Do AI assistants favour local suppliers inside the United States, and what our own readings show
A measured no at national level, a measured yes at metro level, and the published evidence that explains why the two readings disagree.
Published by AI Knows Us · Updated 22 October 2026
At national level, no. At metro level, partly. That is the honest shape of the answer and the two halves come from two separate readings, which we report side by side and never combine. In 62 blind national American readings on 22 October 2026, ChatGPT and Gemini named zero American agencies and named no local firms of any kind, while naming software products 48 times. In 48 city level readings across four metros on 15 and 16 October 2026, two of the four cities answered with a city specific domain as the leading source. So the engines are capable of locality and they exercise it when a city is named in the question and local evidence exists, and they default to national software when neither of those conditions holds.
The two readings, side by side
| National pass, 22 October 2026 | Four metro pass, 15 and 16 October 2026 | |
|---|---|---|
| Bank | 198 American buying questions in 33 batches | 6 questions per city, 4 cities |
| Readings, brand batch excluded | 62 blind, 32 ChatGPT and 30 Gemini | 48, 12 per city |
| Local firms named | 0 | City domains led 2 of the 4 cities |
| What led instead | Software products, 53 namings | A national platform's agency directory led the other 2 cities |
| Citations | 2,866 URLs across 1,143 hosts | 43 of 48 readings returned a named source list |
Different banks, different dates, different denominators. The figures in the two columns answer different questions and adding them would produce a number that means nothing.
The metro detail, because it is the actionable half
- New York, 9 of 12 readings returned a named source list, led by localseo.nyc with 3 citations.
- Los Angeles, 11 of 12, led by agencies.semrush.com with 4.
- Chicago, 12 of 12, led by agencies.semrush.com with 3.
- Houston, 11 of 12, led by houdigitalmarketing.com with 3.
No two cities shared a leading source. Two were local domains and two were a national software platform's directory. For a supplier, the practical consequence is that winning a city answer sometimes means a local asset and sometimes means a profile on a national directory, and which one it is has to be read per city rather than assumed.
Why the national pass named no local firms, and what fixes it
The explanation is in what the engines cited. Across the 62 national readings the top hosts were developers.google.com in 17 readings, support.google.com in 15, searchengineland.com in 13, blog.google in 11, clutch.co in 11, semrush.com in 10, ftc.gov in 9 and blog.hubspot.com in 9, with brightlocal.com and searchenginejournal.com at 7 and goodfirms.co at 6. One entry in the raw data, ccn.web.tr at 8 readings, is the data provider's own fetch infrastructure rather than a publisher and is excluded from source analysis.
That is a list of documentation, trade press, directories, vendor properties and the regulator. It is not a list of local firms, because local firms in the American market publish claims rather than checkable, dated, specific material. The same bank run for Canada on the same day named six Canadian agencies 21 times between them, which demonstrates that the engines will name local firms where the local published evidence supports it. The difference is the evidence, not the engine.
What a local American supplier should do about it
- Name your city in your own published answers. An engine answering a city question needs a document that says the city. Service area pages that name the metro, the neighbourhoods and the drive time do that work.
- Claim the national directory profiles. Two of four metros answered with a national platform's agency directory, which makes a complete profile there a local asset.
- Find the local source that leads your metro. On our readings that was a city specific domain in two of four cities. One local directory or publication can be the whole answer.
- Publish something checkable and dated about your own market. A figure with a denominator and a date is the shape the engines quoted nationally, and it is reproducible locally.
How to test your own city in an afternoon
- Write six questions a buyer in your metro would ask, with the city named in each.
- Put all six to two engines, save the answer text, the named firms and the source list, and record the date.
- Check each named firm for three things. Real, trading, in the metro.
- Count the sources and identify the leading one. That is your first target.
- Repeat for each metro you sell in, because on our evidence the cities do not generalise to each other.
The Houston reading is the one that most clearly shows a non obvious local source winning, and it is written up in a newsroom made the top four. If you need the per city cost implication before you commit to tracking several metros, the unit arithmetic is in why the unit count multiplies before the dollar figure does.
Questions about locality in American assistant answers
So do the engines prefer American suppliers for American buyers?
Not at national level on our evidence, where they named no local firm in 64 readings. At city level they used local sources in two of four metros, so locality appears when the city is in the question and local evidence exists.
Why did Canada get local agencies named and the United States not?
Because the published, citable evidence differed. The same bank on the same day named six Canadian firms 21 times. That is a statement about what each market publishes rather than about the engines.
Will adding my city to every page help?
Adding it where it is true and useful will. A service area page that names the metro and the neighbourhoods is a document an engine can use. The same city word sprinkled into unrelated pages is not.
Do I need to track every metro separately?
If you sell in several, yes, because on our evidence no two of the four cities shared a leading source. Do the multiplication of questions times engines times cities before you price a monitoring plan.
Can I combine the national and city numbers in one chart?
No. Two banks, two dates, two denominators. Put them side by side with their own labels.
Reading your own locality rather than inferring it
The question of whether the engines favour local suppliers has a different answer in every American metro, and the only way to know yours is to read it. 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 rather than summarised. In the United States that is the Growth plan at 299 dollars a month with 100 buyer questions watched, the cheapest tier reading all five engines, read on aiknowsus.com/pricing on 22 October 2026, and the question count rather than the plan name is what to size against the number of metros you sell in.