What 126 Million AI Prompts Reveal About Which Brands Get Recommended

One of the most important lessons is that a brand mention and a website citation measure different forms of visibility. Semrush defines a mention as the brand name appearing within the generated answer. A citation is a website or page used as a source to support that answer. Share of Voice goes further by considering how often a brand is mentioned and where it appears within the response. A business named first and described in detail has a different level of visibility from one placed at the end of a long list, even if both technically count as mentions.

A recommendation is more commercially meaningful still. It occurs when the AI platform presents the company, product or service as a suitable choice for the user’s particular requirements. A business may be cited without being named, named without receiving a link or included in an answer without being positively recommended. This matters because many AI visibility reports combine these outcomes into one percentage. That can make performance look healthier than it is. A B2B consultancy does not simply want its article used to define an industry term. It wants the company to appear when a potential customer asks which providers have the right expertise, location, experience or service model.

The research found substantial differences between the way major platforms select and display sources. Semrush reported that ChatGPT cited an average of 15 sources per answer and often relied on community and reference websites such as Reddit and Wikipedia. Gemini cited an average of three sources and drew from a smaller group that included Wikipedia, Reddit and YouTube. Semrush also found that the overlap between brands mentioned and websites cited by Gemini could be as low as 30%. A platform may therefore rely on one organisation’s content while recommending a completely different brand.

This helps explain why strong visibility in one AI environment does not guarantee equivalent performance elsewhere. ChatGPT may use a company’s detailed guide as supporting evidence, while Gemini may name a better-known competitor based on broader recognition across the web. Google AI Mode and AI Overviews operate within Google’s search ecosystem but can still produce different formats, source combinations and customer journeys. Businesses should not refer to themselves as “visible in AI” based on one successful ChatGPT response. They need to understand which systems matter to their audience and how consistently the brand appears across each one.

A separate Semrush study provides an even clearer warning against treating every citation as a recommendation. The research examined 3,981 domain appearances across 115 prompts, 14 countries and four AI search engines. It found that 61.7% were ghost citations, where the platform linked to a website but never named the associated brand in the answer. Only 13.2% of appearances included both a citation and a brand mention, while 25.1% mentioned the brand without providing a source link. Overall, citations appeared in 74.9% of cases, but brand names appeared in only 38.3%.

For a publisher or research organisation, a ghost citation may still have value because the link can generate referral traffic and demonstrate that the content is being used as evidence. For a service business trying to become better known, however, the outcome is less valuable. A potential customer may read an answer built partly from the company’s expertise without ever learning that the company exists. This is why businesses should track citations and mentions separately. Improving AI visibility is not only about publishing information that systems can retrieve. It is also about building a clear enough brand association for the organisation to be recognised and named.

A combined AI visibility score can be useful for benchmarking, but it can also conceal where a brand is genuinely strong or weak. A company may receive a respectable overall result because its content is cited frequently for broad informational questions, while remaining absent from comparison prompts such as “best”, “recommended” or “which company should I choose?” It may perform well in ChatGPT but poorly in Gemini, or appear consistently in US-focused answers while receiving little recognition in the UK market. Without separating these results, the business cannot tell whether visibility is likely to influence enquiries.

The Semrush research also found major differences in how concentrated visibility was across industries. The three leading brands accounted for 82.9% of visibility in News and Media and 76.9% in Consumer Electronics. By comparison, the top three represented 41.4% in Finance and 42.2% in Industrial markets. Some sectors are therefore dominated by a small number of well-established organisations, while others leave more room for challengers. A useful benchmark must reflect the competitive structure of the relevant market rather than comparing an SME with global brands across unrelated categories.

The study shows that AI visibility is not created by a company website alone. Platforms interpret brands through a combination of owned pages, third-party publications, community discussions, reviews, retailers, directories and reference websites. A clear service page may help explain what the company offers, but external sources can provide the corroboration that makes the claim more credible. When a business is described consistently across its website, media coverage, industry profiles and customer evidence, AI systems have a stronger foundation from which to form a confident answer.

This does not mean businesses should chase large volumes of artificial mentions or attempt to place themselves in every directory. The quality and relevance of the evidence matter. A solicitor may benefit from clear regulator information, respected legal directories, verified reviews and authoritative local coverage. A cybersecurity company may be strengthened by technical research, client case studies, specialist publications and credible industry partnerships. The objective is to create a coherent digital record. If a company describes itself differently across several sources, AI platforms may produce generic, outdated or contradictory summaries.

The main 126 million-prompt dataset is US-focused, so UK organisations should be cautious about applying its exact rankings or market patterns to British customers. The most useful lesson is methodological. Businesses need to monitor the prompts that reflect their own services, locations, customer types and buying journeys. A London recruitment agency should test questions about particular sectors and hiring needs. A Leeds technology consultancy should assess how platforms respond to local and national questions, rather than assuming a broad software visibility score represents its position.

The smaller ghost-citation study also found that brand mention rates varied by country, with the UK sitting in the upper-middle of the markets examined at 41%. Prompt style made a substantial difference too. Short conversational questions produced far more brand mentions than longer structured prompts, while comparison-led questions produced more mentions than informational ones. This reinforces the need for a varied prompt set. A business should test educational questions, local searches, comparisons, recommendations and detailed buyer scenarios because each can produce a different type of visibility.

The first step is to establish a reliable baseline. Businesses should identify the questions customers ask before selecting a provider and test them across the relevant AI platforms. Each result should record whether the company was mentioned, whether its website was cited, where it appeared, how it was described, which competitors were included and which sources supported the answer. The same prompts should then be repeated over time. This provides a much clearer picture than occasionally asking ChatGPT for a recommendation and treating one answer as a permanent result.

The second step is to strengthen the evidence AI platforms can find. Priority service pages should clearly explain what the company does, who it helps, where it operates and why it is credible. Important claims should be supported through genuine reviews, case studies, expert content, reputable directories, media coverage and relevant third-party references. Businesses should then compare their evidence with the organisations already being recommended. The purpose is not to imitate competitors, but to identify why AI systems appear more confident describing and selecting them.

The research does not reveal one secret factor that makes a brand the automatic recommendation. Instead, it shows that AI visibility is fragmented across platforms, prompts and types of appearance. Being cited is not the same as being named. Being named is not the same as being recommended. An overall score can provide a useful headline, but it cannot explain whether a business is visible at the moments most likely to influence a customer.

The brands best positioned to succeed will be those that are easy to understand, consistently represented and supported by credible evidence across the wider web. They will also measure performance at the level where decisions are actually made: individual platforms, priority prompts, competitor comparisons and customer locations.

AwarenessAI’s cross-platform monitoring and optimisation plans help businesses track priority prompts, compare competitors, analyse the sources influencing AI answers and identify practical opportunities to improve how they are understood, cited and recommended.