Between 29 May and 9 June 2026, a geoSurge study measured nearly 4,000 AI model answers across nine sectors: a brand in the model's memory is searched 3.2 times more often than a brand it does not know. Ranking well is no longer enough if the model has never encountered the brand.
The facts
- Across 1,416 brand observations in nine sectors, a brand in the model's memory was searched in 55.7% of cases, against 17.4% for a brand absent from that memory.
- Only 31% of the searches generated by the models name a specific brand; among those, 63% cite one of the five best-remembered brands.
- Over two years, Google search traffic to small publishers fell by 60%, against 22% for large publishers, according to Axios.
Why does the model's memory matter more than the page itself?
A generative model does not read the web for every question. It starts from a list of brands already associated with the topic, then runs searches to confirm it. geoSurge observed this mechanism on a split-payment case: of six searches triggered by a model, four targeted the four brands it already knew, in the exact order of its memory. A brand absent from that list does more than start late. It often starts out of the race.
The problem has nothing to do with content quality. It concerns what happens before the content is read. A well-structured page stays invisible if the model has never associated the brand with its sector, an association built through press coverage, partnerships and repeated presence in comparative content. The resulting loss of traffic is not neutral: it hits the least-remembered players first, the ones who most needed the click to get known.
AIxH's view
This shift is at the heart of AIxH's SEO & GEO (GSO) service: measuring where a brand is already remembered by the models, before discussing the content to produce. If you want to know where your brand sits in that memory, an AI visibility audit gives a first concrete answer.