Brand mentions: why AI won’t recommend your site

A brand mention is defined as any reference to a company written by someone else: a Google review, a Reddit thread recommending it, a comparison in a trade publication or its Wikipedia article. What sets it apart from a link is that it does not need to point anywhere: the name only has to be there, saying what the company does.

For an AI assistant, those mentions do two jobs. They tell it the company exists and is known in its category, and they give it sentences to describe it with.

Auditing which mentions a brand already has, linked or not, is a separate job, covered in the audit of unlinked brand mentions.

When an assistant searches the web, it keeps few sources. In a DoGeo study of 1,230 Gemini runs (October 2026), when the model searched it used about 10 sources per answer. If those ten are comparisons, forums and reviews that do not name a company, that company’s site stays out of the answer, however good it is.

Often the assistant does not search at all. In the same DoGeo study, with the questions on AI visibility and GEO asked in Spanish, 70% of Gemini runs launched no web search; with the same questions in English, 46%. In those answers, whatever the assistant says about a brand comes from what the model already knows, and it learned that from text written by others.

How each assistant decides which sources to consult before citing is covered in how AI models choose their sources. What matters here is the consequence: a company competes to appear in the sources the assistant already uses, not only to have the best page.

BrightLocal’s Local Consumer Review Survey 2026, run on a panel of 1,002 US adults and published on 11 February 2026, found that use of ChatGPT and other AI tools for local business recommendations rose from 6% the year before to 45%. AI is now the third most used source of business recommendations, behind Google and Facebook.

The same BrightLocal survey found that 97% of consumers read reviews for local businesses, and that the average consumer uses six different review sites when choosing. Opinions are spread out, and an assistant summarising a company’s reputation has plenty to look at.

For a company, reviews are the third-party mention that is easiest to influence legitimately: they are requested from real customers and answered, but never written.

Google announced on 22 February 2024 that it now had access to Reddit’s Data API, which delivers real-time, structured content from Reddit. OpenAI announced an equivalent deal with Reddit on 16 May 2024, to bring Reddit content to ChatGPT. Google and ChatGPT therefore have a direct line to what is said on Reddit.

Reddit’s weight shows in citation counts. In May 2026, 5W Research published an analysis of about 600,000 ChatGPT citation events in the US, using Similarweb data from January and February 2026: Reddit was the second most cited domain, at 11.97%, behind only Wikipedia.

The same 5W Research report warns that Reddit’s share in ChatGPT swings a lot from one month to the next, so Reddit is not a channel to build a whole plan on. What does help is taking part under a real name in the industry’s forums; a thread faked by the brand is a bought mention.

Ahrefs analysed ChatGPT’s answers to 750 recommendation prompts in three categories: software, products and agencies (along the lines of “the best web design agencies in London”). “Best X” lists in blog format made up 43.8% of all page types cited, ahead of any other format.

The same Ahrefs study found that brands placed higher on those lists showed up more in ChatGPT’s recommendations than those placed in the middle or at the bottom. It is a correlation, not proof that moving up a list causes the recommendation, but it describes where assistants find their candidates.

For a company, the comparisons in its industry are a concrete target: knowing which ones exist, whether they include it and how they describe it. Paying for the top spot is buying a mention.

On 17 October 2025 the Wikimedia Foundation wrote: “Almost all large language models (LLMs) train on Wikipedia datasets.” In the same post it reported that human pageviews on Wikipedia had fallen by roughly 8% compared with the same months in 2024, and put it down to generative AI and social media, with search engines answering directly, often from Wikipedia content.

Wikipedia does not accept self-promotional articles. Its notability guideline presumes a topic merits an article when it has received significant coverage in reliable sources that are independent of the subject, and its conflict-of-interest guideline strongly discourages a company from editing the article about itself. For a company to get an article, other publications have to write about it first.

Wikidata, the structured knowledge base of the Wikimedia movement, collects data that supports Wikipedia. If the company already has a Wikidata entry, it is worth checking that it matches the company’s site.

1. Write one sentence that says what the company does and for whom, and use it unchanged on the website, business listings, LinkedIn and industry directories.
2. Ask real customers for reviews on the sites the industry relies on, and reply to all of them, including the bad ones. Reviews are the only third-party mention a company can ask for directly.
3. List the “best of” comparisons in its category that already appear in assistants’ answers, and write to their authors with verifiable information: what the company does, for whom and with what results.
4. Publish first-party data nobody else has, such as a study or a case with figures: it is what a journalist or a comparison quotes with the company’s name next to it.
5. Answer questions under a real name in the industry’s forums, without linking to the site in every message.

What an assistant writes about a brand can only be seen by asking it. For the measure to be useful, the prompts have to stay the same, be the ones a customer would ask, and be run several times: that shows whether the brand appears, where, and with which source. How to set it up is in the guide to tracking AI visibility.

What that prompt set does not tell you is whether the assistant has read the site. For that, a rule is created in the firewall or CDN that logs, without blocking, the requests from each AI robot, identified by its user-agent and, where the provider publishes its ranges, verified by IP. Those logs show which pages each robot visits and when.

Search robots, such as OpenAI’s OAI-SearchBot, Anthropic’s Claude-SearchBot or PerplexityBot, show which pages enter each assistant’s index. The ones that open a page when a person asks, such as ChatGPT-User, Claude-User or Perplexity-User, are the closest signal that the assistant has read it in a conversation. This is what DoGeo sets up for each client, alongside the fixed prompts.