How we measure AI visibility.

The engine

Google AI Mode (google.com/search?udm=50), reached from a fresh, logged out Chrome session, not an in app browser and not a paid API. A fresh logged out session is the closest thing to what an ordinary buyer sees, with no account history pulling the answer toward a business they already know. The one exception is the citation source study, which used a logged in session, so we treat those two captures as different methods and never mix them in one figure.

The question

One question per city, "best [trade] in [City, State]". The same fixed template runs in every industry, so the numbers in one index can be read against another. Each city also carries nine more buyer questions (cost, emergency, financing, near me and so on) recorded for later work, but the named list on an index comes from that one question, asked the same way in every city.

What named means

A business counts as named only when the answer's own text said its name. We never asked the AI for a list of five, never supplied candidates, and never filtered or reordered what it said. The order in the data is the order the answer gave. It is a transcript, not a leaderboard we built, and we do not claim the order means anything.

The date

Every answer carries the date it was captured and every report prints it. An answer that reads differently today does not make the data wrong. It means the answer moved, which is the whole point of publishing a dated snapshot instead of a standing claim.

What it reads

The homepage HTML of businesses found while sourcing each city sample: schema.org markup, CRM and quote widgets, tracking pixels, review counts, certification wording, financing, services and footer year. A tool that never appears in the HTML, because it only loads after a click, is not detected. Every share we publish from this crawl is therefore a floor, not a ceiling.

Named against everyone else

Sites are split into the businesses their own city's answer named and the ones it did not. Every comparison between those two groups is a same day reading, never a claim of cause. It shows what the named businesses happened to have on their sites that day. It does not show what got them named.

Blocked and failed sites

Every report prints how many sites were tried, read, blocked by the site itself, and silent. A share is always out of the sites actually read, and that base is printed next to it.

GREEN, the answer is open

The answer names a short list of businesses without calling any one of them the single best or the top choice. Somebody new can still join that list.

YELLOW, one business is ahead

The answer still names several businesses, but its own wording lifts one of them above the rest, for example by calling it the one widely considered the best.

RED, the answer is closed

The answer effectively shuts the field, either by naming one business only or by refusing to name anybody at all.

NOT SCREENED

A city we have not asked about yet. It never gets a guessed tier and never borrows one from a neighbouring city. The live checkers on our reports say so plainly and offer a free check instead.

• Not a ranking. We never rank the businesses an answer names against each other, and no sort on any table on this site reorders them by quality.
• Not an endorsement of a business that appears, and not a criticism of one that does not. A company can do excellent work and still not be the answer AI gave on the day we asked.
• Not a random sample of a whole industry. The cities in each run are the markets that study was built around, and they lean toward small and mid size cities rather than the largest metros.
• Not permanent. Answers change. Each report carries a last updated stamp and a list of what changed, and we re-screen rather than leave an old figure standing.

1. 1

   Open a fresh, logged out browser session.

2. 2

   Ask Google AI Mode, or any AI assistant, the exact phrase "best [trade] in [City, State]" for any city in one of our reports.

3. 3

   Compare the businesses the answer names today against that city's row in the report, and the date the row was captured. A different answer today does not mean the data is wrong. It means the answer moved.

That is how we measure AI visibility, and it is the same run we do for one business in the free check, before any money changes hands.