A Semrush study analysed by Kevin Indig and published on 5 August 2026 on Search Engine Land examined 1,094 AI search categories on ChatGPT in the United States, from January to June 2026. Its finding: a brand stands a good chance of being cited as a source almost everywhere, but AI only recommends it by name on its core business. For a GEO and SEO agency in Luxembourg, this is numerical confirmation that AI visibility is won through depth rather than through spread.
The facts
- In categories close to its core business, a brand is cited in 74% of appearances and explicitly named in 44% of cases; outside its core business, citation stays at 50% but named mentions fall to 25% (Search Engine Land).
- 89.3% of estimated AI search demand sits in categories with no owning brand: in June 2026, only 15.2% of the 1,094 categories analysed had a clear leader (Growth Memo).
- According to a Writesonic study covering around 16 million brand appearances, around 40% of AI citations do not name the source brand in the answer, rising to 52% on Perplexity (Search Engine Land).
Why is an AI citation worth less than a recommendation?
The difference comes down to two distinct mechanisms. A citation proves that a piece of content served as a source; a mention is what the reader actually sees in the answer. Between the two, the loss is massive and varies widely from one engine to another: around 19% of citations without a brand mention on Microsoft Copilot, 25% on Gemini, 37% on ChatGPT and up to 52% on Perplexity. Some engines link a lot but name little, others do the reverse. An aggregate AI visibility score therefore makes little sense: citations, brand mentions and what they produce downstream need to be tracked separately.
The study also brings a nuance that changes the strategy: covering many subjects leaves citations intact, while a shallow presence weakens mentions. A brand present on all five typical queries of a category gains recognition; present on only one, it slips back, even when cited. And once established as the named reference, a brand keeps its top position in 90.4% of month-on-month observations. These figures remain associations measured on ChatGPT in the United States, with no proof of causality, but they align with what we wrote about brand memory in AI models: awareness has become a signal in its own right.
AIxH's view
The window is open: nearly nine AI queries in ten are currently decided in categories with no owning brand. The method we apply as a GEO agency in Luxembourg follows the logic of these studies: identify 10 to 20 subjects genuinely tied to the core business, build a deep presence there rather than shallow coverage, and measure citations and named mentions separately, engine by engine. The foundation is unchanged: this work rests on the organic search we deliver as an SEO agency in Luxembourg, because the pages AI cites are still well-built pages. An audit is enough to find out whether your brand is cited, named, or simply absent.