More and more buyers start with a question to ChatGPT, Gemini or Perplexity instead of a search box. "Who are the best accountants in Leeds?" "Who are the best mortgage brokers in Glasgow?" So we asked them. We took 1,000 independent UK professional firms (accountants, solicitors, financial advisers, mortgage brokers, consultants and coaches) across 15 towns and cities, from Belfast to Norwich, and put every one of them through all three engines.
For each firm we asked two kinds of question. First, the one a buyer asks before they have a name: who are the best firms like this in my town. Second, the one they ask once they have a name: what do you know about this firm, who runs it, where is it based. This is the third and largest round of a test we started in September, after 15 firms and 45 firms.
92% were not recommended by any engine
We asked each engine for the best firms in each firm's own town and field. 949 firms got that question. 871 of them were not named by ChatGPT, Gemini or Perplexity. Not once. Only 13 firms were named by all three.
The engines don't share a shortlist. ChatGPT named 51 of our firms, Perplexity 39 and Gemini 33. Of the 78 firms named at all, 46 were named by just one engine. Being recommended by one tells you little about the other two.
- Named by ChatGPT51 of 949
- Named by Perplexity39 of 949
- Named by Gemini33 of 949
- Named by all three13 of 949
Reviews were the clearest dividing line
We had the Google review count for each of the 900 firms we added in this round. The pattern was hard to miss.
| Google reviews | Firms | Named by at least one engine | Share |
|---|---|---|---|
| None | 332 | 5 | 2% |
| 1 to 9 | 253 | 11 | 4% |
| 10 to 49 | 176 | 13 | 7% |
| 50 to 199 | 111 | 25 | 23% |
| 200 or more | 28 | 10 | 36% |
A firm with 50 or more reviews was around ten times as likely to be recommended as a firm with none. ChatGPT says this out loud: in 31 of its 90 local answers it quoted star ratings or review counts next to the firms it picked.
This is a correlation, not proof. Firms with lots of reviews tend to be busier and better known in other ways too. But reviews are the one signal the engines keep pointing to themselves.
London is the hardest place to be found
None of the 60 London firms we tested was named for its own field in London. Manchester was barely better, with 1 of 60. Asked for London's best solicitors, the engines named Magic Circle firms. Asked for its best financial advisers, they named national wealth managers. With thousands of firms competing in one city, a place in a top five is rare.
Smaller cities were kinder. Norwich came out best (12 of 60 named), then Glasgow (8 of 60) and Southampton (7 of 60). By sector, financial advisers did best (11%), then accountants, solicitors and mortgage brokers (9% each). Coaches (3%) and consultants (1%) almost never made a list.
They know who you are. They just don't recommend you.
Asked by name, all three engines could describe 773 of the 1,000 firms: what they do, who they work with, where they are. Only 20 firms drew a blank from all three. So this isn't mostly a problem of being unknown. The engines have read the website. They just don't have a reason to put the firm on a shortlist.
Half of what AI knows comes from your own website
Across all the answers, 51% of the sources the engines cited were the firm's own website. For 590 of the 1,000 firms, their own site was half or more of everything cited about them. The engines notice. For 596 firms (60%), at least one engine treated the firm's claims as self-reported: results "as stated on their website", rankings it "could not independently verify", a "leading" or "award-winning" line discounted as marketing.
When engines did look elsewhere, they went to the same places again and again. Companies House was cited more than any other outside source, by a long way. Then LinkedIn, the FCA register, the SRA, Trustpilot, the Law Society's Find a Solicitor, Unbiased, VouchedFor and ReviewSolicitors. If your details there are wrong or thin, that is what the engines repeat.
Who runs the firm, and which facts are true
For 289 firms (29%), at least one engine said it couldn't tell who leads the business. Perplexity was the most cautious (182 firms), then Gemini (136) and ChatGPT (93). For a professional services firm, the people are the product. An engine that can't name your founder or managing partner has less to go on when someone asks for an adviser they can trust.
For 320 firms, the three engines gave different answers to the same factual question: different directors, different addresses, different dates. In one case Perplexity described a Liechtenstein fintech with the same name as a Kent finance broker. 60 firms were mixed up with a similarly named company in at least one answer. None of this comes from the firm doing anything wrong. It comes from thin, inconsistent or outdated information elsewhere on the web, and the engines filling the gaps differently.
What to do about it
None of this needs a big budget. Ask ChatGPT, Gemini and Perplexity for the best firms in your field and your town, then ask what they know about you, and write down what is wrong. Get reviews steadily: the jump in our data came at around 50 Google reviews, so ask every happy client, every month. Make sure Companies House, your regulator's register (FCA, SRA, ICAEW), LinkedIn and the main directories in your sector match your website on name, address and who is in charge. Put your people on a clear team page with named leaders and roles. And be specific about where and for whom: a page for the town or client type you serve does more than a general "we help businesses" line.
If you want the same check run on your firm, our free AI Visibility Score asks the three engines about your business and shows you their answers word for word.
How we did it
Firms: 1,000 independent UK professional services firms. The first 100 came from public business directories in September and October 2026. The other 900 came from Google business listings in 15 UK towns and cities, 150 per sector, filtered to independent practices with their own website; chains, national brands and directories were removed. No firm is named here.
Engines and questions: ChatGPT with web search, Gemini and Perplexity, queried from a UK location in English between 24 September and 20 October 2026 through DataForSEO. One "describe this firm" question per firm and engine, and one "best firms in <town>" question per field and town for 949 firms. The first 100 firms also got national "best firms in the UK" questions, which are not counted in the recommendation figures. Every answer was saved word for word.
Scoring: each set of answers was scored against a fixed checklist (does the engine know the firm, does it name a leader, does it discount the firm's claims, do the engines contradict each other, is the firm named in a "best" list) using Claude. We then checked every "recommended" result against the raw answer text and corrected 7 firms the automated scoring had missed. AI answers change from day to day and with the wording of a question, so these numbers are a snapshot, not a ranking.