Do AI assistants get American city level supplier questions right, and the four metros we checked
A measured answer for New York, Los Angeles, Chicago and Houston, what each city answer leaned on, and why a national question set does not predict a city one.
Published by AI Knows Us · Updated 22 October 2026
American city level supplier answers are not a smaller version of the national answer, and that is the single most useful thing a buyer or a supplier can know about them. We checked four metros on 15 and 16 October 2026, six questions per city put to two engines, twelve readings each. Forty three of those 48 readings returned a named source list. No two cities had the same leading source. In two of the four, the leading source was a national software platform's own agency directory rather than a firm in the city at all. So the answer to whether the assistants get city questions right is that they answer them with locally distinct sources and they do not transfer a national answer down to a metro, which means a supplier selling in four American cities faces four separate questions rather than one.
The four metros, as read
| Metro | Readings that returned a named source list | Most cited source in that city |
|---|---|---|
| New York | 9 of 12 | localseo.nyc, cited 3 times |
| Los Angeles | 11 of 12 | agencies.semrush.com, cited 4 times |
| Chicago | 12 of 12 | agencies.semrush.com, cited 3 times |
| Houston | 11 of 12 | houdigitalmarketing.com, cited 3 times |
Two cities answered with a city specific domain and two answered with a national platform's directory. A supplier tracking one national question set reports one picture and misses both of those facts, which is the practical reason the per location line is the one to negotiate hardest with any monitoring vendor.
What right and wrong mean here
There are two different questions behind the headline and they need separating. Does the assistant correctly understand that the buyer means a particular American metro? On our readings, yes, consistently: the source lists were locally distinct rather than national. Does the assistant name the best supplier in that metro? That is unanswerable as stated, because best is the buyer's criterion rather than a fact, and what can be checked is narrower and more useful. Is the supplier it named real, currently trading, and in that metro? Those three are checkable in minutes, and they are the right standard to hold an answer to.
The national pass, for contrast, and why the two do not mix
Separately, on 22 October 2026, we put a 198 question American buyer bank to the ChatGPT and Gemini product surfaces through a data provider, with the United States stated inside each question because the provider accepts gl, hl and location parameters and ignores all three. The brand batch is excluded as a ceiling measurement. That national pass gave 64 blind readings, 2,866 citations across 1,143 hosts, zero namings of us, 24 readings naming a rival and 38 naming nobody, with all 53 namings going to software products led by Semrush at 14 and HubSpot at 12.
These are two separate measurements on two separate banks and the figures must not be combined into one table. The national pass asked 198 questions of the whole country on one date. The city pass asked six questions per metro on two earlier dates. They answer different questions, and the only honest way to use them together is as two readings reported side by side with their own denominators.
What a supplier should do about city level answers
- Count your metros before you buy monitoring. Tracked questions, times engines, times refresh frequency, times cities. Four metros is four question sets, and the bill grows on the location line rather than the plan line.
- Get onto the city specific sources. Two of our four metros answered with a city domain. Those domains are reachable: a local directory, a local publication, a city association. That is concrete work with a short list.
- Claim the national directory profiles too. Two of the four metros answered with a national platform's agency directory, so the profile that wins a city answer may not be a local asset at all.
- Verify what the engines say about you, city by city. An answer that names you with the wrong service area or the wrong phone number is worse than one that names nobody.
How to run the city check yourself
- Pick the six questions a buyer in that metro would actually ask, naming the city in the text of each one.
- Put all six to two engines and save the answer text, the named firms and the source list, with the date.
- Check each named firm for three things. Real, trading, and actually in the metro.
- Count the sources. The leading source in that city is your first target and it will differ from the national one.
- Repeat per metro rather than assuming the first city generalises. On our evidence it does not.
The city readings behind this page are written up individually, and the New York one is the clearest example of a metro answering with its own sources, in two of the four winning sources are the city's own. The Chicago reading is the outlier where every reading returned a source and no single source dominated, in every reading returned a source and no source led.
Questions about American city level assistant answers
Do the assistants know American metros properly?
On our 48 city readings they answered with locally distinct source lists rather than national ones, and 43 of the 48 returned a named source list at all. That is a working understanding of the metro.
Can I rely on an assistant to find a supplier in my city?
As a shortlist generator, with verification. Check that each named firm is real, trading and in the metro before you call, because those three checks take minutes and the answer is not audited by anybody.
Why did a national directory win two of the four cities?
Because the engines read it and it carries city pages. That makes a complete, accurate profile on a national directory a city level asset, which is counterintuitive and cheap to act on.
Should I track every metro I sell in?
Do the multiplication first. If four metros means four question sets at your refresh frequency, that is the number to take into a pricing conversation rather than a plan name.
Can I combine the city figures with the national ones?
No. Different banks, different dates, different denominators. Report them side by side and never add them.
Four cities, four answers, and one reading that covers them
A national monitoring setup told a supplier in four American metros one story on our evidence, and the four cities told four. AI Knows Us reads ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews for the buying questions in a category and a location, reports each engine separately with its denominator, and keeps the raw answer text so a city claim can be checked against what was actually said. In the United States that sits on the Growth plan at 299 dollars a month with 100 buyer questions watched and 1,600 credits, the cheapest tier that reads all five engines, read on aiknowsus.com/pricing on 22 October 2026, and the question count is the line to size against your metro count rather than the plan name.