Ghost Citations: 62% of the Time AI Cites You, Your Brand Name Never Appears

On April 20, 2026, Kevin Indig published a study through Growth Memo that reframes what an AI citation actually is. The study covered 3,981 domains, 115 prompts, 14 countries, and 4 AI search engines. The core finding: 62% of AI citations are what Indig calls ghost citations.

A ghost citation is when an AI platform includes a URL as a source but never mentions the brand name in the response text. The domain gets traffic attribution. The company gets zero name recognition from the people reading the answer.

This matters for one practical reason. If your AI visibility strategy is built on citation count as the primary metric, you are measuring a signal that includes 62% events where buyers cannot see your brand. A citation that drives no brand awareness does not move pipeline.

What the data actually shows

Indig's methodology: 3,981 domains queried across 115 prompts, run across ChatGPT, Perplexity, Gemini, and one additional AI search engine, covering 14 countries.

The headline numbers:

• •74.9% of domains received at least one citation link across the study period
• •38.3% received at least one brand name mention
• •13.2% achieved both a citation link and a brand name mention in the same response

That last figure is the one that should recalibrate how you think about AI visibility. If you track citation presence for your brand, roughly 86% of the time when AI cites your domain, buyers are not seeing your name. They might click through to your site. They will not remember who they clicked through to.

The geographic variance adds another wrinkle. Brand mention rates were highest in India and Sweden at around 50%. They were lowest in Italy, Brazil, and the Netherlands at 18 to 22%. The same content, the same brand, dramatically different name visibility by country. If your customers are in Southern Europe or Latin America, your effective AI visibility may be substantially lower than your global citation count suggests.

ChatGPT and Gemini work in opposite directions

The most operationally important finding from Indig's study is the platform-level behavior split. ChatGPT and Gemini are doing opposite things.

ChatGPT cites sources in 87% of responses but mentions brand names in only 20.7% of answers. Indig describes ChatGPT as functioning "like an academic paper with footnotes": lots of references at the bottom, names rarely said in the text.

Gemini does the reverse. It mentions brand names in 83.7% of appearances but generates a citation link only 21.4% of the time. Gemini reads more like a spoken recommendation: it says the brand name, but does not typically link out.

These two behaviors require different content strategies if you want brand name visibility rather than just traffic credit.

For ChatGPT, a citation link is easy to earn through content quality and crawlability. Getting your brand name into the response text is harder and requires a different content type (covered in the next section). For Gemini, the inverse holds: appearing named is relatively accessible, but earning a citation link requires the kind of structured, reference-grade content that Gemini links to specifically. The Seer Interactive analysis from April 2026 found that Gemini's citation rate dropped 23 percentage points between February and March, while simultaneously moving toward heading-heavy, shorter, table-structured responses. Gemini now favors structured reference content rather than editorial perspective.

For which AI platform to optimize for, this split matters. A brand heavily dependent on Gemini for name recognition needs structured, comparison-driven content. A brand that cares about ChatGPT traffic needs content that earns link citations, then separately builds brand name mentions through evaluation content.

Evaluation content generates 30x more brand name mentions

The most actionable number from Indig's study: evaluation and comparison content generates 30x more brand name mentions than informational content.

"Here are the best tools for X" generates brand names. "Here is how X works" generates source links without names.

This is not a small difference. A factor of 30 means you could publish 30 informational posts and get the same brand name mention volume from a single well-structured comparison page where your brand is named. The content type is the primary variable, not volume.

This connects directly to the third-party content strategy that brand authority research has consistently supported. An Omniscient Digital analysis of 23,000+ AI citations found that 89% come from earned media, not brand-owned channels. When that earned media takes the form of comparison content and "best of" lists where your brand is named, you get the 30x mention multiplier on top of the citation. When it takes the form of informational explainers that link to you as a source, you get a ghost citation.

The implication for content investment: third-party comparison placements where your brand is explicitly named are more valuable for AI visibility than most owned content. A well-placed mention in an independent "best tools for [category]" roundup does more for your brand name recognition in AI responses than the equivalent effort spent publishing informational articles on your own domain.

Why citation count is a broken GEO metric

The ghost citation finding lands harder when you layer in two other data points from the same week.

First: Passionfruit published a research piece on April 21, 2026, citing AirOps data showing that 85% of content AI retrieves is never shown to users. AI platforms pull in far more source material than they surface in responses. A page can be crawled and retrieved as context, never cited, with zero user-facing impact. The 62% ghost citation number is what happens when a page does get cited. Before the citation, there is a much larger set of retrievals that produce nothing at all.

Second: SparkToro conducted a study with 600 volunteers and 2,961 prompts. Finding: there is less than a 1 in 100 chance that the exact same brand list will appear twice for the same prompt run twice. Less than 1 in 1,000 that the same list appears in the same order. This is the volatility problem that citation drift research has been pointing at: AI citation presence is not a stable position. It varies run to run, day to day.

Combine these three findings. 85% of retrieved content never appears to users. Of the content that does get cited, 62% of those citations give traffic credit but no brand recognition. And the specific list of cited brands changes almost every time someone runs the same query. The citation count your GEO tool reports is summing up events that are mostly invisible to buyers, mostly ghost citations when visible, and unlikely to repeat.

This does not mean AI visibility optimization is pointless. It means citation count is an imprecise proxy for the thing you actually want, which is: buyers seeing your brand name when they are in the process of evaluating options in your category.

The three metrics that capture real buyer visibility

The CITE framework distinguishes between Citation Rate and Recommendation Rate because this distinction matters operationally. Indig's data suggests a third metric matters too. Together:

Citation rate (URL appears): How often does your domain appear as a source link in tracked AI responses? This is the metric most GEO tools report. It counts ghost citations along with real brand mentions. Useful for tracking content crawlability and live retrieval access, but not a direct measure of buyer-visible brand presence.

Mention rate (brand name appears): How often does your brand name appear in the text of tracked AI responses, regardless of whether a link is included? On Gemini, this can be high even when citation rate is low. On ChatGPT, this is substantially lower than citation rate. This is the metric that corresponds to buyer-facing brand recognition.

Recommendation rate (brand explicitly suggested): How often is your brand actively recommended as the solution, not just referenced? This is the highest-value visibility event. Comparison pages and evaluation content are the primary drivers of recommendation-level appearances.

Tracking only citation rate will overstate your effective AI visibility by counting ghost citations. Tracking only mention rate will miss the traffic signal and miss ChatGPT's link-heavy behavior. Tracking recommendation rate separately identifies the visibility events most likely to influence buyer decisions.

For teams currently reporting one citation count number, the practical step is to separate these in your next audit. Run your tracked prompts and categorize each appearance: was your domain linked? Was your brand name stated in the response text? Was your brand explicitly recommended? Those three numbers tell a more accurate story than citation count alone.

How to reduce ghost citations in practice

Ghost citations are not random. The content type and platform determine how often you get name visibility versus source-only attribution.

Build more evaluation content. The 30x brand name mention multiplier from comparison content is the most direct lever. If your content mix is mostly informational ("how X works," "what is X"), you are feeding the ghost citation pattern. Add dedicated comparison pages: "[Your brand] vs [Competitor]," "Best tools for [use case]," structured evaluation frameworks where your brand is named and positioned rather than just linked.

Target Gemini's format preferences. Gemini now generates headings in 99.5% of responses and markdown tables in 52% of responses, according to Seer Interactive's April 2026 analysis. Gemini names brands frequently (83.7% of appearances). Content that matches Gemini's current format preference (short, heading-structured, table-driven) has the best chance of getting your brand name said rather than ghostly linked. The comparison page citation research covers this structure in detail.

Invest in third-party named mentions. When external sites, review platforms, and analyst coverage name your brand in comparison and evaluation contexts, those third-party mentions feed the AI training data that drives brand name mentions in responses. The brand authority research found that off-site brand mentions have a 0.664 correlation with AI citation frequency, the strongest single predictor measured. Named mentions in third-party evaluation content compound over time.

Check geographic exposure. If your target markets include Southern Europe or Latin America, where brand mention rates were 18 to 22% in Indig's study, platform-specific optimization matters more. Gemini's brand-heavy behavior may be a larger share of actual buyer-facing visibility in those markets. Audit your prompt performance by region before assuming global citation data represents local buyer experiences.

Track the share of voice measurement separately by content type. A simple split of your citation tracking into "appeared with brand name" versus "appeared without brand name" will surface how much of your current AI visibility is ghost citations. That data tells you where to shift the content mix.

FAQ

What is a ghost citation in AI search?

A ghost citation is when an AI platform includes a website URL as a source link in its response but never states the brand's name in the text of the response. The domain receives credit as a reference, and users can click through to the site, but the brand name does not appear in the answer the user reads. Kevin Indig coined the term in a Growth Memo analysis published April 20, 2026, covering 3,981 domains across 4 AI search engines.

Why does ChatGPT link to sites without mentioning the brand name?

ChatGPT functions similarly to academic citation style: it links sources at the end of passages without necessarily naming the company or product in the prose. Indig's study found that ChatGPT cites a source URL 87% of the time but mentions the brand name in only 20.7% of answers. The mechanism is that ChatGPT retrieves content for synthesis and attributes it with links, but the synthesized answer does not always require naming the source's brand explicitly. To get named rather than just linked, brands need content in contexts where the brand name is part of the answer, such as comparison and evaluation queries.

Does an AI citation help my business if my brand name never appears?

It helps with traffic, potentially. When users click the source link in a ghost citation, they arrive at your site. It does not help with brand recognition in that AI interaction, because the user never sees your name in the response text. For B2B SaaS brands where the buying decision involves an evaluation phase, getting named versus ghost-cited is the difference between being considered and being a background reference. The conversion case for AI search rests on buyers encountering your brand during their research process, not just getting a referral click.

How is Gemini different from ChatGPT for brand name mentions?

The behavior is almost a mirror image. ChatGPT cites URLs in 87% of responses but names brands only 20.7% of the time. Gemini names brands in 83.7% of appearances but generates a citation link only 21.4% of the time. Gemini prioritizes naming brands in response text; ChatGPT prioritizes linking sources. A brand that appears frequently in Gemini responses is likely to have stronger name recognition with Gemini users than its link-citation count suggests. Gemini's format shift toward shorter, table-structured, heading-heavy responses (from the Seer Interactive April 2026 analysis) also indicates Gemini is optimizing for readability over source attribution.

What content gets my brand named (not just linked) by AI?

Evaluation and comparison content generates 30x more brand name mentions than informational content, according to Indig's study. Queries like "best tools for X," "how does [your brand] compare to [competitor]," and category comparison roundups produce responses where brands are named as part of the answer. Informational content ("how X works") produces responses where sources are linked but brands are not typically named in the text. For B2B SaaS brands, the practical implication is that comparison pages, third-party review presence, and category listing appearances carry more brand-name-visibility value than the equivalent effort in informational blog content.

The number that changes how you report

62% ghost citations. 13.2% with both a link and a brand name. 30x more brand name mentions from evaluation content. 85% of retrieved content never shown to users.

Each of those numbers points to the same gap: the AI visibility metric most teams are tracking is counting events that buyers mostly cannot see. That does not make AI optimization irrelevant. It makes the current reporting framework imprecise in a way that matters for how you allocate effort.

The shift is straightforward in practice. Report citation rate, mention rate, and recommendation rate as three separate numbers. Build more evaluation content. Prioritize third-party named mentions over informational owned content. Run your tracked prompts by region if your buyers are in markets where Indig found lower mention rates.

The brands that hold AI visibility through the next round of model changes are the ones getting named in response text, not ghostly linked in footnotes.