Marketers often treat an AI citation as proof that a brand is visible. That assumption can be badly misleading. An AI system may use your page as evidence, link to your domain, and still never say your company name in the answer.
Semrush and Kevin Indig call this a ghost citation. The distinction changes how AI visibility should be measured: citations tell you whether your content is being used; mentions tell you whether your brand is actually entering the user's field of view.
A ghost citation occurs when an AI system cites a page from your domain as a source but does not name your brand in the generated answer.
That makes citation visibility and brand visibility related but separate outcomes. A source can inform the answer without the source brand becoming part of the answer.
The study found that most citations were ghost citations
In a June 2026 study conducted with Kevin Indig and Growth Memo, Semrush analyzed 3,981 domain appearances across 115 prompts, 14 countries, and four AI search experiences: ChatGPT, Google AI Overviews, Gemini, and Google AI Mode.
For every appearance, the researchers tracked two separate outcomes: whether the domain was cited as a source and whether the brand name appeared in the answer text. The split was striking.
Of observed appearances were ghost citations: cited as a source, but the brand was not named.
Were both cited and mentioned, combining source attribution with explicit brand visibility.
Were brand mentions without a citation to the associated domain.
Put another way, 74.9% of appearances in the study included a citation, but only 38.3% included a brand mention. A citation rate almost twice the mention rate means that a dashboard focused only on cited URLs can make a brand look much more visible than it actually is to the person reading the answer.
A citation answers: Did the AI use our content as evidence?
A mention answers: Did the AI actually name our brand?
A recommendation answers: Did the AI present our brand as a viable choice for this buyer?
Why AI can use your content without naming your brand
The behavior is easier to understand when you separate the role of a source from the role of an entity. A source can provide a statistic, definition, explanation, or piece of factual context. The AI system may use that information to construct the answer without needing to identify the publisher that supplied it.
That is especially common with informational content. If a page explains how payroll works, defines a medical term, or publishes a market statistic, the answer may borrow the underlying information while keeping the user's attention on the subject rather than the source brand.
Brand mentions become more necessary when the task itself requires naming market participants. A prompt asking for the best CRM platforms, alternatives to a product, or recommended agencies cannot be completed without identifying companies.
Citation-heavy content
Research, definitions, explanatory articles, reference pages, and educational content can supply raw material for an answer even when the publisher's brand is irrelevant to the user's question.
Mention-heavy content
Comparisons, Best Of lists, alternatives pages, reviews, buyer guides, and recommendations require the answer to identify specific brands and explain where they fit.
Query intent changes the odds of being named
The Semrush/Kevin Indig study found a large difference between informational and comparative prompts. Informational queries such as "what is," "explain," and "how does" had an 89.3% citation rate but only an 18% mention rate. The AI was frequently sourcing information without naming the associated brands.
Comparative prompts such as "best," "vs," and "recommend" produced a 43.3% mention rate, about 2.4 times the brand-mention rate of informational prompts. How-to prompts reached a 42.8% mention rate, while commercial prompts involving pricing or buying intent reached 35.6%.
This does not mean informational content is unimportant. Informational content can build topical authority, earn citations, support traditional SEO, and influence how AI systems understand a subject. It simply performs a different visibility job from content designed to put brands into a consideration set.
Different AI engines behave differently
Another reason to separate citations from mentions is that AI platforms do not use the two signals in the same way. In the study, ChatGPT cited domains in 87% of appearances but mentioned brands in only 20.7%. Gemini behaved almost in reverse: it named brands in 83.7% of appearances but cited a source only 21.4% of the time.
That means a single blended "AI visibility score" can hide material differences. A brand may look strong in ChatGPT because its content is frequently cited, yet weak from a brand-awareness perspective because its name rarely enters the response. The same brand could show a very different pattern in Gemini.
| Outcome | What the user sees | What it tells the marketer | Primary metric |
|---|---|---|---|
| Cited only | A source link to your page, but no brand name in the answer. | Your content is useful to the AI, but your brand is not necessarily gaining consideration. | Citation rate / cited pages |
| Mentioned only | Your brand name appears without a citation to your domain. | The brand is visible in the answer, potentially because of broader market familiarity or third-party evidence. | Mention rate |
| Cited + mentioned | Your brand is named and your site is used as a source. | You are earning both brand visibility and source attribution. | Combined appearance rate |
| Recommended | Your brand is presented as a relevant option for the user's need. | You have moved beyond visibility into consideration. | Recommendation frequency |
Citations and mentions can diverge even at category level
A later Semrush study reinforces the distinction. In research covering more than 50,000 brands in ChatGPT, Semrush found that only 21% of the most-cited domains in a category were also the most-mentioned brand. The relationship between the two signals was slightly negative in that dataset.
This is a separate study with a different methodology, so the numbers should not be combined with the ghost-citation dataset. But the strategic implication is consistent: being the source most often cited for a topic does not automatically make you the brand most often named when AI answers questions about that topic.
Why this matters more for commercial AI visibility
For publishers, research organizations, and educational sites, citation visibility may be a primary success metric. Being used as a trusted source has value even when the organization's name is not central to the response.
For a company trying to sell software, professional services, consumer products, healthcare, financial services, or another commercial offering, that is not enough. The goal is not merely to contribute information to the answer. The brand needs to enter the consideration set when buyers ask which companies, products, or services deserve attention.
This is where the distinction connects directly to mentions, citations, recommendations, and clicks as separate AI visibility metrics. Each one answers a different business question, and none should be used as a substitute for the others.
How to diagnose your ghost-citation problem
A useful audit should start with the prompts that matter commercially rather than with a list of every URL AI has ever cited from your site.
Build a controlled prompt set
Include informational, comparative, "best for," alternatives, use-case, pricing, and recommendation prompts. The goal is to see where citation behavior changes as buyer intent changes.
Classify every appearance
Tag each result as cited only, mentioned only, both, recommended, or absent. Do not collapse those categories into a single score.
Segment by engine and intent
A ghost-citation problem in ChatGPT may not look the same in Gemini or Google AI Mode. Break performance out by platform, query type, geography, and buyer stage.
Inspect what competitors are doing differently
Look at the pages and third-party sources supporting prompts where competitors are named and you are only cited. The gap may be comparison coverage, clearer positioning, reviews, listicles, directories, or another recommendation-oriented format.
Measure movement separately
Track citation rate, mention rate, recommendation frequency, AI share of voice, cited pages, and source mix independently. Improvement in one should not be reported as improvement in all.
How to close a citation-to-mention gap
If your site is frequently cited but your brand is rarely named, the answer is not to weaken the educational content that is already earning citations. Keep it. It is doing its job.
The missing layer is usually brand context. AI systems need enough credible evidence connecting your brand to specific categories, use cases, audiences, differentiators, and competitive situations. That evidence can exist on your own site, but it is especially important across independent sources when the user is asking for market recommendations.
Clarify the entity on owned content
Make it obvious who created the research or expertise, what the company does, which products or services connect to the topic, and why the brand is relevant. Do not force brand references where they do not belong.
Build credible third-party context
Pursue relevant Best Of lists, comparisons, reviews, expert roundups, use-case recommendations, directories, and buyer guides where your inclusion is truthful and useful to the reader.
Target the prompts that require brands
Commercial comparison questions are more likely to require explicit brand names than generic educational questions. Prioritize the recommendation contexts tied to real demand.
Track narrative, not just occurrence
Being named is only useful if the description is accurate. Monitor the attributes, use cases, strengths, and limitations AI associates with your brand.
A citation is evidence that AI found your content useful. A mention is evidence that your brand entered the answer. A recommendation is evidence that your brand entered consideration.
Measure all three. Optimizing one while assuming the others will follow is exactly how ghost citations become an invisible reporting problem.
Measure source authority and brand visibility as two different assets
The most useful way to think about ghost citations is not as a failure of content. A page that earns citations has already demonstrated value. The problem is assuming that source authority automatically creates brand authority in the answer.
For AI search, marketers increasingly need both: content strong enough to be used as evidence and a broader recommendation footprint strong enough for the company itself to be named. That is why a modern measurement system should show citations and mentions side by side, then connect both to recommendation frequency, referral traffic, leads, and revenue where possible.
If the issue is broader than a single prompt, start with a brand mention gap analysis, then build a deliberate brand mention strategy for AI search around the commercial contexts where you are missing.
Find where your brand is being used but not seen.
LSEO AI can help measure mentions, citations, recommendation frequency, competitors, traffic, and business outcomes. When the gap is third-party recommendation coverage, Mention Engine helps brands systematically build credible visibility in the content types AI systems frequently use when buyers compare and choose.
Explore Mention Engine