AI models searched for brands they already knew 3.2 times more often than unfamiliar brands, according to a new study from geoSurge.
| Group | Search rate |
|---|---|
| Familiar brands | 55.7% |
| Brands outside the top 10 | 17.4% |
Memory strongly decided which companies were considered.
Researchers measured model memory and search behavior separately to see whether what a model knew influenced what it searched for.
Across nearly 4,000 responses to 66 U.S. buyer questions, models searched for familiar brands 3.2 times more often than unfamiliar ones.
| Measure | Share |
|---|---|
| Fan-out searches including a company name | 31% |
| Of brand-specific searches, those involving one of the model's five most familiar brands | 63% |
Models searched for familiar brands 41% to 82% of the time, compared with 9% to 23% for unfamiliar brands.
Researchers noted some industries were represented by as few as six prompts. The industries analysed were travel, automotive, finance, business software, education, food and restaurants, luxury, fitness and wellness, and fashion.
In one example, Gemini searched for Lemon Squeezy while answering a question about online payment providers, even though the brand wasn't in its measured memory.
This shows live search can still find unfamiliar brands, especially in categories where models rely less on what they already know.
AI models are more likely to search for brands they already know. This means a brand may have an advantage before the model even starts searching. Strong content can still help newer brands get noticed, but familiar brands appear to have an unfair advantage.
Researchers tested 66 U.S. buyer prompts 60 times between May 29 and June 9.
| Item | Volume |
|---|---|
| Responses analysed | 3,960 |
| Fan-out searches | 13,281 |
| Brand-level observations | 1,416 |
The authors said the results show a relationship between memory and search behavior, but don't prove that one caused the other.
The study: Report: AI Searches What It Remembers
Don't read this as "make more content." The finding is that the field is already tilted before searching begins — not a layer that page optimisation reaches.
Measure being remembered separately from being searched. That separation is why the study measured both — start by establishing whether your brand is in model memory at all.
Treat the 31% as the opportunity. 69% of fan-out searches contain no brand name, meaning most searching is category- and problem-led, where the familiarity gap bites less.
Adjust expectations using the 41–82% industry spread — while noting some industries were represented by as few as six prompts. Check sample size before trusting your own vertical's number.
Read the Lemon Squeezy case as a route, not an exception. Brands absent from memory can still be found by live search, particularly in categories where models rely less on prior knowledge.
Assume causation is unproven. The authors state this limitation explicitly — concluding "get into memory and you get searched" overreads a correlation.
Check entity recognition first. Being remembered requires being recognised as one clear entity — see Entity SEO and AI Search Can't Verify Your Business.
Build memory through earned media. That 82% of AI citations point to earned media is covered in 82% of AI Citations Point to Earned Media — your own site alone rarely gets you into model memory.
3.2 times more often, per geoSurge's study — 55.7% for familiar brands versus 17.4% for brands outside the model's top 10.
No. Only 31% of fan-out searches included a company name. But when a specific brand was searched, 63% of those involved one of the model's five most familiar brands.
Familiar brand search rates ranged from 41% to 82%, unfamiliar from 9% to 23%, across travel, automotive, finance, business software, education, food and restaurants, luxury, fitness and wellness, and fashion — with some industries represented by as few as six prompts.
Yes. Gemini searched for Lemon Squeezy — absent from its measured memory — while answering a question about online payment providers. Live search can still surface unfamiliar brands, especially where models rely less on prior knowledge.
66 U.S. buyer prompts tested 60 times between May 29 and June 9, yielding 3,960 responses, 13,281 fan-out searches and 1,416 brand-level observations. The authors say the results show a relationship between memory and search behaviour but do not prove causation.
To apply what you just read to your own site, start with a free audit of where things are now.
A strategist replies within 24 hours on business days.
An E-E-A-T checker was never automatable because there is no API for it. Here is how an AI coding agent read Google's quality guidance, crawled a page sample, and produced a scored audit — plus where it stops.
A GEO audit is not checking whether your brand appears in an AI answer. Here is a nine-point GEO checklist split into four on-site items, four off-site, and one measurement layer.
LinkedIn, YouTube, Reddit, and TikTok are building systems to identify low-value AI content. Here are the limits of watermarking and what content still gets distributed.