Our own studies · Original study · Medellín, August 2026
The first time we measured where an AI's recommendations come from when someone looks for a local business. The answer is in the sources it cites — and they do not include Google Maps.
Four categories (bakeries, coffee shops, barbershops, veterinary clinics). 22 sources classified, 20 businesses named by the AI. Cost of the study: USD 0.14.
Not one of the 22 sources the AI retrieved was Google Maps. Not one. These are the pages the assistant actually opened to build its answer, and it is the finding that matters: the place where a local business's entire competitive advantage lives is exactly the one the AI does not consult.
Classified by kind, they broke down like this:
The right-hand column is a direct transcript of what the assistant answered. The left is a partial sample of businesses in the category with their verified review counts — useful for order of magnitude, not as a city ranking. For that, see the Bogotá study, which sweeps the whole category.
A Medellín veterinary clinic published a page on its own site titled "the best vet in Medellín". The AI read it and repeated it. It isn't a trick: the bar for being cited is still as low as publishing a decent page about yourself.
Of the 100 businesses we had profiled in Medellín with verified Google data, 1 appeared in the AI's answers — and it got there through an article on its own blog. The other 99 are invisible at the exact moment a potential customer asks an AI about their category.
What about your business? Visquore runs this same comparison for one specific business: your real position on Google, your verified competitors, and whether AI assistants name you.
Run my audit →An assistant with web search answers from what it finds written on indexable pages, and Google Maps is not one of them in practice: of the 22 sources it cited in this measurement, none was. Reviews decide your position on the map; the pages that talk about you decide whether an AI names you.
With one query per category to the official Places API, returning ten relevance-ordered results, shown here with their verified review counts. It is a sample of the category and should be read that way: ten results relevant to one phrasing are not the ten most-reviewed businesses in a city. The Bogotá study uses several phrasings per category, pagination and zones to cover it fully.
We compared Google's real ranking — 1,147 verified businesses across four categories — with the businesses an AI assistant actually names. They are almost never the same.