Semrush study finds AI citations often miss brand mentions

AI systems can send your content traffic, authority, and even trust signals without giving your brand any memory or mindshare in the answer. That is the real sting in Semrush’s latest study with Kevin Indig: a page may be used as evidence, yet the company behind it never gets named. Across 3,981 domain appearances in ChatGPT, Gemini, Google AI Overviews, and Google AI Mode, 61.7% were ghost citations, where the source was cited but the brand name never appeared in the response.

That gap matters because a citation and a mention are not the same outcome. A link can make your domain look useful to the model, but if the user never sees your brand in the answer, you are not winning the recommendation in the way most marketers think about visibility.

Why the distinction matters more than most teams assume

If you have been watching AI search through the old SEO lens, it is easy to obsess over whether a domain appears in the citation stack. Semrush’s framing is stricter and more useful: visibility is not just about being used as a source, it is about being named as the answer. Only 13.2% of appearances in the study produced both a citation and a brand mention, which tells you how often those two signals split apart in practice.

That split changes what success looks like. A cited page can still be invisible as a brand touchpoint, while a named brand can build recognition even when the user never clicks. Semrush’s own AI visibility toolkit reflects that reality by treating mentions, citations, and positioning as separate metrics, not one blended score.

What the study actually measured

This was not a small scrape of a few prompts in one engine. The dataset covered 115 prompts across 14 countries and tracked 3,981 domain appearances across ChatGPT, Gemini, Google AI Overviews, and Google AI Mode. That breadth matters because it shows the pattern is not tied to one interface or one language market.

The headline numbers are blunt. Ghost citations made up 61.7% of appearances, while both a citation and a mention showed up together in only 13.2% of cases. In other words, the default behavior is often “use the source, skip the brand.”

Query shape changes the odds

The study’s most practical insight is that prompt style changes how often AI actually names a company. Short, conversational queries produced 30x to 50x more brand mentions, while long prompts tended to trigger more citations but fewer mentions. That is exactly the kind of nuance that gets lost when teams report only citation counts.

If you are writing for AI visibility, that means you cannot treat every query the same. Compact, natural language prompts seem far better at surfacing brands, while sprawling prompts may still pull in source material without giving the brand a visible role in the answer.

Comparative content beats plain information pages

Semrush also found that comparative content produced 2.4x more brand mentions than informational content. That is a useful signal because it tells you the model is more willing to name brands when the page helps it distinguish between options, rather than just define a topic.

This is where a lot of content strategies miss the mark. A clean explainer can earn citations, but a compare-and-contrast page, a “best for” breakdown, or a direct “X vs. Y” format is more likely to get the model to name the company in the answer. If you want your brand associated with a decision, not just a fact, comparison work carries more weight.

How to build content that gets named, not just used

The tactical takeaway is not to chase every possible mention with gimmicks. It is to make your brand easier for the model to identify when the query calls for a company, product, or service recommendation. Start with the pages where you already have a reason to be compared, then make the brand cues impossible to miss.

• Use comparative pages that plainly connect your brand to alternatives, use cases, and tradeoffs.
• Put the company name in the title, headers, and opening lines where the page is meant to represent the brand.
• Write in short, conversational language on pages that target natural questions, since those queries produced far more mentions.
• Build content around explicit choices, because AI systems seem more willing to name brands when the answer is framed as a decision.
• Track whether a query leads to a citation only, a mention only, or both, since those are different business outcomes.

This is also why brand messaging matters inside the copy itself. If the model can summarize your page without needing to repeat your name, you may get credit without recall. The stronger your entity cues, the more likely the brand survives the condensation process.

The bigger pattern behind the numbers

Semrush’s 2025 AI visibility research already showed that AI search can contradict traditional SEO assumptions about authority and source credibility. That finding pairs neatly with this newer study: the best-known source is not always the one that gets named, and citation behavior does not map cleanly to old authority signals.

Semrush has also reported that AI citations are concentrated among a relatively small set of domains, with preference shifts across different tools. Put those pieces together and the picture gets sharper: AI visibility is not just about ranking in a familiar sense. It is about which brands become visible, remembered, and associated with the answer after the model has done its compression work.

That is why the separate Semrush analysis showing brand mentions in 26% to 39% of responses across five LLMs matters too. Mentions are common enough to measure, but far from guaranteed. Brand naming is not a side effect of citation; it is its own outcome, and teams that treat it that way will make better content decisions.

The cleanest reading of the research is simple: if your pages only feed the model information, you may win a citation and still lose the brand. If you want the answer to carry your name, you need content built for recognition, not just retrieval.