Original measurement

We asked ten buyers' questions from three places. Thirty answers, thirty companies named, and not one company named in all three.

The same ten questions a business owner would type, asked once from Serbia, once from Austin and once from Miami, on the same assistant within two days. Serbia and Austin shared 1 company of 23. Austin and Miami shared 1 of 19. Serbia and Miami shared none at all. Inside Miami one company – AEOGeoAI – was named in 8 of the 10 answers. The list an assistant hands a buyer is rebuilt per place, and inside a place it can be held by one name.

companies named in all three places, out of the 30 named in any of them

Key findings

Asked the same ten buyer questions from three places on the same assistant, not one of the 30 companies named was named in all three places.
Serbia and Austin shared 1 company of 23 named between them; Austin and Miami shared 1 of 19; Serbia and Miami shared 0 of 20.
Inside one metro the answer can concentrate: in Miami, AEOGeoAI was named in 8 of the 10 answers, while the most-named company in Austin appeared in 3.
A question with no city in it is still answered for a place: the ten readings taken from Serbia with no city in the question named 12 companies, and 11 of those 12 were named in neither Austin nor Miami.

How this was measured

When: Thirty readings taken 30 and 31 August 2026. One run per place, no repeats.
Instrument: Our own test setup: a real Chrome browser, controlled by a script, one question per fresh conversation, the full answer text stored verbatim.
Surface: Logged-in ChatGPT, one account. Serbia is the browser's own location with no city in the question; Austin and Miami are named in the question text.
Sample: 10 questions x 3 places = 30 answers. The ten questions are the buyer-voice set in buyer_queries_us.json, asked word for word in each place.
Counting: Every name the answers listed was classified once, by hand, as a company or as a metric, a task or a category phrase, and that classification is published as data (semantic/buyer_vocab.json). A company counts for a place when its name appears in at least one of that place's ten answers.

What we asked, and why three places

Every other measurement we publish asks about a local business and reads which businesses come back. This one turns the instrument around and asks the question the owner asks: who can help me get named by AI. If the answer to that is the same everywhere, a single national page is the thing to build. If it is not, the unit is the metro, and that changes what a company in this market should build first.

So the same ten questions were asked three times: once with no city in them at all, from a browser in Serbia; once naming Austin, Texas; once naming Miami. Same assistant, same account, two days apart at most.

The answer changes with the place, almost completely

Thirty distinct companies were named across the thirty answers. Serbia and Austin shared one of the 23 named between them. Austin and Miami shared one of 19 – InterCore Technologies, once on each side. Serbia and Miami shared none of their 20.

Not one company was named in all three places. A buyer in Miami and a buyer in Austin asking the same words on the same day are handed two different markets.

This is the sharpest version of a result we keep meeting: the named set is not a ranking that a company holds. It is rebuilt for the question, for the reading, and – here – for the place the question is asked from.

Inside a place, one name can hold the answer

Miami's ten answers named eight companies between them, and one of those eight, AEOGeoAI, appeared in eight of the ten. The next was The AD Leaf, in four.

Austin was flatter: twelve companies, the most frequent of them – NSYGHT and Signal Digital – in three answers each. Serbia flatter still: twelve companies and nothing above two.

Concentration tells you what kind of place you are entering. Where one name holds eight of ten answers, the answer has an incumbent; where nothing is above three, it does not. Neither shape is good or bad on its own, but they are different problems, and a company deciding where to compete should know which it is looking at.

A question with no city in it is still answered for a place

The Serbian run named no city at all. It did not produce a neutral, worldwide answer: it produced a third set of companies, different again from the other two.

That is worth stating on its own, because it is easy to read a location-free question as a location-free answer. The answers changed with the place the question was asked from, which suggests the assistant fills in the missing location itself; we did not measure how. Either way, there is no such thing as asking this question from nowhere.

What we are not claiming

We measured that the named set changes with the place. We did not measure why, and the honest candidate is off our own site entirely: the companies that hold a place tend to name that metro in their own brand line and to appear in local directories. Whether a page on our own site can move this is a separate question, pre-registered and being measured now, and this page will be updated with the answer whichever way it comes back.

One run per place is one draw. Our own stability measurement found that two readings of the same question ten minutes apart can share no names at all, so the size of an overlap between two places should be read as small, not as exactly one.

What this means if you are buying

If you sell to local businesses, the market you are competing in is a metro, not a country. A single national page is competing in a different answer from the one your buyer sees.
A company that holds a place holds it for that place only. Being named in Miami says nothing about being named in Austin, and a vendor's screenshot of one answer says nothing about the market you are in.
Anyone measuring this has to record where the question was asked from. A reading with no place recorded is not comparable with anything.

Honest limits of this measurement

One run per place, ten questions, no repeats. Against our own 0% ten-minute overlap finding, a single run cannot separate 'this place is different' from 'this reading is different' – the size of each overlap is soft; that not one company was named in all three is the robust part.
The model tier drifted between runs: Serbia five readings on the full model and five on the mini, Austin two and eight, Miami none and ten. Part of Miami's concentration may be the smaller model returning shorter lists.
The company vocabulary is a floor. Only names the answers put in a list were extracted and classified; a company mentioned inside a sentence and never listed is not counted anywhere.
Three places, one assistant, one account, two days. Nothing here is a claim about other assistants or about how long any of it lasts.

Questions about this research

Does this mean I need a page for every city?

It means the answer your buyer sees is built per place, so a page that names no place is competing in none of them. It does not mean building hundreds of city pages: we cap our own at three metros by ten trades until the first ten of those pages are measured, because a page nobody can distinguish from 300 others is a different problem.

One company was named in eight of ten Miami answers. Can I buy that position?

Not from us, and we would not believe anyone who offered it. What that company has is an off-site footprint: the metro in its own brand line and listings that describe it as operating there. That is a position earned across other people's pages, not a slot for sale.

Why is the Serbian run in here at all?

Because it is the control that shows a question with no city still gets a located answer. Without it, the Austin-Miami difference could be read as an effect of naming a city in the question rather than of where the buyer is.

How do I check this myself?

Ask any of the ten questions from your own browser, then ask a friend in another city to ask the same words. Compare the company names. That is the whole method; ours differs only in that every answer is stored verbatim and the classification of what counts as a company is published.

We run this measurement for a living. If you want it run on your own category and your own city, the free scan is the same instrument on one question, and the report is the full set. Nothing here is behind a form.

More measurements from the same instrument

Every category we measure has its own page, with the questions customers put to an assistant in that category and what a reading costs. See all industries we measure →

Published 20 September 2026, updated 4 October 2026. Every figure on this page is recomputed from raw observation files by a single script, so any of them can be traced back to the readings behind it. If you are named here and believe a reading is wrong, tell us: we will run the measurement again and publish what it returns, whichever way it goes.