Forty eight American city readings, and our own domain was in none of them

Our city scan put six buyer questions to ChatGPT and Gemini for each of four American cities on 15 and 16 October 2026. That is 12 readings per city and 48 American readings in total. 43 of the 48 came back with at least one source attached. Across all 48, the number of readings that cited a page on our own domain is zero.

Every local claim on this site is measured from that floor. Publishing it is a working decision and not a show of humility: anybody who has noticed that we are an interested party goes hunting for the row that costs us something, and the rest of the table is worth more once that row is sitting there to be found. The four city names, the two dates and the denominators are all in the table below so the hunt is short.

The forty eight readings, city by city

City	Readings	Readings that returned sources	Readings citing our domain
New York	12	9	0
Los Angeles	12	11	0
Chicago	12	12	0
Houston	12	11	0
All four	48	43	0

The middle column carries as much information as the last one. Five readings of the 48 came back with nothing behind them, and no page anybody publishes can be part of an answer that used none. So a local result belongs against the sourced readings, and both counts belong in the report rather than one divided into the other.

Why a zero is published before it improves

• A floor recorded afterwards is not a floor. Four city names, four denominators, two dates and a nought are now public, which is enough for a sceptic to hold the next reading against this one without asking us for anything.
• It separates two kinds of absence. Missing from a reading that used sources points at what the pages contain. Missing from a reading that used none points somewhere else entirely, and only a report carrying both columns can tell which one you have.
• It is what the engines themselves asked for. Across our 102 build spec conversations of 15 October 2026 the tallies came out at 68 for showing negative results and 64 for naming the commercial interest on the page.
• It keeps the rest of the table credible. The same scan names the sources that were cited in these cities, at 3 and 4 citations of 12 readings each. Those rows are easier to believe beside a row that reads nought.

What the zero does not mean

Precision matters most where the number is smallest.

• It does not say the engines cite nobody in these cities. Each city produced named sources with counts attached.
• It does not describe any city we did not read. Four American cities have figures here and no others do.
• It does not describe the three surfaces our product covers that were absent from the scan. Two apps were read, ChatGPT and Gemini.
• It is not a statement about being recommended. These readings count a domain appearing in a source list, which is a separate event from a business being named in the text of an answer, measured separately and never added to it.
• It does not forecast your result. It is our own floor, published so our next one can be checked.

Taking your own local baseline

The design is short enough to copy in an afternoon and it is the only number that will survive an argument about your own city.

• Write six buyer questions with the city inside the wording. A location parameter is not a substitute: the surfaces we read take one, report success and then disregard it, which we established before either run was built.
• Name no company in any question, including your own.
• Ask each question on two named apps, which gives you twelve readings.
• Record three things per reading: whether sources came back, which domains they were, and whether yours was among them.
• Write the date, and set the date of the next reading before you look at the result.

That measurement, and the pages that follow it, is what we are paid for. The Google Search Console side of our own record, two domains with their windows and the export date printed on the page, is our proof page.

What a floor is for

A number like this earns its place by being hard to move afterwards. It names the cities, the apps, the dates and the denominators, which means nobody can later reinterpret it as a soft start or a partial reading. It also fixes the comparison: the next scan uses the same questions, so a change in the figure is a change in the world rather than a change in the instrument.

That is the part most reporting in this field skips. A first reading taken loosely cannot be compared to a second taken carefully, so the programme quietly loses its own history and every quarter starts again from an impression. Forty eight readings, written down once, prevents that for as long as the question set stays frozen.

Questions about our own local citation baseline

How many American city readings were there?

48, being 12 each for New York, Los Angeles, Chicago and Houston, taken on 15 and 16 October 2026 on ChatGPT and Gemini.

How many cited your domain?

None. Zero of 48, and zero in each individual city.

How many of the readings returned any sources?

43 of the 48. Chicago returned sources on all twelve, New York on nine, and Los Angeles and Houston on eleven each.

Does this mean local AI visibility work does not do anything?

It means that on those two dates nothing of ours had been picked up in those four cities. The same readings show which pages were used instead, and the top counts in each city sit at 3 or 4 citations of 12, which describes a field nobody has settled.

Which cities and which apps?

Four American cities out of the ten in the scan, read on ChatGPT and Gemini. Our wider category reading is at 597 readings across ChatGPT and Gemini in three markets.

The floor, stated so it can be used against us

Zero of 48 American city readings, 43 of which carried sources, across New York, Los Angeles, Chicago and Houston, on 15 and 16 October 2026, on two named apps, with no company name in any question. Six questions per city and the same two apps are what the next reading will use. A nought second time round gets printed in this table beside the first one.