Most businesses think of AI visibility as a single outcome: either the AI mentions you or it doesn’t. New research shows it is actually two separate outcomes — and most businesses only achieve one of them.
A study by Semrush and Kevin Indig analyzed 3,981 domain appearances across 115 prompts, 14 countries, and four AI search engines: ChatGPT, Google AI Overviews, Gemini, and Google AI Mode. What they found reframes how AI visibility should be measured.
The ghost citation problem
When an AI engine answers a question, it can do two distinct things with a business. It can cite the business — use its website as a source, usually shown as a small link. Or it can name the business — actually say the brand in the answer text as part of a recommendation.
These sound like the same thing. They are not. The study found that 61.7% of AI citations were “ghost citations” — the source link appeared, but the brand name did not. Only 13.2% of appearances were both cited and named.
Of AI citations are “ghost citations” — the source is linked, but the brand is never named in the answer the user reads
Source: Semrush & Kevin Indig — 3,981 AI citations analyzed — semrush.com/blog/the-ghost-citations-study/
The practical consequence: a business can be quietly powering AI answers — trusted enough to be used as a source — while the person reading the answer never learns its name. The citation does the work; the brand gets none of the credit.
Citations show which pages a model trusts as evidence. Mentions show whether the model knows the brand well enough to say its name. Those are two different jobs.
Every AI engine behaves differently
The split between citing and naming is not consistent across engines. Each one leans a different direction, which means a one-size-fits-all approach to AI visibility leaves gaps.
| AI Engine | Behavior | What It Rewards |
|---|---|---|
| ChatGPT | 87% cite, 20.7% name | An evidence engine — rewards trusted, well-structured source pages |
| Gemini | 83.7% name, 21.4% cite | A memory engine — rewards brands it already knows from across the web |
| Short conversational queries | Nearly 100% name the brand | Natural-language questions trigger brand naming |
| Comparison queries | 2.4x more brand mentions | “Best X” and “X vs Y” content drives naming, not just citing |
ChatGPT will happily use your page as a source but rarely says your name. Gemini will name brands it recognizes but rarely links to them. To be both cited and named — the 13% ideal — a business has to satisfy both behaviors at once.
What actually gets your business named
If citations come from your website, where do brand mentions come from? A separate study released the same month answered this directly — and the answer is not your website.
Researchers ran 14,140 queries across five AI engines studying how 12 brands were recommended. When AI engines named a brand as a recommendation, 82% to 100% of those recommendations traced back to third-party sources — reviews, directories, and comparison sites — not the brand’s own website.
Of AI brand recommendations trace back to third-party sources — reviews, directories, and comparison sites — rather than the brand’s own website
Source: Co-mentions study, June 2026 — 14,140 API runs across 5 AI engines — searchengineland.com/co-mentions-ai-recommendation-gap-479829
The starkest example: two competing brands shared the exact same Google business profile description. One was named in 71% of AI recommendations. The other was named in 0%. Their websites and profile data were equivalent. The difference was how much the rest of the web — the third-party sources AI engines trust — talked about each one.
Being cited (evidence): built by a clear, well-structured, technically sound website that AI engines can read and trust as a source. Necessary — but on its own, it produces ghost citations.
Being named (memory): built by reputation across the wider web — strong reviews, accurate listings on the directories AI trusts, and presence on the comparison sites buyers turn to. This is what turns a citation into a recommendation.
What this means for a local business
A good website is necessary but not sufficient
A clean, well-structured website gets you cited. It is the foundation. But the research is clear that being cited without being named is the most common outcome — and it does not produce customers. The website is half the job.
Brand mentions are earned off your site
The 82–100% third-party dependency means the work that gets AI to name your business happens largely away from your own domain: review platforms like Yelp and Angi, business directories, industry associations, and the comparison and “best of” content that buyers and AI engines both rely on. This is reputation engineering across the whole web.
Comparison content is the highest-leverage format
Because comparison queries produce 2.4x more brand mentions than informational queries, comparison and “best [service] in [city]” content is the single most effective format for moving from cited to named. It is also the content buyers read at the moment they are choosing who to hire.
Measure both, not one
The most important takeaway is a measurement one. AI visibility should be tracked as two separate numbers: how often your business is used as a source (citation rate), and how often AI engines actually name you as a recommendation (brand mention rate). A business that looks visible on one measure can be invisible on the other.
Free Report · 5 Business Days
Find out how often AI actually names your business
We test your visibility across ChatGPT, Perplexity, Gemini, and Google AI Mode — measuring both how often you’re cited and how often you’re named — and show exactly where the gaps are.
Ghost citations data: Semrush & Kevin Indig, June 2026 — analysis of 3,981 domain appearances across 115 prompts, 14 countries, and 4 AI search engines — semrush.com/blog/the-ghost-citations-study/.
Co-mentions / recommendation data: Search Engine Land, June 2026 — 14,140 API runs across 5 AI engines over 7 days — searchengineland.com/co-mentions-ai-recommendation-gap-479829.