AI Local Visibility Index

Ask an assistant for the best dentist in your city and it names two or three businesses. Google Maps returns a hundred. This study measures the gap between those two numbers, in one direction only: of the businesses Google is willing to show, how many does an AI ever say out loud?

Across 48 markets the answer is 11.94%. Roughly one business in eight. The other seven exist, rank, collect reviews, and never enter the conversation.

How we measured

The rules below were written down and frozen before any data was collected. They did not change after we saw the result. One amendment was made mid-study and is recorded with its reason and its date.

The question

Of the businesses Google Maps returns for a buyer query, what share does an AI assistant name when asked the same thing? Not “what share of all local businesses” — that denominator is unknowable without a registry, and a number whose denominator is unstated cannot be checked.

Markets

8 mid-size US cities × 6 categories = 48 markets. Austin, Denver, Portland, Nashville, Columbus, Raleigh, Tucson and Boise; dentist, gym, plumber, coffee shop, law firm and hair salon.

Denominator

One Google Maps query per market (“category in city”, pinned to the city coordinate, depth 100), deduplicated by name. 4,657 businesses in total.

Numerator

Three fixed question phrasings per market, asked to four engines. Three samples per question for the chat models, which are not deterministic; one for AI Overviews, which is a search result and repeats itself. 1,440 answers, 0 dropped.

What counts as named

The business name has to appear in the answer as a phrase. Named once, in any sample, counts: the question is whether a business is reachable through AI at all.

Cost

$1.40 of API spend for the published pass, under a hard cap that stops the collector before it can overspend.

The complete dataset, including every business name and which engine named it, is published as a single JSON file. Raw model answers are not republished: they contain the models' own evaluative claims about named real businesses, and we are not willing to amplify statements we cannot verify. They are available on request for methodology review.

What this does not tell you

• The denominator is capped. Google Maps was queried at depth 100 and several markets came back with exactly 100 rows, meaning the real market is larger than what we counted. The share we publish is therefore an upper bound against “all local businesses”, and it is not comparable to third-party figures that use a wider, unstated denominator.
• One country, one language. US cities, English questions. It says nothing about other markets.
• One day. A single collection pass. No trend claim is possible from it.
• Google Maps is not a registry. It is itself a ranked, filtered surface. The denominator is “what Google shows”, not “what exists”.
• Cheap model tiers. Each engine was queried at the tier we run in production. A larger model may answer differently.