Checkatrade was cited in 81 of 120 OpenAI web-search answers when I asked for plumbers across 20 UK cities. That's 67.5% of the answers, but I wouldn't use that figure to tell a plumber to buy a listing.
I wanted to find out whether businesses with more directory listings were more likely to be recommended. What I found gives us a useful picture of the sources appearing in these answers. The evidence wasn't strong enough to answer the original question about the businesses themselves.
What I tested across 20 UK cities
The study started with 500 Google Maps listings classified as plumbing providers, 25 in each of 20 selected UK cities. These were sampled listings and locations, not 500 independently verified plumbing firms. I then ran two versions of a request for local plumbing recommendations, repeating each one three times per city. That produced 120 answers.
The tests used OpenAI's gpt-6-luna model through its API, with web search enabled, on 25 September 2026. Each request started with no conversation history and no list of candidate businesses. This wasn't a test of the consumer ChatGPT app, and it doesn't establish what every AI search tool would recommend.
How often the six directories were cited
I counted the directory domains appearing in the answers' actual citations. Each directory counted at most once per answer, even if the answer linked to it several times.
View exact counts and percentages
| Directory | Answers | Share |
|---|---|---|
| Checkatrade | 81 / 120 | 67.5% |
| Yell | 22 / 120 | 18.3% |
| MyBuilder | 12 / 120 | 10.0% |
| TrustATrader | 11 / 120 | 9.2% |
| Rated People | 3 / 120 | 2.5% |
| Thomson Local | 0 / 120 | 0.0% |
Checkatrade appeared much more often than the other five directories in this panel. Taken together, at least one of the six appeared in 96 of the 120 answers, or 80%. You can't add the bars to get that figure because the counts overlap. These are answers citing a source, not shares of the plumbing businesses being recommended.
Which? Trusted Traders also appeared in 23 answers. It wasn't part of the six-directory panel chosen for the listing comparison, so I've kept that observation separate. It's a useful reminder that the sources an AI answer cites can extend beyond the list you originally set out to investigate.
A citation doesn't prove a listing helped
If I were looking into AI visibility for a plumber, I'd want to inspect which Checkatrade pages appeared and what those pages said about the relevant area. Its presence in these answers is worth investigating.
But a directory being cited doesn't tell us whether joining it gets your business recommended. A city or category page can appear as a source without establishing that the answer used an individual plumber's profile. Even seeing a profile cited wouldn't, by itself, prove that the listing caused the recommendation.
This is where the business comparison ran into trouble. Matching the same business across Maps, directory results and AI answers wasn't reliable enough. Some names were too ambiguous, and some possible matches couldn't be resolved confidently from the retained evidence.
Only 173 of the 500 sampled locations had complete measurements for the main comparison. The study had set a minimum of 400 before interpreting that comparison, and the other checks for whether it could produce a useful answer also failed. So I can't reliably say whether businesses with more discoverable directory profiles were recommended more often.
It doesn't show that directories are useless either. It leaves that question open. I'd rather tell you that than turn a source-citation figure into a promise about your business.
The API work cost about US$4.01
That breaks down as US$1.60 in confirmed DataForSEO charges and about US$2.41 in estimated OpenAI charges, including the technical preflight. That's API usage, not the full cost of doing the research. It excludes storage, other infrastructure and time, and the OpenAI estimate isn't a final invoice.
The dated results, source evidence and cost records are retained privately with a local backup. These findings describe one collection window in selected cities, not a national benchmark or a trend over time. Research collection, analysis and article drafting used AI assistance.
What I'd check before paying for another listing
If I were advising a plumber, I'd start with three fairly simple checks.
- Save a few local search results.
Ask for the service you provide in the area you cover. Save the exact question, date, tool, businesses named and pages cited. Repeat it a few times to see whether the result changes. Treat this as a snapshot of that tool, not a prediction of every customer's experience.
- Check the profiles you already control.
On any sources appearing, look at your business name, service area, website link and services offered. I'd do that as basic housekeeping, not because this study proved that editing a profile would improve AI recommendations.
- Judge a paid directory on enquiries and profitable jobs.
Consider what it can bring you alongside its cost and terms. My study measured neither enquiries nor sales. A directory's citation rate doesn't tell you whether membership will pay for itself.
Before paying for a listing on the strength of an AI visibility claim, I'd ask the provider to show how they're measuring recommendations of your business, then how they'll connect that visibility to actual enquiries. A count of answers citing their directory won't answer either question.