AI visibility for a business in London: the engines answered with UK wide sources and no London named site won a reading

Ask ChatGPT or Gemini who a London business should hire and the sources behind the answer are mostly not London sources. Across twelve London readings taken on 15 and 16 October 2026, ten of which returned a source list, four sources tied at the top: generativeengineoptimizationuk.agency on 3 readings, firstmotion.com on 3, agentcraft-ai.com on 3 and seocompare.co.uk on 3. Not one of the twelve named us. The useful detail sits in those four names. Not one of them describes the city, while two put the category and the country straight into the domain. For a London business that means the words "in London" narrow almost nothing, and the facts underneath them decide the whole answer.

The twelve London readings, and who won them

Sources at the top, and their readings	generativeengineoptimizationuk.agency 3, firstmotion.com 3, agentcraft-ai.com 3, seocompare.co.uk 3
Of those four, how many name London in the domain	0
London readings taken	12, six buyer questions on each engine
Of those, how many returned a source list	10
Of those, how many named our own domain	0
Engines, and the dates they were read	ChatGPT and Gemini, 15 and 16 October 2026

London was one of ten cities in that run, 120 readings in total, and our domain was absent from every one of them. Every figure on this page comes from two engines, and the product follows five. The three missing from this run are Perplexity, Claude and Google AI Overviews, so take these counts as two of the five surfaces we normally read and nothing wider. Sampling, repeat behaviour and the full domain mix for the bigger exercise these city questions belong to are all in what AI assistants actually cite across 597 readings.

Why a London question comes back as a British answer

London is where companies that sell to the whole country put their registered office, and very often their only office. A page written for a London buyer and a page written for a UK buyer are usually the same page. The engines have no reason to treat London as a filter, because most of the material they hold about London is national material with a London address on it.

The same thing showed up from the other direction in the market run on 15 October 2026, whose British findings are set out in the UK cut of 204 readings. The United Kingdom prompts were locked in the wording of the question rather than by a country parameter, and the check was to read the answers back: the UK questions came back answered for London, in pounds. London was the city the engines reached for when asked about Britain at all. That has two consequences, and the second is the better news.

• You are compared with every supplier in the country, not with the firm two streets away. A buyer in Clerkenwell asking an assistant for a shortlist gets national names, so being the best known business in your own postcode does not put you in that list.
• Nobody holds the London answer. Four sources tied on 3 readings each is a field with no leader in it, and the four that led are small operations publishing their own pages about a British category. A slot like that is taken by writing something, not by buying something.

The fact a London page leaves out most often

London holds more group head offices, more UK subsidiaries of overseas parents and more trading names than anywhere else in the country, so the name on the website is frequently not the name on any register. An assistant can repeat a brand it cannot connect to anything, or it can repeat a fact. Give it the fact:

• The registered company name and company number at Companies House, written as text rather than inside a footer image, with the registered office.
• Whether you are a limited company or a sole trader. Buyers are comfortable with either once they can tell which.
• Which company trades as which brand, if those differ, in one plain sentence.
• Your VAT registration position, because that tells a commercial buyer how the invoice will arrive.
• Any register entry your sector carries, named with its number, such as a regulated firm's entry on the Financial Services Register.

This matters more in London than anywhere else in Britain for one reason. The brand a buyer types and the entity a third party can confirm come apart most where groups and subsidiaries cluster, and they cluster here.

Narrowing below London, which is the narrowing that works

If "in London" does not narrow, something else has to. Here the honest narrowing is sector and client size, because London's business base sits in deep pockets: financial services, law, recruitment, property, creative and media, fintech and hospitality. "A London agency" is a national claim with a postcode attached. "A London agency for FCA regulated firms, with a compliance sign off step written into the schedule" is a London claim, and it is answerable.

For a business customers physically visit, the narrowing is the catchment rather than the city: the honest list of boroughs and postcode districts you actually serve, and the ones you do not. The British grain of this, including what differs for UK search generally, is covered in what actually differs for UK search. If you are reading this while also shopping for help, the tier of supplier you belong in decides your entire shortlist before you speak to anybody, and that decision is worked through in which tier you belong in before you approach anybody.

What a London page has to put in writing

Alongside the readings we ran 102 build spec conversations, showing each engine its own answer and asking what a page would have to contain to be cited instead. Three of those counts bear directly on a London supplier trying to stand apart in a crowded market:

• 102 of 102 wanted the price in the reader's own currency. Pounds, with the VAT position written next to the figure. Quote a consumer figure as the amount the customer hands over, and label a trade figure as excluding VAT wherever it is. In a city where most agency sites publish no number at all, a figure with its VAT treatment attached is a differentiator before it is a courtesy.
• 83 of 102 wanted a confidence interval, a variance or a repeat count. If you publish a result, say how many times you measured it. One reading is an anecdote; twelve is a figure.
• 78 of 102 wanted the question set published so it can be re-run. This is the London differentiator hiding in plain sight. Every firm in the market claims the same outcomes, and almost nobody publishes the questions they tested. Printing yours lets a buyer check you instead of believing you.

What London firms ask after reading the source list

Does putting London in the page title help?

On its own, very little. None of the four most cited London sources on 15 and 16 October 2026 had the city in its domain, and two of them named the category and the country there instead. The title carries the city. The body has to carry the sector, the client type and the facts somebody else can confirm.

Which sources did the engines actually use for London?

generativeengineoptimizationuk.agency, firstmotion.com, agentcraft-ai.com and seocompare.co.uk, each in 3 of the 12 readings. Four names, three readings apiece, no leader.

We have one office and it is in London. Are we a London business or a UK business to an assistant?

Whichever one your pages say you are, and most London sites never say. Decide it, then write it: either the page serves customers who come to you in London, in which case name the boroughs, or it serves the country from a London address, in which case lead with the sector and let the address be an address.

We are the UK arm of an overseas group, so which name belongs on the page?

Both, in one sentence, with the company number. Name the UK registered company, give its number at Companies House, then say which brand it trades as. That sentence is what lets an assistant tie the brand a buyer types to an entity somebody else can confirm.

Can we see which page of ours an assistant actually pulled from?

Yes, when the engine returns a source list, because the list gives URLs and not just domains. Record the full URL each time, and you learn which page is doing the work and which pages the engines have never reached for. On 10 of our 12 London readings there was a list to record.

Taking one of the four London slots

A flat field is an invitation. Four sources tied on three readings each means nobody has settled the London version of this question, and the names that led were small operations publishing their own pages about a British category. Pick the one buying question your London customers actually ring with, answer it with your sector named, your registered entity written out, your price in pounds with the VAT position stated, and the questions you tested printed underneath. We earn our money from that measurement and the publishing that follows it, which is exactly why the London figure printed above is the one showing our own domain on zero: AI Knows Us would rather hand a London buyer a number they can check against us than a claim they cannot.