W
Reviewed by Jacob Whitmore, Whito · Fact-checked for accuracy
This research is free to read and the method is published in full. The complete dataset behind studies like this one, every business named and the raw answer logs, is available as a report. Whito Intelligence

Finding

Thirty identical ChatGPT questions asking for a trustworthy plumber in three UK towns produced 180 recommendations naming 51 different businesses in Whito's July 2026 study, and only one business was named every time. 17 of the 51 were named exactly once.

Source: Whito, whito.co.uk/research/ai-recommendation-lottery-study, July 2026. Free to quote with a link to this page.

Last Updated on September 5, 2026

Ask ChatGPT to recommend a trustworthy plumber. Then ask again. Then ask eight more times.

We did exactly that, word for word identical, in three UK towns. Thirty questions in total.

ChatGPT named 51 different plumbers.

In a July 2026 Whito study, 30 identical ChatGPT questions across three UK towns produced 180 recommendations naming 51 different plumbers, and only one business was named every time.

If AI has recommended your business this week, that is nice. It probably will not tomorrow. Here is what we found, and what it means if you want customers to actually find you.

The test

On 17 July 2026 we asked ChatGPT (free tier, the version most customers actually use) the same question ten times in a row, each time in a fresh chat: “Recommend a trustworthy plumber in [town]”. We ran it in Nottingham, Luton and Preston, three towns from the same panel we used in our electrician study.

Then we ran a second test in five more towns (Reading, Leeds, Swansea, Doncaster and Norwich): the same question asked four ways. Plain. With “please” and “thank you”. As a pensioner. As a landlord. Twenty more questions.

Test one: the recommendation lottery

Thirty identical questions produced 180 recommendations naming 51 different businesses.

Only one business out of 51 was named every time. Aquasure Plumbing in Nottingham appeared in all ten of that town’s answers. Nobody else managed it. In Luton and Preston, not a single business appeared in all ten.

A third of the businesses recommended (17 of 51) were named exactly once. One appearance across ten asks, then gone.

In Preston, two consecutive identical questions produced almost completely different answers. Same wording, seconds apart, different plumbers.

The town-by-town numbers:

TownDifferent plumbers named (10 asks)Named in all 10 answersMost consistent business
Nottingham151 (Aquasure Plumbing)10 of 10
Luton1809 of 10
Preston1808 of 10
Total511 of 51

One Luton answer also included a plumbing supply shop on its recommendation map. We asked for a plumber. It suggested somewhere to buy a tap.

Test two: the politeness test

Changing how you phrase the question changed roughly 70% of the names recommended.

In two of the five towns, adding “please” and “thank you” produced an answer with zero businesses in common with the plain question. In Reading, the plain question shared no names at all with any of the other three phrasings.

And one for the pub quiz: Doncaster Plumbing & Heating Ltd was recommended to the polite asker, the pensioner and the landlord, but never to the person who just asked plainly.

To be fair to the tool, the advice around the names did adapt sensibly. Pensioners got warnings about upfront payments and rogue traders. Landlords got talk of invoicing, tenant communication and compliance. The framing is smart. The actual businesses churn like a raffle.

What this means if you run a business

Two things, and they pull in opposite directions.

First: a single AI recommendation means almost nothing. If someone tells you ChatGPT recommends your business, that was one draw of the lottery. Fifty of the 51 plumbers in our test would be wrong to assume they “rank” on AI. There is no rank. There is a pool, and every question shuffles it.

Second: the pool itself is not random. The businesses that came up again and again share a profile: large review volumes, presence on checked directories (Checkatrade, ThreeBestRated, Which? Trusted Traders came up constantly as ChatGPT’s sources), consistent business information everywhere they are listed. The lottery has weighted tickets. Structure buys you more of them.

That is the same pattern we found with electricians, where the three big AI tools could only agree on 6% of businesses, and some recommendations turned out to be dissolved companies.

How to get more tickets in the draw

You cannot make AI recommend you every time. Nobody can, and anyone selling you guaranteed AI rankings is selling smoke. What you can do is show up in more draws:

1. Review volume wins. The most-recommended businesses in every town had the deepest review histories. Ask every satisfied customer. This is the single strongest signal.

2. Be on the directories AI actually reads. ChatGPT cited checked trade directories in almost every answer. If you are not listed and checked somewhere it trusts, you are not in the pool.

3. Keep your details identical everywhere. Same name, same number, same area on every listing. Inconsistent information is how you end up as the plumbing supply shop on the map.

This is exactly why we built the Whito directory: a place where UK businesses get found and checked, so that when the lottery draws, you hold more tickets. Listing is free.

Method notes

All 50 questions were asked on 17 July 2026 on ChatGPT free tier in fresh chats, exactly as a real customer would. During collection ChatGPT occasionally dropped to its lighter fallback model and at one point rate-limited us entirely. We recorded everything as-is, because that is what a free-tier user actually experiences. Full raw logs, including every business named in every answer, are available on request.

Cite this research

Whito Research (2026). We Asked ChatGPT for a Plumber 30 Times. It Named 51 Different Companies. Whito. https://whito.co.uk/research/ai-recommendation-lottery-study/

Key finding: 30 identical ChatGPT queries produced 51 different plumber recommendations. AI recommendations are a lottery with weighted tickets: review volume, checked directory listings and consistent details buy more tickets.

This is original Whito research. You are welcome to reuse these figures with a link to this page as the source.

The takeaway

AI recommendations are not a ranking. They are a lottery with weighted tickets.

You cannot win the draw. You can only own more tickets: reviews, checked listings, consistent details. Structure before scale, as ever.

Related reading
Where does your own business stand?

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 Check
Corrections and right of replyThis page reports research that may name real businesses. Data is collected and checked on the dates stated. Company financials, insolvency and regulator action are taken from Companies House, company announcements or the regulator’s own notice.

If 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.
author avatar
Jacob Whitmore Founder and Lead Analyst
Jacob Whitmore founded Whito. He works on what UK businesses are actually charged for marketing, and on getting those numbers published with the date they were checked.