What Are Ghost Citations When AI Cites a Domain but Not a Brand?

What did the study find?

The study ran 115 prompts in ChatGPT, Google's AI Overviews, Gemini and Google's AI Mode, and sorted every domain appearance into three groups:

Engines behaved very differently:

• ChatGPT cited brands' domains 87% of the time they appeared, but named them in the text only 20.7% of the time.
• Gemini did the reverse, naming brands 83.7% of the time but citing them only 21.4% of the time.

Other findings: comparative prompts, such as "best", "vs" and "recommend" questions, named a brand about 2.4 times as often as informational prompts. Informational content earned citations but rarely mentions. Mention rates also varied by country, from about half of answers in India and Sweden to 18–22% in Italy, Brazil and the Netherlands.

Why do ghost citations happen?

• Your page supports a fact, not a recommendation. Assistants use informational pages as evidence for general statements, like "most heat pumps last 15 to 20 years", without attributing them.
• The brand is not part of the fact. If your page states a figure without tying it to your name, the answer can repeat the figure without the name.
• Engine design. Some engines show sources as links next to the text; others weave names into the answer. The same content produces different outcomes in different engines.
• Long, detailed prompts tend to produce more citations and fewer brand mentions, according to the study.
• Generic domain names that do not match the brand name make the source less recognisable, even when the link is seen.

Do ghost citations matter?

They matter less than mentions for brand building, but they are not nothing:

• They show trust. Your content is being selected as evidence.
• They can bring clicks. Some readers follow source links, especially for detailed topics.
• They are an opportunity. A page already cited is close to earning a mention, if the brand becomes part of what the answer repeats.

For businesses that depend on being shortlisted, such as B2B software, services and consumer brands, mentions in the answer text matter most. A buyer who never sees your name cannot choose you.

How do you measure them?

For each tracked question and run, record two things separately:

1. Cited: does any link in the answer point to your domain?
2. Mentioned: does your brand name appear in the answer text?

From these, calculate:

• Citation rate: share of runs citing your domain.
• Mention rate: share of runs naming your brand.
• Ghost share: share of your citations where the brand is not named.

Break them down by engine and by content type. A high ghost share on informational guides is normal. A high ghost share on comparison pages suggests the brand is not coming through. Why AI tools and analytics struggle with brand citations is covered in whether GA4 or Semrush can measure AI brand citations.

What about mentions without citations?

The reverse case, where the brand is named but no link to its site appears, made up about a quarter of appearances in the study. It is common in Gemini and in answers drawn from a model's general knowledge. For brand recognition, it is often the better outcome: the buyer sees the name. But it has two weaknesses. There is no link to click, so the visit happens later through search or a typed URL, which analytics usually credits elsewhere. And without a cited source, you cannot easily see what the description is based on, which makes errors harder to trace. Track these unlinked mentions as part of your mention rate, and check their accuracy as closely as linked ones.

How should you report them to stakeholders?

Show citation rate and mention rate side by side, by engine, rather than a single "visibility" figure; the method is covered in why to measure brand mention rate separately from URL citation rate. A report that says "cited in 40% of answers, named in 12%" tells leadership something a combined number hides. Add a short note on which content types drive each. That turns the ghost citation gap into a clear brief for the content team.

How can you turn citations into mentions?

• Make the brand part of the fact. Publish original data under a name: "The [Brand] 2026 Heating Cost Index found..." Answers that repeat the figure are more likely to repeat the name.
• Use named methods and frameworks where they genuinely describe your approach, so the method cannot be described without the name.
• Write attributable statements: "In [Brand]'s analysis of 2,000 installations, ..." rather than an unattributed claim.
• Publish comparative content. In the study, answers to comparison prompts such as "best" and "vs" named a brand in 43.3% of cases, 2.4 times the rate for informational prompts (how-to prompts were close, at 42.8%). Pages built for those questions put your brand where naming happens.
• Earn third-party mentions. When other sites name you alongside the facts, assistants see the brand and topic together. Ahrefs' study of 75,000 brands found branded web mentions correlated strongly with visibility in AI Overviews.
• Keep naming consistent across your site, profiles and press, and make your domain recognisably yours where possible.

Avoid stuffing the brand name into every sentence. It reads badly and is unlikely to help. The aim is to make the brand inseparable from the information, not repeated around it.

Whether syndication helps is covered in whether syndicating content to trusted sources improves AI citations.

What might an illustrative case look like?

Consider an illustrative payroll software company whose guides on employment law are cited often in ChatGPT answers, but whose name almost never appears. Its tracking shows a citation rate of 34% and a mention rate of 6% on 40 unbranded questions. The guides state facts, such as filing deadlines, without any branded element. The company publishes an annual survey of small business payroll errors, named after the company, and adds its figures to the guides with attribution. It also creates comparison pages for its category. Two quarters later, its citation rate is similar, but its mention rate has risen to 15%, mostly on comparison questions and questions where the survey figures are used.

What do people ask about ghost citations?

Is a ghost citation better than nothing?

Usually yes. It shows the content is trusted and can bring some clicks. But it does little for recognition on its own.

Why does Gemini name brands without citing them?

Engines present sources differently. Gemini tends to weave brand names into the answer, while ChatGPT tends to show source links. Track each engine separately.

Should we stop writing informational content?

No. It earns citations and supports your authority. Pair it with original data and comparison content that carry the brand.

How do we know if our brand is being named?

Record the answer text for each tracked question and check for your brand name and common variants, separately from the cited links.