A buyer asks ChatGPT which CRM suits a company their size. The answer names three vendors. One of them is your competitor.
You open the sources panel and your site is there. The comparison guide your team spent six weeks on supplied the framing the model used to build its answer. Your name appears nowhere in the text the buyer actually read.
That is a ghost citation: the engine links your page as a source and never mentions the brand behind it. It is a specific way to lose that almost nobody measures, because it does not look like losing. It looks like nothing at all.
An analysis by Writesonic covering roughly 16 million brand appearances over a 30-day window put the rate at about 40% of citations. Worth noting the study was co-authored by Writesonic's own CEO, so treat the precise figures as directional. The pattern holds across every engine measured, and none is close to zero.
Every engine plays by different rules
The rate splits sharply: Perplexity 52%, Google AI Mode 49%, AI Overviews 41%, ChatGPT 37%, Gemini 25%, Grok 22%, Microsoft Copilot 19%.
That spread is not noise. Writesonic's analysis found a structural divide it labels namers and citers. Namers (Gemini and Copilot) put brands in the answer text but link to source pages less often. Citers (Perplexity and Google's answer engines) link generously but are inconsistent about naming the brands behind the links. ChatGPT and Grok sit between the two without doing either especially well.
Both failure modes cost you something different. A mention without a citation makes you visible but sends no one to your site. A citation without a mention hands you a link nobody clicks, because nobody knows it is yours. Getting both at once is the rarest outcome of the three.
Being known is not the same as being named
The more revealing data comes from a separate study by Victorious, published in July 2026, covering 175 brands across five verticals and eight AI platforms.
Ask an AI directly about a brand and it does fine: 96% were described accurately. The models know who these companies are.
Then ask the questions real buyers ask, the ones that don't name anybody, and 89% of those same brands never appeared in the answer at all. At the problem-awareness stage, where a buyer is describing a symptom rather than shopping for a category, brand mention rates fell to 0.10%.
The gap between those two numbers is the whole problem. AI understands your business. It does not bring you up. Most companies audit the first thing, feel reassured, and never test the second.
Your own website is not the lever
Here is where the standard response fails. Told they have an AI visibility problem, most teams write more content on their own site.
Victorious found that of the 49,391 citations analyzed, 99.99% pointed to third-party websites rather than the brand's own domain. That is effectively all of them, not merely most.
Muck Rack's ongoing study of what AI systems actually read (more than 25 million links across ChatGPT, Claude, and Gemini) lands in the same place: earned media accounts for 84% of AI citations. Journalism alone is 27%. Paid and advertorial content is 0.3%.
That 0.3% deserves a second read. You cannot buy your way into being named. The substrate AI draws on is coverage, reference, and third-party corroboration, the stuff that used to sit under "PR" in the budget, filed well away from anything measured on clicks. Getting named is a PR problem that most teams still hand to the SEO team.
Which points at the one lever you fully control. Publish research that carries your name inside the finding. A statistic from "the 2026 Khalife Group pricing study" is far harder for a model to strip than a statistic from "one analysis found," because the attribution is load-bearing in the sentence rather than sitting in a link. Build the credit into the work and it travels with the work.
Victorious also found a threshold effect: brands with fewer than 2,000 indexed pages mentioning them appeared in AI answers just 3% of the time, with mention rates climbing steeply above that. What moves the number is how often others write about you, not how much you publish yourself.
The fragmentation problem
One more finding, from Kevin Indig's H1 2026 analysis: 91% of citations appear in only one engine.
Being cited by ChatGPT tells you almost nothing about whether Perplexity or Gemini will cite you. There is no single AI ranking to win. There are several largely independent ones, drawing on overlapping but distinct source sets, and a strategy tuned to one is close to unhedged against the others.
What this actually asks of you
Start with the audit, because most teams run the wrong one. Ask AI what it knows about your company and it will reassure you; it knows plenty. The question that matters is the unbranded one your buyers actually type. Run it across at least four engines and record whether you surface at all. That number is your real baseline.
Where you spend follows from where the citations come from. If 84% are earned media and 0.3% is paid, the highest-leverage work is digital PR, analyst relations, and getting referenced where your category is already discussed. That lands badly for anyone who has spent three years building a content engine on their own domain, which is exactly why it needs saying.
Then change what you count. Reporting built on citations overstates the thing you care about by roughly 40%, so separate mentions from citations and stop logging a cited-but-unnamed page as a win. It is the opposite of a win: proof your content is doing the work in an answer that belongs to someone else.
Your content may already be shaping what AI tells your market. Whether your name travels with it is a different problem, and your own website is not where it gets solved.