Finding
Across six professions Whito tested on 14 and 17 August 2026, between 55% and 85% of the firms AI engines named were named by only one engine, one of 55 dated figures from 34 Whito studies on this page.
Source: Whito, whito.co.uk/research/uk-ai-search-statistics. Free to quote with a link to this page.
Last Updated on September 17, 2026
First published 15 September 2026. Every figure below was counted by Whito, dated, and linked to the study it came from. Reviewed with each quarterly edition of the Whito AI Visibility Index; the next is due December 2026.
A survey tells you what people say they do. It cannot tell you what an AI engine actually recommended on a given day, or whether the business it named is on a register.
That second kind of figure can be checked. You cannot re-run someone else’s poll. You can ask ChatGPT the same question again, look the answer up on Companies House, and see whether you get the same result.
So this page holds only what Whito counted: AI answers collected on a named day, websites and search results checked, businesses matched against a named register, files read on a named date. There are 55 dated figures from 34 Whito studies, counted between May and September 2026. Each line gives the date, what was asked, the base, and the study to read before you quote it.
The short version: AI engines often name different businesses for the same question, a measurable share of the businesses they name cannot be matched to a register, and Google AI Mode leans on business websites and Google listings far more than on regulators.
The key figures
Sources: the Whito studies linked from each numbered line below.
Jump to: Do AI engines agree on who to recommend? · Can you check the businesses AI recommends? · Where AI engines get their answers · Ratings and adverts inside AI answers · Are UK business websites ready for AI crawlers? · Do AI citations send visitors? · Trust beyond AI: registers, directories and search
Do AI engines agree on who to recommend?
In Whito’s June test, three engines named three different top picks on 11 of 12 questions. Ask one engine the same question again and the names still move.
1. Two of three AI engines shared a top pick in 1 of 12 identical questions, 8%, and all three agreed in none. 33 different businesses were named as the number-one pick across the 36 answers.
Source: Whito’s AI engines disagree study. Counted 30 June 2026. ChatGPT, Gemini and Perplexity, free tiers. Four trades in Leeds, Manchester and Bristol, first-named business only, 36 answers.
2. Only 11 of 182 electricians, about 6%, were named by all three of ChatGPT, Gemini and Google AI Mode. In 8 of the 15 towns the three engines did not agree on a single electrician.
Source: Whito’s electrician AI study. Counted July 2026. Prompt: “Recommend a trustworthy electrician in [town]”, once per engine per town. 15 towns, 45 answers, 237 mentions.
3. Across six professions Whito tested on 14 and 17 August 2026, between 55% and 85% of the firms AI engines named were named by only one engine. Vet practices sat at the low end: the study reports 45% named by more than one engine, so 55% by one only. Accountancy firms sat at the high end, 85%. Mortgage advisers were 62%, dental practices 66% and law firms 74%. Estate agencies were 63%, which is 100 minus the 37% the study reports. The vet and estate agency figures both err high, because Gemini did not answer in every city.
Sources: vets, accountants, mortgage advisers, dentists, solicitors, estate agents. Counted 14 August 2026 (accountants, solicitors, dentists, vets) and 17 August 2026 (mortgage advisers, estate agents). Six cities each. ChatGPT, Gemini and Google AI Mode; the accountants test used Perplexity in place of Google AI Mode. The figures are shown side by side, not pooled.
4. ChatGPT, Gemini and Perplexity named 109 accountancy firms across six UK cities and agreed on exactly one. 93 of the 109, 85%, were named by one engine only.
Source: Whito’s accountant AI study. Counted 14 August 2026. Prompt: “Who is the best accountant for a business in [city]?” 18 answers, 126 recommendation slots.
5. 3 of 61 law firms, 4.9%, were named by all three AI engines, and 74% were named by one engine only.
Source: Whito’s solicitor AI study. Counted 14 August 2026. ChatGPT, Gemini and Google AI Mode. Prompt: “Who is the best solicitor for a business in [city]?” Six cities, 18 answers, 80 slots.
6. 5 of 59 dental practices were named by all three AI engines, and 66% were named by one engine only.
Source: Whito’s dentist AI study. Counted 14 August 2026. ChatGPT, Gemini and Google AI Mode. Prompt: “Who is the best private dentist in [city]?” Six cities, 84 recommendation slots.
7. 45% of 60 vet practices were named by more than one AI engine, a figure the study says is understated. In the four cities where all three engines answered, 8 of 43 practices were named by all three. The accountants test the same day, run with Perplexity in place of Google AI Mode, gave 15%, so that comparison is directional.
Source: Whito’s vet AI study. Counted 14 August 2026. ChatGPT, Gemini and Google AI Mode. Prompt: “Who is the best vet in [city]?” Six cities, 16 of 18 answers delivered, 95 slots.
8. 5 of 71 mortgage firms, 7%, were recommended by all three AI engines, and 62% were named by one engine only.
Source: Whito’s mortgage adviser AI study. Counted 17 August 2026. ChatGPT, Gemini and Google AI Mode. Prompt: “Who is the best mortgage advisor in [city]?” Six cities, 18 answers, 103 slots.
9. 81% of 75 recruitment agencies were named by one AI engine only, and 5, 6.7%, by all three.
Source: Whito’s recruitment agency AI study. Counted the evening of 3 September 2026. ChatGPT, Gemini and Google AI Mode. Six cities, 18 answers, 94 recommendation slots.
10. 10 of 57 interior designers, 17.5%, were named by all three AI engines, the same evening only 5 of 75 recruitment agencies were.
Source: Whito’s interior designer AI study. Counted the evening of 3 September 2026. ChatGPT, Gemini and Google AI Mode. Prompt: “Who is the best interior designer in [city]?” Six cities, 18 answers, 98 recommendations.
11. Thirty identical ChatGPT requests for a trustworthy plumber produced 180 recommendations naming 51 different businesses. Only one business was named every time in its town, and 17 of the 51 were named exactly once.
Source: Whito’s AI recommendation lottery study. Counted 17 July 2026. ChatGPT free tier, fresh chat each time. Nottingham, Luton and Preston, ten asks each.
12. Rephrasing the same plumber request changed roughly 70% of the businesses ChatGPT named.
Source: Whito’s AI recommendation lottery study. Counted 17 July 2026. Four phrasings (plain, polite, as a pensioner, as a landlord) in five towns, 20 asks.
13. Only 5 of 99 businesses were named by both ChatGPT and Google Gemini across eight Manchester trades.
Source: Whito’s multi-engine local recommendation study. Counted 14 September 2026. One fixed trustworthy-firm question per trade, consumer chat sites. Manchester only.
14. TSB, Metro Bank and Virgin Money were not named once across 40 ChatGPT and Google AI Mode answers to UK consumer questions. Octopus Energy was named in all eight energy answers.
Source: Whito’s AI-recommended brands study. Counted 7 August 2026. 20 questions across flights, broadband, energy, current accounts and mobile, asked on both engines. Eight answers per category.
Can you check the businesses AI recommends?
Every figure here compares what an engine recommended with a public register. An unmatched name is a reason to check, not proof of anything.
15. 38 of 728 Google AI Mode recommendations, 5.2%, could not be verified as an active firm trading in the town named. Whito treats 5.2% as a floor: 31 close calls were resolved in the business’s favour.
Source: Whito’s unverifiable AI recommendations study. Answers counted 22 August 2026, checked 28 August 2026. Prompt: “Recommend a trustworthy [sector] in [town]”. 20 towns, 8 sectors. Checked against Companies House, then against public trading footprint where no company matched.
16. Only 305 of the same 728 recommendations could be tied by name to a live registered company.
Source: Whito’s unverifiable AI recommendations study. Checked 28 August 2026 against Companies House. Sole traders are not on the register, so an unmatched name is not a fake one.
17. 11.2% of Google AI Mode’s dentist recommendations, 11 of 98, could not be verified, the highest rate of the eight sectors.
Source: Whito’s unverifiable AI recommendations study. Answers counted 22 August 2026, checked 28 August 2026.
18. A single name match against Companies House failed 58% of the time on those 728 recommendations; four passes of verification failed 5.2% of the time.
Source: Whito’s trade names register check. Checked twice on 28 August 2026, same businesses, same register. Recommendations collected 22 August 2026.
19. 49% of AI-recommended electricians could be matched to an active registered company in or near the town. In 6 of the 15 towns, at least one engine recommended an electrician whose company was dissolved or in liquidation.
Source: Whito’s electrician AI study. Counted July 2026. 182 businesses from ChatGPT, Gemini and Google AI Mode, checked against Companies House.
20. Only 51% of 173 AI-recommended roofing firms could be verified as active registered companies in or near the town, and 36% had no Companies House match at all. 92% were not NFRC members.
Source: Whito’s roofer AI study. Counted 10 July 2026. ChatGPT, Gemini and Google AI Mode. Prompt: “Recommend a trustworthy roofer in [town].” 15 towns, 45 answers. Checked against Companies House and the NFRC member directory.
21. 36 of 79 roofers Google AI Mode recommended across 20 UK towns, 46%, could not be matched to any registered company on a one-pass check. In three towns the only local name match was dissolved or in liquidation.
Source: Whito’s AI-recommended roofers Companies House check. Answers counted 22 August 2026, checked 11 September 2026. One pass per name, so read 46% as how hard these firms are to verify, not a count of fake firms.
22. 69% of 77 AI-named Manchester businesses that could be checked matched an active company on Companies House. 23 had no matching company on the register at all.
Source: Whito’s multi-engine local recommendation study. Counted 14 September 2026. ChatGPT and Gemini, eight trades. Companies House does not cover sole traders.
23. 17 of 88 mortgage firms named by ChatGPT and Google AI Mode, 19.3%, could not be matched to an active FCA permission in, or reasonably serving, the city they were recommended for. 50 of the 88, 56.8%, were appointed representatives trading under a name that is not the authorised firm, which is lawful.
Source: Whito’s AI mortgage adviser FCA register check. Counted 8 September 2026. Prompt: “Who is the best mortgage advisor in [city]?” 12 cities, 24 answers. Checked against the FCA Financial Services Register the same week. Not found does not mean unauthorised.
24. In 0 of 24 answers did ChatGPT or Google AI Mode tell the reader to check a mortgage firm on the FCA register.
Source: Whito’s AI mortgage adviser FCA register check. Counted 8 September 2026. 12 cities, one answer per engine per city.
25. Asked for a trustworthy architect, Google AI Mode named 92 businesses and individuals; for 39 of them, 42%, no registered architect could be found on the Architects Register under the name the answer gave. The ARB register and RIBA were cited as a source in none of the 20 towns.
Source: Whito’s AI architect ARB register check. Counted and checked 10 September 2026. Prompt: “Recommend a trustworthy architect in [town]”. 20 towns. The register’s practice field is self-reported, so this understates true registration and alleges nothing.
26. 3 of the 12 CQC ratings Google AI Mode stated for care homes did not match the regulator’s own file. All 72 homes it named in 17 English towns matched a currently registered CQC care home; 7 of the 72 were rated Requires improvement.
Source: Whito’s AI care home study. Counted 8 September 2026. Prompt: “Recommend a trustworthy care home in [town]”. 20 towns. Checked against the CQC care directory file dated 2 September 2026, the ratings file dated 1 September 2026 and the homes’ live CQC pages.
27. 2 of the 10 roofing review profiles behind AI’s most trusted recommendations belonged to firms whose registered companies are dissolved, and 8 of the 10 scored the equivalent of 9.9 out of 10 or better. A firm can close a limited company and carry on as a sole trader, so a dissolved company is not proof of anything on its own.
Source: Whito’s review forensics study. Counted 10 July 2026. Ten firms sampled from the 173-firm roofer study, profiles on Checkatrade and TrustATrader. Patterns, not prevalence.
Where AI engines get their answers
What the engines cited or displayed beside their recommendations.
28. The business’s own website was cited in 143 of 159 Google AI Mode answers, 90%, and Google reviews or Business Profiles in 128, 81%. In at least twelve answers the engine cited the website of a firm it did not recommend, so 90% is not a pure self-citation rate.
Source: Whito’s AI citation sources study. Counted 22 August 2026. Google AI Mode only. Prompt: “Recommend a trustworthy [trade] in [town]”. 8 sectors, 20 towns, one answer missing.
29. Trustpilot appeared in 2 of those 159 answers, and no news outlet and no Companies House page appeared in any.
Source: Whito’s AI citation sources study. Counted 22 August 2026. Google AI Mode only.
30. Checkatrade appeared in 50 of 60 Google AI Mode trade answers, 83%, but in only 1 of 79 professional-service answers.
Source: Whito’s AI citation sources study. Counted 22 August 2026. Trades: electricians, plumbers, roofers. 20 towns.
31. A review platform sat behind Google AI Mode’s roofer recommendation in 100% of towns; for dentists it was 5%. Community sources, mostly local subreddits and town noticeboards, fed 51 of the 160 answers.
Source: The Whito AI Visibility Index. Counted 22 August 2026. Google AI Mode, 8 sectors, 20 towns, 160 answers, 734 businesses named. The Index counts 160 answers; the preserved logs used for the three lines above hold 159.
32. Reddit was ChatGPT’s most-cited source for a trustworthy local business, appearing in 10 of 18 answers. Gemini showed no sources at all in 13 of its 18 answers. Yell, Yelp and Facebook were cited zero times in the log.
Source: Whito’s AI sources study. Counted 19 July 2026. ChatGPT and Gemini. Prompt: “Recommend a trustworthy [business] in [town]”. 36 answers across six categories and three towns.
33. Google Business Profile listings were cited as a source in 17 of 20 towns’ care home answers; the CQC was cited in one.
Source: Whito’s AI care home study. Counted 8 September 2026. Google AI Mode, 20 towns.
34. 116 of the 184 source cards behind Google AI Mode’s architect answers, 63%, were Google Business Profile cards.
Source: Whito’s AI architect ARB register check. Counted 10 September 2026. 20 towns.
35. The recruitment trade bodies REC and APSCo were cited zero times across 18 AI answers about recruitment agencies.
Source: Whito’s recruitment agency AI study. Counted 3 September 2026. ChatGPT, Gemini and Google AI Mode, six cities. Gemini’s six answers showed no citations at all.
Ratings and adverts inside AI answers
Star ratings as the engines showed them, and the paid slots sitting under the answers.
36. 28 of the 32 mortgage firms ChatGPT gave a star rating had a perfect 5.0.
Source: Whito’s mortgage adviser AI study. Counted 17 August 2026. Six cities. Ratings as shown by ChatGPT, not verified by Whito.
37. 28 of 71 rated roofer recommendations in Google AI Mode carried a perfect 5.0.
Source: Whito’s AI-recommended roofers Companies House check. Counted 22 August 2026. 20 towns, 79 roofers, 71 with a rating card.
38. The median review count AI engines quoted was 368 for dental practices, 113 for law firms and 37 for accountancy firms.
Sources: Whito’s dentist AI study, solicitors and accountants. Counted 14 August 2026. Bases: 54 dental, 13 law and 21 accountancy recommendations that quoted review data; the accountants test used Perplexity in place of Google AI Mode. Figures as quoted by the engines, not audited.
39. 11 of 12 ChatGPT answers about mortgage advisers carried a sponsored placement.
Source: Whito’s AI mortgage adviser FCA register check. Counted 8 September 2026. ChatGPT free tier with web search, 12 cities.
40. Adverts sat beneath 5 of 6 ChatGPT answers about estate agents and 5 of 6 about recruitment agencies. For estate agents Whito saw no evidence the adverts changed which firms were recommended; for recruitment, no advertiser appeared in any organic list.
Sources: Whito’s estate agent AI study and Whito’s recruitment agency AI study. Counted 17 August 2026 (estate agents) and 3 September 2026 (recruitment). ChatGPT free tier, six cities each.
41. 12 of 20 ChatGPT answers to UK consumer buying questions carried a paid advert underneath.
Source: Whito’s AI-recommended brands study. Counted 7 August 2026. ChatGPT free tier, five categories.
Are UK business websites ready for AI crawlers?
What UK business websites tell AI crawlers, and what they give them to read.
42. 92% of 120 independent UK business websites had no AI crawler policy of any kind.
Source: Whito’s AI readiness audit. Counted 2 July 2026. 12 sites in each of 10 sectors, chains and franchises excluded. robots.txt, llms.txt, homepage structured data, visible FAQ and pricing checked.
43. 1 site in 120 had a hand-written llms.txt file; 20 of the 21 found were generated automatically by Wix, Shopify or SEO plugins.
Source: Whito’s AI readiness audit. Counted 2 July 2026. Same 120 sites.
44. Only about 6% of those 120 sites used correct business-type structured data.
Source: Whito’s AI readiness audit. Counted 2 July 2026. Schema counts are floors.
45. 1 of 49 UK business websites actively blocked any AI crawler, and 45 of 49 gave AI crawlers no instruction at all. 3 of the 28 independents, all on Wix, exposed a live MCP endpoint that lets AI agents read details and start a booking.
Source: Whito’s AI Door Test. Counted 20 to 26 June 2026. 21 national brands and 28 independents across seven cities. A deliberate spread, not a random national sample.
46. 19 of 55 top-rated UK electricians, plumbers and roofers, 35%, had no website at all, only a Google listing.
Source: Whito’s trades invisible to AI study. Counted June 2026. 55 top-rated trades from Whito’s seven-city directory research.
Do AI citations send visitors?
Whito’s own site, read from Microsoft’s and Google’s own tools. One website is not a market, which is why the base says so.
47. Microsoft Copilot and its partners cited whito.co.uk 17,400 times across 127 pages while Bing web search sent 167 clicks in the same window, 104 citations for every click.
Source: Whito’s AI citations and traffic study. 29 May to 28 August 2026, read 29 August 2026. Citations from Bing Webmaster Tools AI Performance (beta, which Microsoft labels a sample); clicks from its Search Performance report. One website, Whito’s own.
48. 518,000 Google impressions produced 1,410 clicks for whito.co.uk, a click-through rate of 0.3%.
Source: Whito’s AI citations and traffic study. Three months to 27 August 2026, read 29 August 2026 from Google Search Console. One website, Whito’s own.
Trust beyond AI: registers, directories and search
The same register checks applied to directories, ordinary search results and regulators’ own files.
49. None of nine standard Google searches for an electrician showed an AI answer, and Checkatrade, TrustATrader, Yell and MyBuilder were in the first results in all nine cities.
Source: Whito’s how the UK finds a tradesperson study. Counted June 2026. Google UK standard search, nine cities. Results are personalised and change over time.
50. 48 of 164 UK trade names from business directories and ordinary search, 29%, did not resolve to an active company on Companies House. Failure ran from 0% on one directory to 55% on another.
Source: Whito’s trade names register check. Counted July 2026 (directories 4 to 5 July). Five directories and ordinary search results, limited companies only for the directory sample. An upper bound.
51. 19 of 64 tradespeople found by ordinary search in eight UK cities, 30%, could not be verified as an active company. 2 were trading under a dissolved or wound-up company.
Source: Whito’s tradesperson verification study. Counted July 2026. Electricians, plumbers, roofers and builders, checked against Companies House. Some will be sole traders, who are not on the register.
52. 4 of 200 electrical installation companies could be found on the ICO register of data protection fee payers under their name, against 92 of 200 accounting and bookkeeping companies. Plumbing and heating companies were 7 of 200, hairdressing and beauty 10 of 200. A looser same-postcode check lifts the three trades only to about 8 to 10%.
Source: Whito’s ICO register check. Counted 10 September 2026. First 200 active companies per SIC code, matched against the full ICO register downloaded that day. Exemptions exist; a low rate is not an accusation.
53. 27,551 of 57,082 care services on the CQC register, 48.3%, had no website recorded. When Whito spot-checked ten blank home care entries, three had a working website the register did not hold.
Source: Whito’s CQC register study. CQC care directory file dated 26 August 2026, England only, read 30 August 2026.
54. 20 of 28 businesses advertising tooth whitening were not dental practices, and none of those 20 named a GDC-registered professional or published a GDC number on its website.
Source: Whito’s tooth whitening GDC check. Counted July 2026. Leeds, Manchester, Birmingham, London and mobile services. A small sample; no national percentage is claimed and no wrongdoing is alleged.
55. Only 3 of 15 genuine heating engineer websites checked showed a readable Gas Safe number.
Source: Whito’s heating engineer search study. Counted July 2026. Websites of local firms from first-page results in 15 towns. Numbers shown only inside logo images were not counted.
What these figures are, and what they are not
Unless a line says otherwise, each AI figure comes from one run per prompt. Lines 11 and 12 repeat prompts on purpose to measure how much the names move.
Samples differ, so lines are never added together. Where two figures sit in one line, they are shown side by side, not pooled.
Could not be matched means the checker could not find it. It does not mean a business is fake. Sole traders do not appear on Companies House.
The engines used differ by study and are named on each line.
The study page is the authority. If a study is corrected, the matching line here changes the same day, with a dated note under Corrections.
Survey figures from other publishers are left out on purpose. For those, Whito keeps separate pages: UK search engine statistics and UK AI adoption statistics.
What to do with this
Before paying anyone to make a business visible in AI answers, look at what the engines above were reading, the website and the Google listing, and at what they mostly skipped, the company record and the register entry. Then check those four agree with each other.
That costs nothing.
The free Whito Visibility Check runs the website side of it. Seventeen checks, about twenty seconds, no email, no sales call.
The files behind the studies
Several of these studies have a full sector file on the Intelligence shelf: every business named, every register result, town by town. £1,280 per edition, and the studies themselves stay free.
Corrections
None. First published 15 September 2026.
How to cite this page
Whito, “UK AI Search and Trust Statistics 2026”, whito.co.uk/research/uk-ai-search-statistics/, checked 15 September 2026. Free to quote with a link to this page or to the study named beside the figure. Quote the collection date with any number. Each line number is a link to that line.
Common questions
Do AI engines agree on which UK business to recommend?
Not often on the top pick. In Whito’s June 2026 test, two of ChatGPT, Gemini and Perplexity shared a top pick in 1 of 12 identical questions, and all three agreed in none. Across six professions tested in August 2026, between 55% and 85% of the firms named were named by only one engine; line 3 on this page shows how each figure was derived.
Can you trust a business an AI engine recommends?
Check it first. Of 728 Google AI Mode recommendations collected on 22 August 2026, 38 (5.2%) could not be verified as an active firm trading in the town named, after four verification passes. On a single name match, 58% of the same recommendations failed. Sole traders are not on Companies House, so an unmatched name is a reason to check, not a verdict.
Where does Google AI Mode get its local recommendations from?
Mostly from businesses’ own websites and Google. In 159 answers collected on 22 August 2026, the business’s own website was cited in 143 (90%) and Google reviews or Business Profiles in 128 (81%). In at least twelve answers the cited website belonged to a firm the engine did not recommend. Trustpilot appeared in 2 answers and no news outlet appeared in any.
Do AI citations send traffic to a website?
On Whito’s own figures, citations and clicks are very different numbers. Microsoft Copilot and its partners cited whito.co.uk 17,400 times between 29 May and 28 August 2026, while Bing web search sent 167 clicks in the same window: 104 citations for every click. That is one website, not a market.
Are these survey figures?
No. Every figure on this page was counted by Whito from AI answers it collected, websites and search results it checked, registers it matched against or files it read, and each line names its date, base and source study. Survey figures from other publishers are kept on separate Whito pages.
- How accurate are AI business recommendations? The UK evidence, measured
- The Whito AI Visibility Index
- UK electrician statistics
- UK roofing statistics
- UK care sector statistics
The structural checks behind this research run on any website in about twenty seconds. Enter yours and read the full result on the page. Free, and we do not ask for your email.
Run the free Visibility CheckIf your business is named here and you think we have a fact wrong, email hello@whito.co.uk and we will check it and correct it. If you want to respond, send us a statement and we will publish it alongside. Full detail: Corrections, right of reply and removals.
