Does AI Search Optimisation Work in the UK? The British Target List Is Not the American One

The British count, measured on 16 October 2026: aiknowsus.com appeared in 3 of 420 United Kingdom readings. The split matters more than the total. All 3 came from ChatGPT, where 210 questions produced 3 citations of us. Gemini answered the same 210 and cited us in none. That is the British baseline we hold ourselves to.

The part that changes what a British operator should do about it is in the sources. On 15 October 2026 the UK readings reached for British institutional domains that the American readings never produced at all: gov.uk was cited in 57 UK readings and ico.org.uk in 28. Neither of those appears in the US set. So a British before and after run is not chasing a smaller version of the American list. It is chasing a differently composed one, and some of the slots on it belong to public bodies.

The five domains that filled British slots

Two kinds of source dominate the UK readings, and they call for two different responses.

Source	Cited in N UK readings, 15 October 2026	What it is
gov.uk	57	British public sector, absent from the US set
whito.co.uk	46	Small British agency, own site
rank4ai.co.uk	39	Small British agency, own site
aeoscore.co.uk	32	Small British agency, own site
ico.org.uk	28	British regulator, absent from the US set

The institutional rows are not a target. Nobody displaces gov.uk on a question about registration, consent or complaints, and spending a quarter trying to is budget spent for sport. What they tell you is which questions are already answered by somebody official, so your question set should not be built mostly out of those.

The agency rows are the ones that move. Three small British firms were cited dozens of times each on their own pages, which is where most of these citations go: across 1,260 readings on 16 October 2026, 82.0% of the 20,143 citations recorded landed on a vendor's or an agency's own website, and review directories took 0.7%. Every one of those counts, and the rule used to produce them, sits in our published benchmark where a reader can check it.

Pick the questions where a British page can move the answer

This is the practical consequence of the table and it decides whether your rerun shows anything. Split your frozen question set into three groups before the baseline, not after it.

Questions a public body already owns. Anything whose real answer is a rule, a register or a complaints route. Measure them if you like, but do not count them in the set you expect to move, because the slot is held by an institution.

Questions about how the engines themselves work. These go to platform documentation and stay there. On 16 October 2026 developers.google.com was cited in 446 readings, support.google.com in 328 and help.openai.com in 269. A British page explaining Google's own mechanics is competing with Google.

Questions a British buyer asks about buying the thing. This is the winnable group and it is where whito.co.uk, rank4ai.co.uk and aeoscore.co.uk earned their counts. Prices in pounds with the VAT position stated, the British terms for the work, the date on the page. All 102 of the build spec conversations on 15 October 2026 asked for pricing in the reader's own currency and for a date and refresh frequency, and 98 of 102 asked for the engines to be named individually.

Weight your question set towards the third group and the before and after has something to measure. Weight it towards the first two and a successful quarter of work will still read as a flat line, because the slots were never available.

The city layer is a third answer again

One more design point for a British test. The UK market questions and the UK city questions did not produce the same lists. In the city scan on 15 and 16 October 2026, six questions per city on both engines, London, Manchester and Birmingham returned 36 readings between them, 33 of which came back with a named source list, and none of the 36 cited us. The winners differed by city.

So if you sell in a named British city, your question set needs a city column rather than a national average standing in for it. Deciding which catchment you are actually competing in before you write the list is the same decision as deciding the catchment before you write the keyword list, and it has the same effect on what the numbers end up meaning.

Two British rows that replicate, and the row that is missing

They establish a baseline twice over. The second run was designed eleven days after the first, asked a different and much wider set of questions, drew twice the sample, and landed in the same place on all three of the measures that matter here.

• Our own domain turned up in 3 of 597 readings first time, then in 6 of 1,260.
• 85.6% of citations landed on somebody's own marketing site the first time, and 82.0% the second.
• Review directories took 0.8%, then 0.7%.

Rewriting the questions and doubling the sample moved none of those three anywhere interesting, and that stability is what makes the British row worth building a year of work on.

They do not establish a change caused by anything, because nothing was altered between them and the second set of questions was not the first set. What is missing is the British percentage point movement with a baseline sample and a follow up sample printed beside it. That is a measurement still to be made rather than one being withheld, and making it takes a frozen British question set, a dated change log, a month of waiting and an identical rerun.

There is no reason to wait for our round before starting yours. The before and after measurement protocol carries the whole method, including the blind rule and the arithmetic that tells you whether a movement is large enough to be visible at the sample size you can afford.

We sell the measurement this page tells you to run, and the British result we are reporting about ourselves is 3 readings in 420.

What British firms ask about the evidence behind this

When should we expect the second round to move?

Set the rerun at least thirty days after the baseline and take weekly readings in between, so you see the shape rather than two endpoints. Anything closer together than a month is mostly reading crawl and retrieval lag.

Can SEO changes improve AI recommendations here?

The British pages earning citations on 15 October 2026 were small agency sites publishing the British version of the buying question, with prices in pounds and the VAT position stated. That is ordinary good British content practice, and 102 of 102 build specs asked for the currency and the date by name.

Why do gov.uk and the ICO matter to this?

Because they hold slots. gov.uk was cited in 57 UK readings and ico.org.uk in 28, and neither appears in the American set. Those questions are answered by an institution, so leave them out of the group you expect your own work to move.

Does a national British question set cover my city?

Not reliably. London, Manchester and Birmingham returned 36 readings on 15 and 16 October 2026 and the winning sources differed by city. If your trade is local, put a city column in the question set before the baseline rather than after it.

Can you show a British client result?

Not yet, and we will not manufacture one by setting two different question sets side by side and calling the gap a result. The British figure on this page is a starting line, 3 citations in 420 readings on 16 October 2026, and the intervention round is dated work rather than a claim.

Nothing in the current British answer carries a sample size

What a British buyer actually hears is confident, correctly British in tone, and built from a handful of .co.uk pages plus some government guidance, with not one auditable figure in it. That is not a complaint about the engines. It describes what is available for them to read. The supply of British pages on this subject that print a numerator, a denominator and a date is close to nil, which explains both why the field looks the way it does and why a dated British measurement is cheap to produce and difficult to argue with. The one AI Knows Us publishes about itself currently reads 3 in 420. The next carries its own date, and the distance between the two is the only honest answer to the question in the title above.