93% of UK professional firms are not recommended by ChatGPT, Gemini or Perplexity, study of 1,000 firms finds
Independent accountants, solicitors and advisers with few Google reviews are almost never named when buyers ask AI for “the best firm in town”. London firms fared worst: none of the 68 tested was named.
Kelriva AI, a London AI consultancy, has published a study of how three AI assistants recommend UK professional services firms. The research checked 1,000 independent accountants, solicitors, financial advisers, mortgage brokers, consultants and coaches across 15 UK towns and cities. The full anonymised dataset is openly available on Zenodo, the research archive run by CERN, under DOI 10.5281/zenodo.23950083.
Researchers asked ChatGPT, Google Gemini and Perplexity the question a buyer asks before they have a name, such as “Who are the best mortgage brokers in Leeds?”, then checked whether each firm appeared in the answer for its own town and field.
Key findings
- 928 of the 1,000 firms (93%) were not named by any of the three assistants. Only 11 were named by all three.
- The assistants rarely agreed. ChatGPT named 47 of the firms, Perplexity 34 and Gemini 28.
- Google reviews were the clearest dividing line. Firms with no reviews were named 1% of the time. Firms with 50 to 199 reviews were named 23% of the time, and firms with 200 or more 38%.
- Location mattered. None of the 68 London firms tested was named for its own field in London. Norwich firms did best, with 13 of 67 named.
- The assistants mostly knew who the firms were. Asked about a firm by name, all three described 757 of them. But for 56% of firms at least one assistant could not say who leads the business, and for 33% the three gave conflicting facts such as different directors or addresses.
- Half of the sources the assistants cited about a firm were the firm's own website. Companies House was the most cited outside source, followed by LinkedIn and the regulators' registers.
“The AI knew most of these firms. It just didn’t trust them enough to recommend them. AI doesn’t recommend the firm that says it’s the best. It recommends the firm everyone else says is the best.”
Costin Bucuci, co-founder of Kelriva AI
The study found the gap was in recommendation rather than recognition, which suggests that consistent public facts and a steady flow of genuine reviews matter more than paid listings. Kelriva AI notes that review counts and recommendations are correlated, and the study makes no claim that one causes the other.
Methodology
1,000 independent UK firms were drawn from Google business listings in six sectors and 15 towns and cities, filtered to independent practices with their own website. All questions were sent in English from a UK location between 24 September and 4 November 2026. Answers were scored against a fixed checklist, and every “recommended” result was checked by hand against the original answer. No firm is named in the published data. AI answers change over time, so the results are a snapshot, not a ranking.
Read the full study. The dataset, report and exact questions are at doi.org/10.5281/zenodo.23950083 and are free to reuse with credit (CC BY 4.0).
