September 2026: over half of what Google cites in the UK is not in the results
Forty B2B buying queries put to Google AI Overviews, four sectors, UK market. The reading was taken on 2 September 2026 and cost 0.16 dollar. Fast Growth Advisors publishes the questions, the answers and the cited sources, with the link to run each search again. Google AI Overviews answered all forty, without exception.
Ten answers per sector, 296 sources across 190 distinct domains. Only 46.5 % of them appear anywhere in the organic results.
Does ranking buy as much here as it does in France?
No. Fast Growth Advisors ran the same measurement on the French market the day before, with an equivalent question set and the same sector structure, and the two results are not close.
| What we measured | United Kingdom | France |
|---|---|---|
| Cited sources that appear in the organic top | 46.5 % | 70.2 % |
| Queries where the 1st organic result is cited | 57.5 % | 74 % |
| Queries where one of the top 3 is cited | 77.5 % | 96 % |
The gap holds at every level of the table. One of the top three results is cited on 77.5 % of UK queries, against 96 % in France.
What does ranking first get you on the UK market?
On the UK market, ranking first gets you cited on fewer than six queries in ten. In France it is nearly three in four.
A UK site that holds its positions and loses its traffic therefore has less to reclaim through ranking work than a French one. More of the source slot is decided by something other than position: whether the page answers the question, and whether third parties describe the vendor.
Fast Growth Advisors reads that difference as a budget signal rather than a verdict on SEO: ranking work still pays in the UK, it simply buys fewer citations.
This is a single reading on a single month. It sets the level, it does not prove a trend, and the next edition will say whether the gap with France holds.
Who does a UK vendor compete with inside Google AI Overviews?
Mostly American publishers. Here is how the 296 cited sources break down in the Fast Growth Advisors reading.
| Nature of the source | Citations | Share |
|---|---|---|
| Websites, vendors and media combined | 234 | 79.1 % |
| Video | 43 | 14.5 % |
| Forums | 17 | 5.7 % |
| Public sector | 2 | 0.7 % |
Websites, vendors and media combined carry 234 of the 296 citations, or 79.1 %.
Only 10.5 % of the cited domains end in .uk. Among the most cited names: youtube.com, reddit.com, zapier.com and gartner.com. A British vendor is not competing with British competitors for the source slot, it is competing with American publishers.
Video holds 14.5 % of citations here.
On the French market, measured the day before on an unrelated question set, it held 15.1 %. Two markets, two question sets, and almost the same share, which suggests video deserves a test in any B2B content plan that currently ignores it.
Forums weigh more than in France: 17 citations, 5.7 % of the total. Public sector sources, on the other hand, barely exist: 2 citations, 0.7 %.
What we do not conclude from it. We measured a presence, not a cause. No control group lets us say that a video gets cited more easily than a page. This figure justifies a test, not a budget reallocation.
Does the French rule hold across three readings?
Only partly. In France, the vendor Google names on a comparison query was never among its cited sources.
In the UK, the three Fast Growth Advisors readings below give three different outcomes.
best payroll software uk. Google recommends Xero Payroll and Sage UK for small and growing teams. Its sources include xero.com and sage.com, alongside reddit.com, rippling.com and money.co.uk. The vendors it recommends are also cited.
comparison hr software uk. Google names Breathe HR for small and medium UK businesses. Its sources are money.co.uk, capterra.co.uk, roothr.co.uk, smallbusiness.co.uk and video.
Here the named vendor is not cited at all.
vulnerability management saas. Vendor pages are the source, exactly as on the French market: Google explains what the category does, then cites qualys.com, cycognito.com, sophos.com, orca.security, tufin.com and accuknox.com.
Being named without being cited happens here too, but not systematically. A vendor cannot assume either outcome without measuring it on its own questions, in its own market and spelling.
Some categories even behave differently across the Channel. Business password managers get an answer on the UK market, citing reddit.com, pcmag.com and gartner.com, while two French phrasings produced no AI Overview at all. Same category, same month, two behaviours.
Privacy-focused analytics is the other case. The query google analytics alternative gdpr returns Plausible, Fathom, Matomo and PostHog, and every cited source is a vendor page. In French, the equivalent query returns an answer written in English, which we reported as an anomaly rather than correcting it silently.
How was the September 2026 edition measured, and how stable is a citation?
40 questions, ten per sector, written by hand by Fast Growth Advisors and never generated.
Five comparison and five needs queries in each sector, the same structure as the French set, so the two markets can be compared sector by sector.
They were chosen from 52 candidates, all measured before selection. None failed, where the French set lost three out of 65. British spelling was used deliberately: measured on 1 September, optimisation and optimization returned two entirely different sets of sources with no domain in common.
These two markets do not start from the same point. Google brought AI Overviews to the UK in August 2024 (TechCrunch), while the French rollout began on 22 July 2026 (Blog du Modérateur).
Every reading keeps the raw payload, question by question and date by date, with its own measurement time. Cost: 0.004 dollar per question, 0.16 dollar for this edition.
A citation is not a position you hold. We replayed the same question eleven times in a row, seconds apart: on the UK market, best hr software for small business returned the same eight domains, eleven times out of eleven.
On the French market, the same experiment on comparatif logiciel sirh pme produced the opposite: sources changed from one call to the next, with a median overlap of 62 % and 38 % in the worst pair. Only three domains appeared in all eleven readings.
So there is a stable core and a rotating fringe.
Sitting in the core is worth far more than appearing once, and a tool that measures a single time cannot tell the two apart. A caveat, though: two questions, one per market. Stability is measured question by question, and the next edition, which starts the series, will publish it.
Sources
Links checked on 30 September. The two external sources below date the arrival of AI Overviews on each market, which matters when comparing them: the UK had lived with AI Overviews for about two years when these readings were taken, France for six weeks. Every other figure on this page is Fast Growth Advisors data, from the readings of 1 and 2 September 2026 (50 French questions, 40 UK questions). Neither external source feeds a measurement: both only date the context.
- TechCrunch, “Google is bringing AI overviews to India, Brazil, Japan, UK, Indonesia and Mexico” (16 August 2024): the UK rollout.
- Blog du Modérateur, “Google déploie AI Overviews et AI Mode en France” (22 July 2026, in French): the French rollout.
FAQ
On how many questions did Google write an answer?
On all forty.
Ten questions per sector, four sectors, no exception. On these UK B2B buying categories the question is no longer whether Google AI Overviews answer instead of the sites, but whom they cite when they do. Fast Growth Advisors found the same on the French market, where Google answered all fifty questions.
Does organic ranking still matter on the UK market?
Less than in France, and the gap is wide.
Only 46.5 % of the sources cited appear anywhere in the organic results, against 70.2 % on the French market. Ranking first gets a UK site cited on fewer than six queries in ten, against nearly three in four in France.
Why build a UK question set instead of translating the French one?
Because a translated query measures a market nobody searches in.
We also measured that British and American spellings return two entirely different sets of sources, with no domain in common, so the wording itself decides which market you reach. A French set translated into English would have measured neither market properly.
Is the UK market more stable than the French one?
Nothing in this edition proves it.
One UK question returned the same eight domains eleven times out of eleven, while one French question saw its sources move from call to call. Two questions, one per market, cannot rank two markets. Stability is measured question by question, and the series starts with the next edition.
How much does a UK edition cost?
0.16 dollar for the forty questions.
That is 0.004 dollar per question, published with every edition. At that price a measurement has no reason to be replaced by a guess, and anyone can rerun each search from the link kept with the reading.