Can a brand be cited by an AI without being recommended?

In the AI visibility audits we conduct, one scenario comes up often. A company sees its site appearing in the sources for Perplexity or ChatGPT and concludes that it is well-positioned. Then, we ask the AI the question its customers are asking, such as "which agency should I choose for...", and its name appears nowhere.

An AI can pull a figure or a method published on your site, link to your page, and never write your company's name. Being cited as a source does not mean being recommended at the moment a customer makes their choice. Our GEO agency works on the gap between the two.

For this article, we reviewed published studies that analyze large volumes of AI responses and compared them with what we observe in our own content.

Key takeaways

• An AI can cite a page, mention a brand, list it among others, or recommend it.
• According to a Semrush study, 62% of citations in AI responses are not accompanied by any brand mention.
• AIs cite a brand's pages for its prices or features, and rely on third-party sources to say which one to choose.
• At least 92.7% of the brands recommended by ChatGPT appear in the content it cites.
• To be recommended, you must be present in these sources, have a consistent brand entity, and answer purchasing-related questions.

Cited, mentioned, recommended: what is the difference?

The following four levels do not have the same value for a company.

The same brand moves from one level to another depending on the question asked. For "how much does... cost," it will be cited for a price. For "which agency for...", it will be recommended or absent.

What do analyses of real-world responses show?

62% of citations are not accompanied by any mention

Semrush and Kevin Indig studied how AI engines cite their sources and name brands. In their sample, 62% of citations resulted in no brand mention, and only 13.2% of brands were both cited and mentioned. Google AI Overviews cites brand sources in 87% of cases and names a brand in only 20.7% of responses. (Semrush study)

The authors also note that well-known brands are mentioned more frequently, while comparison sites and directories are cited more often as sources.

ChatGPT mentions brands primarily in response to purchase-related queries

BrightEdge analyzed thousands of prompts: ChatGPT mentions brands 3.2 times more often than it cites them with a link. However, 44% of prompts contain no brand names, and commercial phrasing (e.g., "best," "where to buy") generates 4 to 8 times more mentions than informational queries. (BrightEdge analysis)

Thecommercial prompts therefore determine the recommendation. A brand that is highly visible in general queries may disappear as soon as the user asks what to choose.

The brand's website is used for facts, while choices are based on third-party sources

By observing what ChatGPT retrieves and displays, Suganthan Mohanadasan found that supplier pages are cited for their own information, such as pricing or specifications. When it comes to recommending the best offer, ChatGPT refers to a third-party source. (Suganthan Mohanadasan analysis)

A brand can therefore be considered reliable for describing its own offering without being recommended.

Almost all recommended brands appear in the cited pages

Visible compared the brands recommended by ChatGPT with those present in the content it cites: at least 92.7% of recommended brands appear there. A brand present in 6 or more cited URLs is 6 times more likely to be recommended than a brand present in only one. Furthermore, 96% of brands that appear for the first time in a cited page are not recommended. (Visible analysis)

Appearing in a single article makes almost no difference. You need to be featured in several third-party sources, and in a prominent position within those sources.

Without web search, major brands dominate

Part of ChatGPT's responses comes from its internal memory. Onely cites a University of Toronto study of 50 unbranded prompts: major brands received 56.3% of mentions, while niche brands received 12.3%. When ChatGPT performs a search, 57% of the pages cited in the Onely sample were the websites of the brands mentioned, and 22% were lists. (Onely study)

An SME or a specialized brand is therefore more likely to be recommended in answers generated from a web search if its pages, and the pages that mention it, answer the question precisely.

Why is a brand mentioned but not recommended?

Its pages answer general questions

An article that explains pricing or a methodology is good informational content. The AI picks it up for a figure. But the page doesn't tell the AI that you are the right provider for the person asking the question.

It does not appear in rankings and comparisons

If "best agency for..." articles and directories in your sector don't mention you, the AI has no external source to rely on to recommend you.

Its entity is poorly defined

When the name, offering, figures, or address vary from one site to another, the AI struggles to link your content to a single entity. It may cite a page without knowing which company to attribute it to.

Its content describes the company without adding any value

In our GEO Playbook, we explain that AIs rarely use introductory sentences like "X is a company that does Y." They use analyses, viewpoints, and proprietary data that answer a real question.

What we observe with Growth Room's content

We publish many informational articles: Meta Ads agency pricing, budget for a ChatGPT Ads test, budget allocation between Meta and Google. AIs use this type of page for facts. On these topics, our goal is to be cited, and then named as the source of the figure.

For a question like "which growth marketing agency to choose," the pages that matter are those where others talk about us. Impli ranks us among the best growth marketing agencies in 2026, Zeliq presents us as a benchmark agency for B2B SaaS, and we have profiles with customer reviews on La Fabrique du Net and Sortlist. These pages give an AI the elements to recommend us that our own articles cannot provide.

On our site, we apply what our GEO Playbookdescribes, downloaded by over 4,000 people: publishing figures from our campaigns, giving our opinion on our topics, and tagging pages with structured data. We are now tracking citations, mentions, and recommendations separately across a fixed panel of prompts, and we will publish the results.

How do you go from being mentioned to being recommended?

At the same time:

• harmonize your company name, offerings, and key figures across your website, LinkedIn, directories, and partner profiles;
• mark up your pages with structured data (Organization, Service, FAQ), following the priorities outlined in our article on structured data for AI;
• publish studies based on your own data;
• tailor your content to specific industries and cities, as users include their own context when asking questions.

How do you measure the gap between citation and recommendation?

If your dashboard GEO only tracks citations, it overestimates your visibility. We monitor four indicators across a fixed panel of prompts:

• the citation : the share of responses where one of your pages is included as a source;
• the mention : the share of responses where your name appears in the text;
• therecommendation rate : the share of responses where you are recommended, including your position;
• the sentiment associated with your brand.

The panel combines three types of prompts:

• choice-based questions: "what is the best agency...", "which provider for...";
• informational questions about your topics: "how much does... cost", "how to do...";
• questions about your brand: "what do you think of...", "X or Y for...", and a question about your boundaries.

Run each prompt on ChatGPT with search, Perplexity, and Gemini in a new conversation, and repeat the tests on a fixed date. We detail the method in our article for proving your content works on ChatGPT and Perplexity.