Follows The UK AI Visibility Report for Solicitors — July 2026, Report TAI-R-2026-001.
In July we measured who AI cites when someone asks for a solicitor. This month we measured something harder: who AI actually names.
Between 18 July and 24 August 2026 we ran the same fixed panel of 68 prompts across 17 UK cities against Perplexity, ten repeats per prompt, and recorded — for every one of 1,214 SRA-regulated firms in those cities — whether that firm was named in the answer.
Each firm was assessed against the four prompts for its own city, ten repeats each: 40 eligible answers per firm, 48,560 firm-answer observations in total.
The result is stark enough that it changes how a firm should think about the problem.
Finding 1: 83% of firms were never named once
Of 1,214 tracked firms, 1,003 were never named in any of their 40 eligible answers.
Not rarely named. Never.
At the other end, three firms — Taylor Walton, Wake Smith and Ware & Kay — were named in all 40. Rothera Bray reached 39 of 40. Hay & Kilner 38 of 40.
Because each firm was only tested against its own city's prompts, this is a measure of local visibility. A firm at 100% is winning its own local market outright. A firm at zero is invisible in the town it practises in.
| Firms tracked | 1,214 |
| Eligible answers per firm | 40 |
| Firm-answer observations | 48,560 |
| Never named once | 1,003 (82.6%) |
| Named in all 40 answers | 3 |
| Overall mention rate | 2,279 of 48,560 (4.69%) |
This is a threshold, not a ranking. There is no gentle slope from position 40 to position 20. There is a small set of firms AI names, and a large majority it does not name at all.
In this measurement, visibility behaved more like a threshold than a smooth ranking. Whether that holds beyond this panel is untested.
Finding 2: ChatGPT and Perplexity name different firms
In wave 1, ChatGPT named 120 distinct tracked firms; Perplexity named 258. Mention rate was 2.50% and 3.70% respectively.
| ChatGPT | Perplexity | |
| Mention rate | 2.50% | 3.70% |
| Distinct firms named | 120 | 258 |
The two engines also read different webs. ChatGPT's most-cited sources were law firms' own websites and google.com. Perplexity's were directories and review platforms — reviewsolicitors.co.uk, solicitors.com — and Reddit.
This replicates our July finding on a second, independent collection: a firm can be visible on one engine and absent from the other.
Finding 3: which source AI cites does not predict which firm it names
This is the finding we least expected.
reviewsolicitors.co.uk appeared in 89.6% of Perplexity's answers — by far the most-read source in UK solicitor recommendations. So we asked the obvious question: when Perplexity reads a given directory, does it name our tracked firms more often?
It does not.
| Source cited | Firm named (source cited) | Firm named (source not cited) |
| reviewsolicitors.co.uk | 4.78% | 4.00% |
| solicitors.com | 5.45% | 4.09% |
| reallymoving.com | 4.29% | 4.95% |
| samconveyancing.co.uk | 4.24% | 5.01% |
| thesolicitordirectory.co.uk | 4.33% | 5.01% |
| reddit.com | 4.27% | 4.96% |
| comparemymove.com | 4.32% | 4.80% |
| legal500.com | 3.94% | 4.85% |
| yell.com | 4.44% | 4.72% |
| vouchedfor.co.uk | 4.45% | 4.74% |
Every figure sits between 3.9% and 5.5%, against a 4.7% baseline. Half are slightly higher when the source is cited; half are slightly lower. Legal 500 goes the wrong way.
We measured which sources were cited, not whether each firm was listed on them — those are different questions, and only the first is answered here. But the simple version of the story that circulates in AI-visibility marketing — get listed on the big directories and AI will recommend you — finds no support in this data.
Finding 4: things move on their own
We ran a control group of firms with no intervention of any kind between waves. Their mention rate rose from 3.59% to 4.19% over five weeks — up 0.60 percentage points, for no reason we caused.
This matters more than it sounds. Any firm measuring its own AI visibility across two dates will see movement. Most of that movement is not attributable to anything the firm did. Without a control, a 0.6-point rise looks like a result.
It also sets a floor: an intervention has to beat baseline drift before it means anything.
What we cannot tell you
We do not know why some firms cross the threshold.
We examined the websites of three of the most-cited firms. All three share deep, nested service page structures, a questions-and-answers page under every practice area, and CQS and Lexcel accreditation. Two of the three publish question-shaped content — pages titled the way a person would actually ask.
But all three are also established multi-office regional practices. We cannot separate "structures its website this way" from "is a well-known firm", and we did not check how many of the 1,003 invisible firms share the same traits. Three visible firms having something in common is a hypothesis, not a finding.
Anyone telling you they know which website change makes AI recommend you is ahead of the evidence — including us.
Why we are publishing a result that doesn't help us sell
TendorAI sells AI visibility services to UK regulated firms. The commercially convenient finding would have been that a specific, purchasable action moves the number.
We tested one of our own: whether a firm having a profile page on tendorai.com made AI more likely to recommend it. Treatment firms improved 1.39 points between waves; control firms improved 0.60. The difference — 0.79 points — is small, and across 48,240 treatment-arm observations spanning both waves Perplexity never cited a tendorai.com profile page once. Whatever produced that difference, it was not the mechanism we hypothesised.
We are publishing it because a research programme that only reports its successes is marketing.
What a firm can sensibly do with this
- Check whether you are in the 83%. Most firms have never looked. The answer is usually not what they expect.
- Check both engines separately. They name different firms. One measurement is half a picture.
- Distrust single-point measurements. Your number will move on its own. Anything measured once, or compared across two dates without a control, is not evidence.
- Be sceptical of anyone selling a mechanism. Including us. Ask what they measured, how many times, and against what control.
What this is and isn't
- One engine for the headline finding. The 83% figure is Perplexity only. Our ChatGPT arm could not be re-run: OpenAI retired the model our earlier measurement used, mid-study. We report this rather than substituting a different model and pretending the series continued.
- One profession, one jurisdiction, one window. SRA-regulated solicitors in England and Wales, 17 cities, July–August 2026.
- Constructed prompts. Plausible client queries, fixed before testing, containing no firm names. Not observed consumer behaviour.
- Mention detection is automated. Firm names are matched against answer text by a classifier. It is imperfect at the margins, particularly for firms with common surnames in their name.
- Two collection dates. Wave 1 opened 18 July 2026; wave 2 completed 24 August 2026. Two points are not a trend.
Added 2 September 2026. The prompt set is weighted towards property work. Four prompts per city across 17 cities: "Best conveyancing solicitors in [city]", "Recommend a solicitor in [city] for buying a house", "Which solicitors in [city] have good reputations?", and one local specialism prompt. Two of the four templates are conveyancing, so 34 of the 68 prompts are property queries by construction; the published panel tags 35 of 68 (51.5%) as conveyancing, because one city's specialism prompt — Cardiff's — is conveyancing as well. Every firm was assessed against the four prompts for its own city, so a firm doing no property work was measured largely on prompts outside its practice. These figures describe visibility for property-related queries more closely than visibility for solicitors' work in general.
Method
Fixed panel of 68 prompts across 17 UK cities: four prompt types per city (best-in-city, purchase intent, reputation, practice-area specialism). Firms were drawn from the SRA's public register of regulated firms. Prompts were fixed before any collection began and contain no firm names. Ten repeats per prompt per engine per wave.
Wave 1 opened 18 July 2026 and covered Perplexity and ChatGPT. Wave 2 completed 24 August 2026, Perplexity only.
Each of the 1,214 firms was assessed against the four prompts for its own city, ten repeats each — 40 eligible answers per firm. Firms were assigned to treatment and control groups before wave 1. Every answer was recorded in full, with all cited URLs.
Treatment group: 603 firms. Control group: 611 firms. Each assessed against 40 eligible answers per wave.
The measurement instrument — classifier, collection script and prompt configuration — was verified byte-identical between waves by file hash.
Data availability
- The full 68-prompt panel (CSV)
- The city and firm-count panel (CSV)
- The deviations log, including corrections to our own errors
The panel is unchanged between waves, so the first two files are the ones published with the July report.
Raw AI responses and cited URLs from all runs are retained and available for audit. They are not published in full, as they contain a large volume of third-party firm data collected without those firms' involvement.
How to cite this report
TendorAI. (2026). Most UK Solicitors Are Never Recommended by AI: August 2026 Measurement. Report TAI-R-2026-002. https://www.tendorai.com/resources/ai-visibility-report-solicitors-august-2026
Published ungated. Reproduction permitted with attribution and a link. Figures and findings may be quoted in full.
Gated research cannot be cited.