEO today track citation rate: how often they appear as a source. That number can rise while commercial outcomes flatline, because citation is not the metric that moves buyers. The discipline now is to track two numbers separately:
Citation rate tells you whether your content is good enough to be used. It is a content-quality signal.
Mention rate tells you whether your brand is strong enough to be recommended. It is an entity-strength signal.
A healthy AEO programme moves both, and a programme that moves only citation rate is optimising the half that does not convert. Treating them as one number hides the gap that the Semrush data just made impossible to ignore.
How to Become the Named Brand
Closing the gap is an entity problem, not a content-volume problem. Four levers move mention rate.
Entity consistency. A model names brands it can resolve confidently. That requires the same brand facts, name, category, founding details, core offering, repeated consistently across your site, your structured data, and the third-party sources the model trusts. Inconsistency across these sources is the single most common reason a frequently-cited brand goes unnamed.
Brand-entity strength over backlinks. Independent research into AI citation behaviour finds brand search volume to be the strongest single correlate of citation frequency, ahead of traditional backlinks. Building a recognised entity, the thing people search for by name, feeds the same signal models use to decide who to name.
Definitional, attributable content. Write the sentences a model can lift as an answer with your name still attached: clear "X is Y" definitions, named frameworks, and claims specific enough that paraphrasing them without attribution would lose meaning. Generic content gets absorbed; distinctive, named content gets named. We covered the underlying signals in brand authority signals AI agents actually trust.
Structured data that ties content to entity. Schema that connects each page back to a coherent Organization entity helps the model link the source it is citing to the brand it should name. Citation without that link is exactly how a page gets used and the brand gets dropped.
The Starting Point
The fastest way to find your own ghost citations is to run your top commercial queries through ChatGPT, Gemini, and Google AI Mode and record two things for each: were you cited as a source, and were you named in the answer. The pattern usually emerges within a dozen queries. Where you are cited but not named, the fix is rarely more content; it is tightening the entity, the consistency of your brand facts across the web, and the strength of the named-search demand that teaches models who you are. Treat the named-but-absent queries as the priority backlog, because those are the answers a competitor is currently winning with your research underneath them.
What to Watch
Two shifts will decide how much this matters. The first is whether the answer engines begin surfacing more named recommendations by default as commercial intent features mature; the more the interface foregrounds named brands, the more expensive the ghost citation gap becomes. The second is measurement tooling: as mention-rate tracking separates from citation tracking across the AEO platforms, the brands that have been quietly cited-but-not-named will see the gap quantified on their own dashboards.
Being cited proves your content earned its place in the answer. Being named is the only version of that win the buyer ever sees. The brands that understand the difference will spend the next year closing it. The rest will keep congratulating themselves on a citation rate that never reaches a customer.
