AI VISIBILITY FOR SMALL BUSINESSES  |  5 MODELS MEASUREDREPORT IN 3 DAYS
15 Oct 2026 · 4 min read

Cited is not the same as named

A source link under an AI answer feels like a win. In most of the appearances Semrush examined, the domain was linked and the brand was never mentioned.

Open any AI answer and you see two things: the text, and a row of source links. Visibility reports count both as "being in the answer". They are different events. A citation means your page was used as a source. A mention means your name stands in the sentence the reader actually reads. Someone asking for a tax advisor in Regensburg reads the sentence.

The gap, measured

In June 2026 Semrush, together with Kevin Indig's Growth Memo, looked at 3,981 domain appearances across 115 prompts, 14 countries and four engines: ChatGPT, Google AI Overviews, Gemini and Google AI Mode (Semrush). For each appearance they recorded whether the domain was linked, whether the brand was named, or both.

61.7%
of appearances were "ghost citations": the domain was linked as a source, the brand name never appeared in the answer. Only 13.2% were both cited and named; another 25.1% were named without a link.
Semrush / Growth Memo, 3,981 appearances, June 2026

The engines behave very differently. ChatGPT linked the domain in 87% of appearances and named the brand in 20.7%. Gemini is the mirror image: it named the brand in 83.7% and linked it in 21.4%. For 22% of the prompt-and-domain combinations (100 of 454), the engines disagreed on whether the brand deserved a name at all. So "we get cited by ChatGPT" and "ChatGPT recommends us" can be two separate sentences about the same company.

The type of question matters too. Informational prompts produced a citation in 89.3% of cases but a mention in only 18%. Comparative prompts, the "which is better" kind, produced a mention in 43.3% (Semrush). A page that explains a topic gets used. A question that asks for a choice gets names.

One engine is not the market

A second study, run by the agency Fractl, ran 96 prompts across eight industries, 15 times each on three models, which gave 4,320 responses. Only 900 brands, 11% of the total, were referenced by all three models; 77% appeared in just one (Search Engine Land). Read it as direction, not as decimals: the author co-founded Fractl, and the percentages in the article do not fully reconcile with the brand counts.

11%
of brands were referenced by all three models tested. Three in four appeared in only one.
Fractl via Search Engine Land, 4,320 responses, Aug 2026
A citation says the model read you.
A mention says it is willing to put your name in a sentence. Only the second one is something a customer can act on.

Why the name is the whole prize

Nobody follows the links anyway. Pew Research tracked 68,879 real Google searches by 900 US adults: when an AI summary appeared, users clicked a link inside it in 1% of visits (Pew Research). We covered that in Nobody clicks the link. If the link is rarely opened, a ghost citation delivers almost nothing: the reader sees an answer, remembers whichever name was in it, and never learns that your page supplied the facts.

What tends to put a name into the sentence

Semrush's own advice for consumer brands is to track unlinked mentions on third-party sites such as news and listicles next to backlinks, and to add clearer brand references and comparison sections to pages that are cited but not named. The correlation data points the same way. Ahrefs analysed 75,000 brands and found branded web mentions correlating at 0.66 to 0.71 with AI visibility, and YouTube mentions at about 0.737, the strongest single factor (Ahrefs). That is correlation, not proof that mentions cause visibility, and it measures the topic broadly rather than the cited-versus-named gap specifically.

What this means with five employees

No study here tests a fix. Semrush measured the gap; it did not show how to close it. The sample is small (115 prompts), Perplexity is not included, and Semrush sells a tool that measures exactly this. What you can do is cheap and does not depend on the exact percentages.

First, read the right half of the answer. Ask ChatGPT, Gemini and Google the questions your customers ask. Ignore the source list at first and look for your name in the sentences. If your page is in the sources and your name is not in the text, you are a ghost on that question.

Then, make the name hard to leave out. A page that says "we" and "our team" gives the model facts without a label. Put the firm name, the service and the place into the headings and the first paragraph, and add a short comparison section: who the service suits, who it does not, what the alternatives cost. That is the structure Semrush suggests for pages that are cited but not named.

Finally, get named where you do not control the page. A mention in a regional article, a supplier's reference list or a trade video is a sentence someone else wrote about you. For a firm this size, that is the slower and harder half of the work, and the one the data keeps pointing at.

// free_preview

See whether AI names you, or only reads you.

We ask ChatGPT, Perplexity, Google AI and Claude the questions your customers actually ask, and show where your name appears in the answer text, not only in the source list. No account, no card.

run_check →

Sources

  1. Semrush & Growth Memo — Why 62% of AI citations don't lead to brand mentions (3,981 appearances, 115 prompts, 14 countries; 61.7% ghost citations; ChatGPT 87% cited / 20.7% named; Gemini 83.7% named / 21.4% cited)
  2. Search Engine Land (Fractl) — brands vanishing from AI search (4,320 responses; 11% of brands in all three models, 77% in one)
  3. Pew Research — Google users are less likely to click on links when an AI summary appears (68,879 searches; 1% click inside the summary)
  4. Ahrefs — AI brand visibility correlations (75,000 brands; branded mentions 0.66–0.71; YouTube mentions ~0.737)