Do AI Assistants Recommend the Businesses Google Ranks? A Study of 500 UK Local Businesses
Agencies now get asked a new question by clients: "do we show up when someone asks ChatGPT?" Most of us answer from instinct. We wanted a number, so we measured it.
This is a small study, one snapshot in September 2026, with the method and its limits set out below. It is not a ranking of AI engines. It is a picture of how often the businesses Google already ranks are named when an assistant answers the same question, and what the named ones have in common.
What we did
We chose 5 trades that people search for locally and buy after a short comparison: emergency plumber, dentist, accountant, solicitor and estate agent. We paired each with 10 UK cities: London, Birmingham, Manchester, Leeds, Glasgow, Edinburgh, Bristol, Liverpool, Cardiff and Newcastle. That gave 50 searches.
For each search we took Google's top 10 Maps results from the city centre as the ground truth: 500 businesses in total, the ones Google itself considers the best local answers.
Then we asked each assistant the natural question a person would ask, "What is the best emergency plumber in Leeds?", twice, on 6 September 2026. A business counted as mentioned if the assistant cited its website or named it in the answer, in either run.
| Engine | How we asked | Runs per search |
|---|---|---|
| Perplexity | Sonar model, its default web search | 2 |
| ChatGPT | GPT-4o mini with web search on | 2 |
| Claude | Claude Haiku 4.5 with web search on | 2 |
| Google AI Overviews | UK desktop search results, AI Overview captured when Google showed one | 1 |
The headline: 26 percent
Of the 500 businesses in Google's map packs, 130 were named by at least one assistant. 370, or 74 percent, were never named by any of them.
LocalPulse UK AI visibility study, 6 September 2026. 500 businesses, 50 searches.
Perplexity named at least one map-pack business on 39 of the 50 searches. ChatGPT did on 33. Claude did on 11. Google showed an AI Overview on 1 search out of 50 (solicitor in Edinburgh); for the other 49 it showed the map pack and no AI answer, which is itself a finding about UK local search in September 2026.
Two things to say about Claude's number before anyone quotes it. In 73 of its 100 answers Claude opened with a disclaimer, and in 75 it gave no citations at all. Its low score reflects a policy of not recommending specific local businesses without current data, not a lack of knowledge. Perplexity was the most consistent between runs (82 businesses named in both runs, 5 in only one); ChatGPT was the noisiest (18 in both, 29 in only one).
What the named businesses have in common
We compared the 130 named businesses with the 370 that were not.
LocalPulse UK AI visibility study, 6 September 2026. 50 businesses at each rank.
Map position helps, but less than you might expect. Google's number one was named 37.5 percent of the time, which means 62 percent of the businesses Google ranks first were never mentioned by any assistant. The top three together were named 32 percent of the time, ranks four to ten 23.5 percent.
A website matters more. Of the 466 businesses with a website, 27 percent were named. Of the 34 without one, 9 percent. An assistant with web search has nothing to cite if there is no site.
Reviews matter most. The named businesses had a median of 146 Google reviews. The unnamed had a median of 42. Star rating made almost no difference: 4.70 against 4.65.
| Signal | Named by an assistant | Not named |
|---|---|---|
| Has a website | 27% of 466 | 9% of 34 without |
| Median Google reviews | 146 | 42 |
| Average star rating | 4.70 | 4.65 |
| Google Maps top 3 | 32% named | 23.5% named for ranks 4 to 10 |
By trade
100 businesses per trade, 10 cities. LocalPulse UK AI visibility study, 6 September 2026.
Plumbers and dentists did best. Both trades are heavily reviewed and heavily listed on directories the assistants read. Emergency plumber was also the only trade where Claude named businesses in any number (14 percent, against 0 to 3 percent for the other four), presumably because "emergency" reads as a request for practical help rather than a recommendation.
Estate agents, solicitors and accountants did worst, at around one in five. For those trades the assistants leaned on national directories and comparison sites rather than on the local firms themselves.
By city
50 businesses per city, 5 trades. LocalPulse UK AI visibility study, 6 September 2026.
London was the least visible city by a distance: 3 of its 50 map-pack businesses were named. Birmingham was next lowest. The likely reason is competition. In a city with thousands of dentists, an assistant answering "best dentist in London" reaches for a directory or a "top 10" article rather than a specific practice near the centre.
On 8 of the 50 searches no map-pack business was named by any assistant: dentist, solicitor and estate agent in London, solicitor in Manchester, emergency plumber in Birmingham, solicitor and accountant in Liverpool, estate agent in Cardiff. At the other end, 8 of the 10 estate agents in Newcastle's map pack were named.
What the assistants cite instead
This is the part agencies can act on.
Perplexity leaned on Reddit, cited in 27 of the 50 searches, and on directories: Bark, Clutch, Checkatrade, ReviewSolicitors, WhatClinic, MyNextDentist, TheSolicitorDirectory, Rightmove, GetAgent, VouchedFor, Unbiased, Doctify and Zoopla all appeared. A business that is well reviewed on the directory an assistant trusts for its trade gets named. One that is not, does not, whatever Google thinks of it.
ChatGPT cited Google Maps links and business websites directly, with few directories. So for ChatGPT the business's own site, and its Google profile, carry the weight.
Claude mostly declined to recommend at all. When it did, it behaved like ChatGPT.
What this means for an agency
Four things, in order of effort.
Make sure every client has a real website with the service and the town in plain words. The no-website group was named 9 percent of the time.
Keep reviews flowing. The gap between 146 and 42 median reviews is the clearest signal in the data, and it is the one clients can influence every week.
Get listed, with reviews, on the directory the assistants use for that trade. For a solicitor that is ReviewSolicitors and TheSolicitorDirectory; for a dentist WhatClinic, MyNextDentist and Doctify; for trades Checkatrade and Bark; for accountants VouchedFor and Unbiased; for estate agents Rightmove, Zoopla and GetAgent.
Measure it monthly. AI answers change between runs, as the ChatGPT numbers show, so a single check tells you little. A monthly check across the four engines, in the client's report, is what turns "do we show up in ChatGPT?" from a shrug into a line on a chart.
Limits of this study
One month, one phrasing of the question, city-centre coordinates, two runs per engine, and fuzzy name matching (29 of the 173 engine-level mentions came from the business name appearing in the answer text rather than a cited domain). The Google AI Overview figure comes from one desktop results page per search. Treat the numbers as a snapshot of September 2026, not as a league table of engines, and expect them to move.
Full disclosure: LocalPulse includes an AI visibility check across these four engines in every client report, and we ran this study with the same checks. The table of all 500 businesses, with each engine's result, is available on request to marian@digitalizer.uk for anyone who wants to check our working or run their own analysis.
What to do next
Pick three clients and ask ChatGPT and Perplexity "what is the best [trade] in [town]?" today. Note who gets named and which sites are cited. Then look at those sites: that is the list of places your client needs reviews.
If you would rather have that done every month for every client, with the map grid and organic ranks in the same report, LocalPulse does it at £15 per property per month. You can see the AI visibility section in the sample report.