When BrightLocal pulled apart 26,672 AI citations for plumber searches, 54% pointed at Google Business Profiles, 13.8% at Yelp, and the plumber's own website was somewhere in the remaining third.
That is from a study of 1.9 million citations published in August 2026, and it is the closest thing we have to a map of where AI engines go when a homeowner asks who to call.
We have not run our own 50-prompt test yet, so this article does not pretend to. It reports what the published studies found when buyer questions went to AI engines for trades and local services, with every number labelled by source. Then it sets out the 50-prompt method a Northampton plumber, electrician or roofer can run in an afternoon, with the prompts written out and a table to record the results.
What the published studies found
Four studies between December 2024 and August 2026 ran local or trade prompts through AI engines and published their numbers. None is perfect and only one touches the UK, so the caveats sit next to each figure.
The biggest is BrightLocal's August 2026 analysis. It ran five prompts at nine map points around each of 1,355 business locations in the US, UK and Australia, 60,970 checks in all, across Google AI Overviews, Google AI Mode and ChatGPT. Home and trade services were 10.98% of the locations. It did not test Gemini or Perplexity, and it gives no separate UK figures.
Across every category, Google Business Profile pages took 28.63% of all citations and appeared for 94.17% of businesses. Yelp took 9.53%. Around 93% of the distinct domains cited were individual business websites, which sounds encouraging until you notice that is a share of domains, not of citations. Lots of business sites were cited once. A handful of platforms were cited hundreds of thousands of times.
The engines diverged sharply. The shares below are worked out from the raw citation counts BrightLocal published for each engine.
| Engine (BrightLocal, Aug 2026) | Top cited source | Share of that engine's citations | Second source |
|---|---|---|---|
| Google AI Overviews | Google Business Profile | about 67% (108,018 of 161,094) | Yelp, about 7.6% |
| Google AI Mode | Google Business Profile | about 53% (455,123 of 861,263) | Yelp, about 8.8% |
| ChatGPT | Yelp | about 61% (99,281 of 162,966) | Tripadvisor, about 6.3% |
BrightLocal puts ChatGPT's Yelp habit down to the OpenAI and Yelp partnership, and notes Yelp "appeared as a source in 80% of all answers from ChatGPT". Google Business Profile did not make ChatGPT's top five at all. For a UK trade with no Yelp listing, that is the most useful line in the study.
For plumbers specifically, the split was Google Business Profile 54%, Yelp 13.8%, Facebook 3.1%, Angi 2.2%, then HomeAdvisor, BBB, Trustpilot and MapQuest at 0.9% each. Electricians and builders are not broken out.
xychart-beta
title "Share of 26,672 AI citations for plumber searches (BrightLocal, Aug 2026)"
x-axis ["Google Business Profile", "Yelp", "Facebook", "Angi", "Trustpilot", "BBB"]
y-axis "Per cent of citations" 0 --> 60
bar [54, 13.8, 3.1, 2.2, 0.9, 0.9]
The second study is BrightLocal's earlier ChatGPT work from December 2024: 800 manual ChatGPT searches, 20 verticals, 20 US cities. Business websites were 58% of sources overall and 72% for transactional terms such as "plumber", against 43% for "best plumber" style questions. The one trade outlier was "best electrician", where directories took 62% of sources. This predates the Yelp partnership, which is probably why the two BrightLocal studies disagree so much on ChatGPT.
The third is from a PR agency, 5W Public Relations, published in 2026. It ran more than 320 consumer prompts ("best plumber in [metro]", "emergency dentist [metro]") through ChatGPT, Claude, Perplexity, Gemini and AI Overviews across ten US cities and six categories including plumbing, electrical and HVAC. Its headline: 78% of independent local operators had effectively zero citation share. Fewer than 10 of about 50 independent plumbers tested appeared in any citation, and Roto-Rooter won emergency plumber prompts at a rate "approaching 80%".
Treat those numbers with care. 5W sells generative engine optimisation, calls its shares estimates, does not publish per-engine sample sizes, and uses its 22% figure two different ways on the same page. The direction of the finding (chains and aggregators crowd out independents) matches BrightLocal. The precision does not deserve the same trust.
The UK evidence is thin, and what there is points at directories
Only one published piece ran trade prompts in UK towns, and it is small. Whito, a UK research publisher, audited nine cities in June 2026: a standard Google search for an electrician in each, plus two Google AI Mode prompts, "best electricians in Leeds" and "best plumbers in Manchester".
Checkatrade, TrustATrader, Yell and MyBuilder appeared in the Google results for all nine cities. The AI Mode answer for Leeds named four firms that did not match Google's local results, cited Reddit, ranked a 4.3-star firm first, and placed a Pudsey business in Northampton, 43 miles from where it actually was. The Manchester plumber answer overlapped with Google's local pack on two of four names.
"The AI is a confident guess stitched from reviews, registers and forum posts, and it is not always right." Whito, June 2026
Two prompts is not a dataset. Whito's follow-up guide says later rounds across 15 UK towns found the engines "rarely agree with each other" and that a firm can be recommended while dissolved at Companies House, but the prompt lists and counts behind that are not published.
The only UK figure with any scale is not trade-specific. LLM Listed checked 500 UK business websites across ChatGPT, Claude, Gemini and Perplexity and reported that 8% appeared consistently across all four, with 76% outranked by a competitor. No methodology was published and the firm sells AI visibility services, so it is a signal of scale, not a measurement.
On the demand side, BrightLocal's March 2026 consumer survey of 1,002 US adults found 45% had used an AI tool for a local business recommendation in the past year, up from 6% the year before. ChatGPT was used by 31% and Google AI Mode by 23%. Again US only. I could not find an equivalent UK trade survey.
Why one prompt is not a test
One prompt in one engine on one day tells you almost nothing, because the sources behind AI answers move constantly. Semrush tracked 3,000 keywords that triggered AI Overviews every day through October 2024 and found 0% kept the same cited URLs for the month. The URLs changed on 56% of the days an Overview appeared, and only 6.7% of the URLs from the first week were still there in the last.
That volatility is why the method below asks for 50 prompts, four engines and a repeat run a fortnight later. You are measuring a pattern, not a position. If your firm shows up in 12 of 200 answers in week one and 14 in week three, you have a baseline. If a competitor shows up in 80, you know the size of the gap. Running answers many times is the discipline the survey of 45 generative engine optimisation studies insists on before any number means something.
The 50-prompt test for a Northampton trade business
The test is 50 prompts, ten per buyer intent, run through ChatGPT (with search on), Google AI Mode, Gemini and Perplexity, with every answer logged in one spreadsheet. The prompts below are written for a plumber. Swap the trade and keep the structure; a dental practice or private clinic can run exactly the same test.
Before you start, open a fresh chat for every prompt, use an account that has never discussed your business, and make sure the engine knows you are in Northampton (say so in the prompt, or check the location setting). Record the date. Four engines by 50 prompts is 200 answers. At two minutes each that is an afternoon.
Emergency intent (the calls that pay best):
- I have a burst pipe in Northampton right now, who can come out tonight?
- Emergency plumber Northampton, 24 hour
- My boiler has stopped working in Kingsthorpe and it is freezing, who do I call?
- Who does emergency plumbing in Wellingborough on a Sunday?
- Water leaking through my ceiling in Duston, need a plumber today
- Emergency plumber near Northampton General Hospital
- Toilet overflowing, Northampton, need someone within the hour
- Gas smell in my kitchen in Northampton, who is Gas Safe registered and available now?
- Emergency heating engineer Kettering tonight
- No hot water, Northampton, same day plumber
Comparison intent:
- Who are the best plumbers in Northampton?
- Compare the top three plumbing companies in Northampton
- Checkatrade or a local independent plumber in Northampton, which should I use?
- Best boiler installers in Northamptonshire
- Which Northampton plumbers do landlords recommend?
- Most reliable plumber in Northampton according to reviews
- Best bathroom fitters in Northampton
- Who is better for a new boiler in Northampton, a British Gas engineer or a local firm?
- Top rated heating engineers in Daventry
- Best plumber for an older house in Abington, Northampton
Price intent:
- How much does a plumber charge per hour in Northampton?
- Cost of a new combi boiler fitted in Northampton 2026
- What is a fair call-out fee for an emergency plumber in Northampton?
- How much to replace a bathroom in Northampton, and who gives fixed quotes?
- Cheapest reputable plumber in Northampton
- Boiler service price Northampton
- Is £90 an hour reasonable for a plumber in Northamptonshire?
- Which Northampton plumbers publish their prices online?
- Cost to fix a leaking radiator in Northampton
- Do any Northampton plumbers offer boiler finance?
Trust intent:
- Which plumbers in Northampton are Gas Safe registered?
- Northampton plumbers with the most Google reviews
- Are there any Northampton plumbers I should avoid?
- Which Northampton plumbing firms are Which? Trusted Traders?
- Family run plumbing business Northampton
- Northampton plumber that offers a written warranty on its work
- How do I check a Northampton plumber is legitimate?
- Plumbers in Northampton that have been trading more than ten years
- Is (your own firm's name) a good plumber?
- Is (your main competitor's name) a good plumber?
Local intent:
- Plumber near me (with location set to Northampton)
- Plumbers in Far Cotton, Northampton
- Heating engineer covering Towcester and Brackley
- Plumber in Moulton or Boughton, Northampton
- Who covers Earls Barton for plumbing?
- Plumbers in Northampton town centre
- Local plumber Weston Favell
- Plumber that covers Rushden and Higham Ferrers
- Northampton plumber who works in Sixfields and Upton
- Plumbers serving NN1 to NN6
Prompts 39 and 40 are the only branded ones. Everything else is unbranded on purpose, because a buyer who already knows your name is not the buyer you are losing.
How to record the results
Log every one of the 200 answers in a table with the columns below, and score each row the same way so the totals mean something.
| Column | What to enter |
|---|---|
| Prompt number and intent | 1 to 50, plus emergency, comparison, price, trust or local |
| Engine | ChatGPT, AI Mode, Gemini or Perplexity |
| Date and run | Run 1 or run 2, with the date |
| Businesses named | Every firm named, in the order given |
| You named? | Yes or no |
| Your position | 1st, 2nd, 3rd, or further down |
| Sources cited | Each domain in the sources list or footnotes: your site, Google Business Profile, Checkatrade, Yelp, Reddit, a competitor's site |
| Your site cited? | Yes or no, separate from being named |
| Facts about you | Correct, wrong, or absent (address, phone, Gas Safe status, hours) |
| Wrong location? | Yes if a firm was placed in the wrong town, as in Whito's Pudsey case |
Being named and being cited are different things, which is why the template separates them. An engine can recommend you from a Checkatrade page without touching your website, and it can cite your website in a sources list without recommending you.
Run the set again two weeks later. Then add up, per engine and per intent, how many of the ten prompts named you, how many named your nearest competitor, and which three domains were cited most often.
Reading the totals
The source column decides what you do first. If Google Business Profile dominates the AI Mode and Overviews rows, as the BrightLocal data says it will, the profile is the job: categories, service areas, photos from real jobs, and recent reviews. Whito's line, "ten reviews from this year beat fifty from 2020", is a judgement, not a measured result, but it is the right instinct.
If Checkatrade, TrustATrader or MyBuilder keep appearing in the ChatGPT and Perplexity rows and you are not on them, the engines are reading those directories as the register of who exists. That is a listings decision, and it belongs in the same plan as the rest of your local search work.
If you are named but your facts are wrong, fix the mismatch between your website, your profile and your directory entries first. A wrong phone number in an AI answer is worse than no answer. If you appear nowhere across 200 answers, check whether Bing can crawl your site, since ChatGPT leans on sources outside Google's index, and check whether your website says in plain text which towns you cover and what you do.
The emergency rows deserve their own look. In the 5W data, one national chain took close to 80% of emergency plumber prompts. If a national brand or a lead-generation site owns those answers in Northampton too, that is where the hardest fight is, and a smaller firm might reasonably decide not to fight it.
The second run matters more than the first. Make your changes, run the 50 prompts again in a month, and compare the totals. That is how you find out whether AI search visibility work is doing anything, and the answer in your own spreadsheet beats a vendor's dashboard.
If you would rather we ran the 200 prompts and read the results with you, that is the first step of our AI visibility service. Get in touch with your trade and your postcode area.