Our published source tables name small agencies in two markets and none in the third

In the 15 October 2026 run of ChatGPT and Gemini, the British cut surfaced three small agency domains with counts: whito.co.uk at 46 citations, rank4ai.co.uk at 39 and aeoscore.co.uk at 32. The Indian cut surfaced four: upgrowth.in at 52, themediaverse.in at 48, nicodigital.com at 44 and hawkorbit.com at 33. The American cut we published carries no equivalent list of United States only agency domains.

Before anybody builds an argument on that absence, the honest description of it is this: it is a limit of what we published, and not a measurement of what exists in the American market. Those are two different claims and a reader is entitled to be told which one a table supports.

What the two named lists do support

Market	Named domains with counts in our published cut
United Kingdom	whito.co.uk 46, rank4ai.co.uk 39, aeoscore.co.uk 32
India	upgrowth.in 52, themediaverse.in 48, nicodigital.com 44, hawkorbit.com 33
United States	No United States only agency domain is named in the published cut

The British and Indian rows are positive findings. Seven small, market specific domains were cited between 32 and 52 times each in a run of 597 usable readings, and none of them is a large company. That is a measured statement about what a domain built around one market can do.

What the empty row does not support

• It says nothing either way about whether an American agency domain was cited. The run recorded 10,652 citations across 2,708 distinct domains. A domain can be in that set without being in a published list of seven.
• It is not a ranking of markets. The lists were assembled as examples of a pattern and not as a complete inventory of any market's winners.
• It is not a claim about American agencies in general. Nothing in the run measures the size, age or quality of any firm.
• It is not transferable to another category. Our bank asks about our own market, and a bank about machinery or conveyancing would surface a different local layer in all three countries.

What makes the distinction practical rather than pedantic is how often an absence gets read as a result. A published table is the part of a dataset somebody chose to print, so the honest question to ask of any table, ours included, is what the selection rule was and whether a row is missing because nothing was there or because nobody listed it.

The one place in our data where American local domains are named

They are in the separate city scan, taken on 15 and 16 October 2026, with six buyer questions per city on both engines and twelve readings each. The American cities in it produced small, city specific winners: localseo.nyc at 3 citations of New York's 12 readings, chicagoseoagency.co at 3 of Chicago's 12 and houdigitalmarketing.com at 3 of Houston's 12.

So the right reading of the gap is about instrument scope rather than geography. A bank of national questions and a bank of city questions ask different things, and the second one is where a local American domain shows up. If you want the local layer of a market, you have to ask local questions, which is also the reason a market level figure cannot be used to describe a city. The habit of deciding which variable a reading actually describes before acting on it is the same one set out in geographic segmentation, the variables and a worked example.

Four questions to put to any source table

• How were the listed rows chosen? Top by count, examples of a pattern, or everything above a threshold. Each produces a different table from the same data.
• How many rows exist that are not shown? Ours is 2,708 distinct domains behind a printed list of ten or fewer.
• What is the unit? Ours is one domain in one reading, deduped, and the same run counted by raw address occurrence would give a different total entirely.
• Which cut is each row from? A market cut, the whole run, or a separate scan. Rows from different cuts in one table need labels or they will be compared wrongly.

The edges of this run

Two apps, ChatGPT and Gemini, on 15 October 2026, with 102 questions per market of which 72 were worded identically across markets and 30 were written for the market. 597 readings were usable of 612 attempted. Three of the five surfaces our product covers were asked nothing. No question in the bank named an industry, so nothing in it supports a figure for one. Every count records a domain sitting in a source list and not a business named in an answer's text.

We sell this measurement and the publishing work after it, so the row that goes against us is printed as well: our own domain was cited 3 times in those 597 readings, with none of them in the United States on either engine. The Google Search Console side of our record, with both windows and the export date, is our proof page.

How the two named lists were used, and how they were not

Those seven domains have a job on this site and it is a narrow one. They are the evidence for a single claim: that a domain built around one market, carrying pages about that market, can be cited dozens of times in a run where the top of the table belongs to platform documentation. Three British sites between 32 and 46 citations and four Indian ones between 33 and 52 is enough to support that claim and nothing larger.

What they were not used for is a league table of agencies, a judgement about any of the seven, or a forecast for anybody else's domain. None of the seven was contacted, tested or assessed. Their counts are the only thing we know about them, which is worth saying plainly because a named domain with a number next to it reads like an endorsement unless somebody says otherwise.

Questions about what a published source table leaves out

Which small domains are named in the British cut?

whito.co.uk at 46 citations, rank4ai.co.uk at 39 and aeoscore.co.uk at 32, from the run of 15 October 2026.

And in the Indian cut?

upgrowth.in at 52, themediaverse.in at 48, nicodigital.com at 44 and hawkorbit.com at 33, from the same run.

So no American agency was cited?

That is the inference the table cannot carry. What is true is that our published American cut names no United States only agency domain, out of 2,708 distinct domains recorded across the whole run.

Where do American local domains appear in your data?

In the separate city scan, where New York, Chicago and Houston each produced a small city specific domain at 3 citations of 12 readings. That scan asked city questions, which the national bank did not.

Which engines and date produced the named lists?

ChatGPT and Gemini on 15 October 2026 across three markets. The full run, with its method and complete source mix, is published as 597 readings across ChatGPT and Gemini in three markets.

Say what the table is a table of

Two of our three market cuts carry named local winners and the third does not, and the reason is the selection we made when we printed it. Writing that down costs a sentence and it stops seven real findings being turned into an eighth that was never measured. Ask the same of the next source table anybody hands you, starting with how its rows were chosen.