AI brand visibility in the UK, measured: 420 readings of ChatGPT and Gemini, and the best and worst result in the dataset arriving on the same day

Hand the same 210 British buying questions to two engines on one day and you can take your best result and your worst out of the same data. On 16 October 2026 that is what Britain produced, across 420 UK readings with none discarded: our own domain was cited in 3 of the 210 British ChatGPT readings, the best of the six market and engine cells we recorded, and in 0 of the 210 British Gemini readings. Any British business about to buy a single AI visibility score should start from that fact. Averaging those two cells gives a number that describes neither engine, and it is the one figure most reports in this category lead with.

The six cells, so the British ones can be read in context

The run covered the United States, the United Kingdom and India, 210 questions per market in 21 batches, 630 questions and 1,260 readings with none discarded. The question sets went deliberately into eight directions an earlier run had missed, including procurement depth, questions that name a competitor, questions asked by a job title rather than a buyer, comparison against the large SEO suites, and timing. The earlier run is published in what AI assistants actually cite across 597 readings of ChatGPT and Gemini in three markets, and its British cut, with the small agency domains that won it, is in the UK cut of that benchmark.

Market and engine	Readings	Readings citing our domain
United Kingdom, ChatGPT	210	3
United Kingdom, Gemini	210	0
India, ChatGPT	210	2
India, Gemini	210	0
United States, ChatGPT	210	1
United States, Gemini	210	0

Read down the Gemini column. Nothing, three times. Add the earlier run's 291 Gemini readings and the total is 921 readings without a single citation of our domain. It is the most repeated single result in this programme. What each engine did with sources overall is counted in how often each engine returned sources at all.

One score across two engines averages two different games

The two engines are not two samples of the same thing. On British questions they behave differently enough that a publisher faces two separate problems.

ChatGPT concentrates. In the earlier run its single most cited domain appeared in 176 of its 306 readings, which means a short, entrenched list holds most of the slots and anyone publishing is competing against it directly. Gemini spreads. Its most cited domain appeared in 19 of its 291 readings, so there is no entrenched list to displace and the field is flat and wide. A British agency being told its ChatGPT visibility and its Gemini visibility have been averaged into one number is being handed the average of two different games.

That is also why the British cells came out as they did. A domain can get onto a concentrated engine's list and still be absent from a flat one, and a single score cannot show you which of those happened. Ask for the figure per engine, per market, with the denominator printed. If a British supplier will not break a score down to that level, the score is a presentation rather than a measurement.

What was cited for British buyers, and what that costs to compete with

20,143 citations came out of those 1,260 readings, spread over 4,947 distinct domains. Sorted by who owned the page, 82.0% sat on a vendor or agency own site and 0.7% on a review directory. By readings, the ten most used domains were developers.google.com in 446, support.google.com in 328, developers.google.cn in 325, blog.google in 309, help.openai.com in 269, github.com in 243, developers.openai.com in 230, semrush.com in 199, theverge.com in 197 and searchengineinfo.org in 147.

Two things follow for a British publisher. The first is that the top of that list is platforms answering questions about themselves, so a British page explaining how an engine picks sources is bidding against the company that built the engine. The second is more encouraging: with 82.0% of citations going to self published pages, the slots that are not held by platform documentation are held by publishers rather than by directories. Review directories took 0.7% of citations, which is the measured answer for any British firm deciding between a year of directory profiles and a quarter of well evidenced pages.

British questions also reach for British institutions. Two of them, a government domain in 57 British readings and a regulator domain in 28, did not surface in the American set at any point in the earlier run. For anything touching data, consent, complaints or compliance, the British page that names the relevant regulator page and then answers the commercial question the regulator does not answer is sitting beside a source these engines already use.

How the British readings were produced

British phrasing throughout, with the United Kingdom written into each sentence rather than switched on as a parameter, and every answer read for pounds and British context before it was counted. No question carried our own business name, which would have turned the reading into a test of recall. The counting rule throughout is one domain per reading, deduped, and the scoring, the classifier and the handling of proxy hostnames are documented with the method rather than summarised here.

On the publishing side of this, prices on a British page belong in pounds with the VAT position stated, which all 102 build spec conversations in the earlier run asked for in one form or another, and a comparison belongs inside what the page can prove on its face, since the CAP Code governs British marketing claims and a complaint can be put to the Advertising Standards Authority. Fixing the market before the questions are written is the same discipline deciding the catchment first asks of a British advertiser, for the same reason: the scope decides what the numbers can mean.

What this British dataset does not contain

Trade by trade, it says nothing. No trade was named in any of the 210 questions, so a professional services figure or a retail figure could only reach this page by being made up. City figures are not in this run either; London, Manchester and Birmingham were each read separately in a scan of ten named cities, which is where any city claim has to come from. Six cells is the whole table, and three engines are missing from it on purpose: Google AI Overviews, Perplexity and Claude were never asked these questions. We read five surfaces for customers and these counts reach two, which is why nothing here is called an AI visibility score.

Questions about the UK AI brand visibility readings

Which engine names British businesses more often?

Three of the 210 British ChatGPT readings used our own domain as a source and none of the 210 British Gemini readings did, on one day, so in this market ChatGPT was the engine that cited and Gemini was not.

Run the same comparison on your own frozen British questions, because a cell this size tells you about this domain rather than about British business in general.

Does British AI visibility vary by industry?

This dataset cannot answer it, because no trade was named in the questions and so no trade level British rate was ever scored. It is the next thing to measure rather than something to read out of these counts.

Why publish three citations rather than a percentage?

Because a percentage taken off 210 readings implies a precision those 210 readings cannot support, while three of 210 with a date on it is a fact anybody can check. Three of 210, with the date and the engine attached, is a fact a reader can check.

Is a .co.uk domain required to be cited for British questions?

What the British winners had in common in the earlier run was British specificity in the content. A UK facing page on a .com can carry the pound price, the VAT line, the named regulator and the British version of the question just as well.

How often are the British readings taken again?

Often enough that an engine update shows up as a movement rather than as a surprise, and always on the same six cells so the comparison is like for like.

The British result to carry into your next supplier conversation

Three of 210 and 0 of 210, same questions, same day, two named engines. Whatever a British supplier tells you about your AI visibility, that is the shape the honest answer has: a fraction per engine, per market, with a date. We are paid to run this measurement and to write the pages that follow it, which is the context these counts belong in, because the best cell in the dataset is still three readings out of two hundred and ten and it is our own. It is printed here with its denominator so the same question can be put to us on the next round.