Ask an AI assistant to recommend something in your category and you will often get the same handful of names. If yours is not among them, the instinct is to assume your content is not good enough. New research points somewhere less comfortable: models tend to go looking for brands they already recognise, before anyone’s content is weighed at all.
That changes where the work sits. Here is what the research found, what it does not establish, and what a lesser-known brand can realistically do.
What Did the Research Find About AI and Familiar Brands?
Models searched for brands they already knew far more often than brands they did not.
In a recent study, researchers tested 66 U.S. buyer prompts, running each prompt 60 times over a multi-day period. That produced 3,960 responses, 13,281 fan-out searches and 1,416 brand observations. The headline findings:
This is third-party research into how models behave. It is not a statement from Google, OpenAI or any other model provider, and it describes US buyer prompts over a twelve-day window rather than search everywhere.
Why Would an AI Model Reach for a Brand It Already Knows?
Because a model answers from what it has already absorbed about a category, and widely-covered brands are simply more available to it.
When an assistant handles a commercial question, it often breaks the request into several narrower searches — the study calls these fan-out searches — and decides what to look for. A brand that appears consistently across the material a model has learned from is more likely to be one of the things it thinks to look for. Recognition acts as a shortlist before the shortlist.
That is a description of observed behaviour, not a published ranking system. No model provider documents how brand recognition feeds retrieval, and anyone claiming precise mechanics is guessing.
Does This Mean Smaller Brands Cannot Get Cited?
No — and the geoSurge’s study is the reason to say so confidently.
Only 31% of the searches the model ran named a brand at all. The other 69% went after a product, a problem or a category, which leaves that ground open to whoever answers that need best.
The researchers highlight one case where the model searched for a brand it did not already know. Asked which payment provider a startup should use, Gemini 3.5 Flash first searched the three names it knew best in that category — Stripe, PayPal and Square. It then ran a fourth search, for Lemon Squeezy, a brand that had never appeared in its memory at all. The study’s full write up notes that this happened in finance, one of the categories where familiar brands dominated the searches most heavily. Where the model knew a category less well, brands outside its memory appeared more often. As the researchers put it: “Memory is the more reliable way in, but it isn’t the only one.”
Familiarity is an advantage, not a gate. It is worth being precise about what the research supports, because the honest version is more useful than the fatalistic one:
For lesser-known brands, the key takeaway is that there is still room to compete. The opportunity lies in the unbranded majority of queries, where the question is what to buy rather than who to buy it from.
What Actually Builds the Familiarity Models Respond To?
Familiarity is built through consistent visibility across the web — not through a single page, platform or optimization tactic.
The study did not test how brands can build that familiarity, so the points below are our interpretation of what its findings mean in practice. If repeated exposure helps a brand become recognizable to models, then the practical implication is to build a stronger presence across the wider web:
Taken together this is the case for generative engine optimization (GEO): treating your presence across the whole web as the asset, rather than optimising a page and hoping. We view AEO and GEO as relatively new components of an overall Search Engine Optimization (SEO) strategy, not independent of one another. It is widely considered that a solid “old school” SEO strategy and implementation are good foundations for this new era of AI Search.
How Long Does Building That Familiarity Take?
Longer than a campaign, and nobody can give you an honest number.
Recognition compounds. It accumulates through repeated presence across many sources, and no research we have seen measures how long that takes to register with a model — nor would a single figure travel across categories, budgets and starting positions. Treat anyone offering a fixed timeline with suspicion.
What can be said is that the work is cumulative rather than switchable, which makes starting sooner worth more than starting bigger.
How Do You Tell Whether It Is Working?
Start with a baseline and then track whether your brand gets named for your category over time, and watch what happens to demand.
Ask the assistants the questions your buyers ask, at a regular interval, and record whether you appear. Run the same prompts each time so the comparison means something. Alongside that, watch branded search and direct traffic — if recognition is genuinely growing, people start arriving having already heard of you. Mentions are the leading indicator but enquiries are the one that matters. Make sure that your analytics are set up to measure traffic sources as granularly as possible.
Build the Recognition That Compounds
The uncomfortable reading of this research is that AI assistants are reflecting an advantage the market already handed to the established names. The useful reading is that the advantage is recognition, and recognition is buildable — slowly, through presence and substance, in the same places it has always been built.
Most queries still do not name a brand. That is the opening, and it belongs to whoever answers the question best and appears often enough to be worth checking. If you want a clear-eyed view of where your brand currently shows up in AI answers and what would move it, contact us today.
