Last Updated on August 25, 2026

Whito Research
What AI actually cites when it recommends a UK local business
We counted every source named in 159 AI answers across eight sectors and twenty towns. The two sources that decide almost everything are the two you already own.
Most of the “get found by AI” advice sold to UK small businesses starts in the wrong place. It points at review platforms, PR, directory subscriptions and backlink packages. All of it assumes AI goes looking for reputation signals scattered across the web.
So we checked. Every answer in the Whito AI Visibility Index was logged with the sources the engine actually cited or named. That is 159 answers, one prompt each, collected on 22 August 2026. We went back through every one of those source lines and counted them.
The result is not close.
The mistake most businesses are making
The mistake is treating AI visibility as a new channel that needs new spending. It is not a channel. It is a summary of what the web already says about you, and the web mostly says what your own website and your Google profile say.
If those two are thin, wrong or out of date, no amount of directory spend fixes it. If they are strong, most of the rest is optional.
That is structure before scale, and the data says it plainly.
An answer is counted once per source type, whether the type appeared once or five times.
| Source type | Answers | Share of 159 |
|---|---|---|
| The business’s own website | 143 | 90% |
| Google (reviews and Business Profiles) | 128 | 81% |
| Trade directories and review platforms | 85 | 53% |
| Regulators, registers and trade bodies | 71 | 45% |
| Reddit and Facebook | 51 | 32% |
| Auto-generated “best of” pages | 32 | 20% |
| News outlets and press coverage | 0 | 0% |
Percentages do not total 100 because most answers cite more than one type.
Every named source we found
Eighty-four separate external sources appeared across the study. Here is everything that turned up in more than one answer, plus the ones people expect to see and we did not find.
| Source | What it is | Answers | Share |
|---|---|---|---|
| The business’s own website | Owned | 143 | 90% |
| Owned profile plus reviews | 128 | 81% | |
| Checkatrade | Paid trade directory | 62 | 39% |
| Forum | 38 | 24% | |
| TrustATrader | Paid trade directory | 26 | 16% |
| Groups and pages | 18 | 11% | |
| Which? Trusted Traders | Paid trade directory | 15 | 9% |
| ThreeBestRated | Auto-generated “top 3” pages | 14 | 9% |
| WaterSafe | Register | 11 | 7% |
| MyBuilder | Lead platform | 10 | 6% |
| NICEIC | Register | 10 | 6% |
| Gas Safe Register | Statutory register | 10 | 6% |
| RCVS | Regulator | 10 | 6% |
| ICAEW | Professional institute | 8 | 5% |
| ACCA | Professional institute | 7 | 4% |
| GetAgent | Agent comparison site | 6 | 4% |
| TrustMark | Government endorsed scheme | 6 | 4% |
| Yell | Directory | 4 | 3% |
| Rightmove | Property portal | 4 | 3% |
| CQC | Regulator | 4 | 3% |
| bestlocalrated.co.uk | Auto-generated listings | 4 | 3% |
| Trustpilot | Review platform | 2 | 1% |
| Yelp | Review platform | 0 | 0% |
| Companies House | Statutory register | 0 | 0% |
| Any news outlet | Press | 0 | 0% |
Six things the count shows
1. The two assets you own decide almost every answer
The business website appeared in 143 answers. Google appeared in 128. Both appeared together in 115. Only three answers out of 159 used neither.
Neither of those is a platform you rent. One you build, one you claim. Both are free to keep accurate.
2. Checkatrade is a trade tax, not a UK business tax
Checkatrade appeared in 50 of the 60 trade answers, which is 83%. Across the 79 professional service answers, vets, dentists, accountants and estate agents, it appeared exactly once, and that was an estate agent answer pulling fee data from it.
If you are a plumber, roofer or electrician, directory presence is close to unavoidable. If you are an accountant paying for trade directory listings on the theory that AI reads them, the data does not support the spend.
| Sector | Answers | Own site | Checkatrade | Any directory | Reddit or Facebook | “Best of” pages | |
|---|---|---|---|---|---|---|---|
| Electricians | 20 | 75% | 75% | 85% | 90% | 20% | 20% |
| Plumbers | 20 | 85% | 85% | 80% | 95% | 30% | 15% |
| Roofers | 20 | 75% | 80% | 85% | 100% | 35% | 40% |
| Cleaners | 20 | 90% | 55% | 55% | 65% | 40% | 15% |
| Vets | 19 | 95% | 95% | 0% | 16% | 42% | 5% |
| Dentists | 20 | 100% | 95% | 0% | 5% | 20% | 15% |
| Accountants | 20 | 100% | 90% | 0% | 15% | 15% | 40% |
| Estate agents | 20 | 100% | 70% | 5% | 40% | 55% | 10% |
3. Regulators get name-checked, not read
A regulator, register or professional body appeared 111 times across the study. In 95 of those appearances it was simply named in the text as a thing to look for. In only 16 was it cited as a source the engine had actually drawn on.
Gas Safe Register appeared in ten answers and was never once a cited source. RCVS, the same. ACCA, TrustMark, CORC and the Property Ombudsman, the same.
The practical version of this: the engine tells the customer to check the register, but does not check it itself. Your registration number is worth publishing on your own site, where the engine will read it, rather than assuming the register will speak for you.
4. Reddit and Facebook shape a third of answers
Fifty-one answers, 32%, drew on Reddit or Facebook. It is highest in estate agents at 55% and vets at 42%.
This is where negative sentiment enters. In the estate agent sector the engine relayed warnings about named chains, sourced to Reddit threads, and in one town recommended as first choice the same chain it had flagged as a red flag in another.
You cannot buy your way out of this and you should not try to game it. You can know what the threads in your town say about you, because right now the engine does and you probably do not. We have written before about why Reddit earns its place in these answers.
5. One in five answers leans on a page nobody wrote
Thirty-two answers cited an auto-generated “best of” page. These are sites like ThreeBestRated, bestlocalrated.co.uk and a long tail of single-purpose domains such as topratedroofers.co.uk, dentistlocator.co.uk and accountantsup.co.uk.
They exist to rank, not to inform. Nobody visited them to choose a tradesperson. The engine reads them anyway, which means a page assembled by a script can put a competitor in front of a customer who asked an honest question.
6. The platform you are being sold is not the one being read
Trustpilot appeared in 2 answers out of 159. Yelp did not appear once. Companies House did not appear once. No news outlet or newspaper appeared at all, in any sector, in any town.
That last one matters if you have been quoted a price for press coverage on the promise it will improve your AI visibility. In this dataset, press did not feed a single recommendation.
The one that should worry you most
“The business’s own website” does not always mean the recommended business’s website.
In at least twelve answers, the engine cited the website of a company it did not recommend. In the Belfast dentists answer, all three cited practice websites belonged to practices other than the three it recommended. In the Nottingham roofers answer, one cited source was a roofing company’s own blog, used as neutral background while four other firms were named.
Read that the right way round. If your site explains the job well, it can end up as the evidence behind a recommendation for someone else. That is a wasted asset, and it is fixable.
What to do about it
Work in order. Skipping stages is how the money gets wasted.
Start
Claim and complete your Google Business Profile. Correct category, correct service area, correct hours, real photos, and reviews you actually ask for. It is in four out of five answers and it costs nothing.
Then fix your own website so it answers the question a customer asks. Where you work, what you charge or how pricing works, what you are registered with and the registration number, and who you are. The engine reads sentences, not slogans.
Build
If you are in a trade, treat Checkatrade or TrustATrader as a distribution cost and check your profile is accurate and rated. If you are in a professional service, do not buy trade directory listings for AI visibility. There is no evidence here they earn it.
Search your own town’s subreddit and local Facebook groups for your business name and your competitors’. What you find there is already in the answer.
Put your credentials on your own pages in plain text. Gas Safe number, NICEIC membership, ICAEW firm number, whichever applies. The register itself will not do the talking.
Scale
Track which sources feed answers in your sector and town over time, because they move. Then decide spend against that, rather than against whatever a supplier is selling this quarter.
Press and reputation work still have a case, on their own merits. Just not on this one.
Want this for your sector and your towns?
Full sector detail: every business named, every town, every engine, every source. £595, built to order, delivered within one working day.
Method
This study is a second pass over the dataset behind the Whito AI Visibility Index, Q3 2026. No new answers were collected for it.
One prompt shape, used identically everywhere: “Recommend a trustworthy [trade] in [town]”. One run per prompt, no retries, no regeneration. Google AI Mode, signed in, UK, udm=50. Eight sectors: electricians, plumbers, roofers, cleaners, vets, dentists, accountants, estate agents. Twenty towns: Reading, Sunderland, Swansea, Leeds, Nottingham, Coventry, Luton, Bristol, Doncaster, Ipswich, Preston, Stoke-on-Trent, Milton Keynes, Norwich, Plymouth, Glasgow, Belfast, Mansfield, Blackpool, Wolverhampton. All answers collected on 22 August 2026.
Every answer was logged at collection with the sources it cited or named. For this study each of those source lines was transcribed and counted twice, once directly and once by an independent recount working only from the original logs. The two counts were reconciled before publication and the differences are set out below.
Counting rules. A source counts once per answer regardless of how often it appears in it. “Own website” covers any business or brand website or profile cited in the answer, including businesses that were not recommended. “Cited” means the source was drawn on. “Named” means the source was mentioned in the text without being drawn on, and those are counted separately wherever the distinction matters.
Limitations, stated plainly
One engine. This is Google AI Mode only. ChatGPT, Gemini, Perplexity and Copilot will differ, and a Perplexity comparison is planned for the Q4 edition. Do not read these figures as “AI” in general.
159 answers, not 160. The vets sector has 19 towns, not 20. The Wolverhampton vets answer failed to generate and was never captured, so it is not in the dataset. Every percentage here uses 159 as the denominator.
One run per prompt. AI answers vary between runs. A single run per town captures what one user would have seen, not an average.
Definitional edges. The regulator count of 71 answers uses a strict definition covering statutory registers, licensing schemes, council trusted trader schemes and professional institutes. A wider definition that also counts the Competent Person Scheme and an NHS health board portal gives 73. The “best of” pages count of 32 is the strict figure. Counting comparison and lead-generation sites such as GetAgent, HaMuch and WhatClinic in the same bucket gives up to 46. We have published the conservative number in both cases.
Source attribution is as the engine presented it. Where an answer named a source without linking it, we recorded it as named rather than cited, and said so.
Whito, “What AI actually cites when it recommends a UK local business”, August 2026. Based on the Whito AI Visibility Index, Q3 2026: 159 Google AI Mode answers across 8 sectors and 20 UK towns, collected 22 August 2026. Underlying town-level source logs are available to journalists and researchers on request.
What to do with this
If your own site is the source cited in nine answers out of ten, the practical question is what it actually says that a machine can use. We have set out the eight most common gaps, each with its measured frequency and a free fix, in common mistakes that make UK businesses invisible to AI.
Common questions
No. It means the answer depends on your sector. In the trades, a directory appeared in 95% of answers and Checkatrade alone in 83%. In professional services, directories appeared in 19% of answers and Checkatrade in one out of 79. Same spend, very different return.
We can only report what the engine cited, not why. What we can say is that in 159 UK local business answers, Trustpilot was cited twice and Google was cited 128 times. If you are choosing where to put your review-gathering effort for local search, that gap is the finding.
Almost certainly not, and we have not measured them. This is Google AI Mode only. A second engine is planned for the Q4 2026 edition of the index.
There is no submission process and nobody can guarantee it. What the data shows is that the website is the most cited source type in the study at 90%, so the practical step is making sure yours states your service area, your pricing approach and your registrations in plain sentences a machine can read.
The underlying index is collected and published quarterly. The next edition is December 2026, and the source breakdown will be recounted with it.
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.
Check your business freeIf 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.

