What ChatGPT and Google AI actually say about solicitors — our own sweep data
We put 50 real buyer questions about Southampton solicitors to three engines and tallied every answer. One engine named firms hundreds of times and recommended precisely none of them.
Steen Stones · Reviewed 5 Aug 2026
Ask which solicitor to use for a Southampton house purchase, a divorce, a probate case, and three different engines answer in three genuinely different ways. We put 50 such questions to ChatGPT, Gemini and Claude and tallied every firm that came up — 78 in total. The starkest finding needs no interpretation: across all 78 firms and all 50 questions, Gemini named a solicitor hundreds of times and put one forward as its specific pick precisely zero times.
Avouch multi-engine market sweep, 17 July 2026 — 50 buyer questions, Southampton solicitors, ChatGPT + Gemini + Claude, 78 firms tallied
One engine lists. The other two pick.
Moore Barlow, the most-cited firm in the sweep, was named by Gemini 34 times and recommended zero of those times — every single mention was Gemini presenting it as one name among several, never as the answer. ChatGPT and Claude, asked the identical questions, converted real shares of their own mentions into a specific pick.
Moore Barlow, the most-cited firm in the sweep — same questions, three engines
The same shape holds across the other firms in the tally, not just the leader: Paris Smith converted 63% of its ChatGPT mentions and 65% of its Claude mentions into a recommendation — and 0% of 21 Gemini mentions. If you are measuring “AI visibility” as one blended number across engines, Gemini's flat zero is quietly dragging every average down, and hiding real, usable results on the other two.
What actually earns the specific pick
When ChatGPT and Claude did recommend a firm by name, the reasoning was never generic. Every one of the sweep's recommendation quotes ties the pick to something specific and checkable — a named team, a named specialism, a stated capability — not “this firm is well regarded”.
“My genuine recommendation is Eric Robinson Solicitors — intestacy administration is a named specialism on their site, and their Southampton team works through exactly this process.”
Three more from the same sweep, same pattern: Trethowans won a jointly-owned-property divorce question because “their Southampton family team deals with jointly owned property routinely”. Moore Barlow won a settlement-agreement question because its “Southampton employment team specifically handles settlement agreements”. And on a complicated residential purchase, ChatGPT picked Moore Barlow “despite the likely higher fee” — naming the trade-off explicitly rather than pretending it away. In every case, the deciding detail was a specific practice-area claim the firm's own site could support, not a general reputation statement.
Who the sweep actually surfaced
Legal500.com and chambers.com both appear in the top ten most-cited sources across the sweep, alongside reviewsolicitors.co.uk — the same pattern the pillar page found in a completely different pair of markets. Two independent sweeps, two different sets of firms, the same three kinds of source doing the citing.
Common questions
- Why does Gemini behave so differently from ChatGPT and Claude?
- We can see the pattern clearly in this sweep but not the cause — Gemini appears to default to a listing behaviour for this kind of local-services question rather than a single-pick behaviour, at least at the time we ran this. It is exactly why we report every engine separately rather than blending them: a single average would make Gemini's real, large presence look like a smaller, more moderate number, and hide that it functions completely differently from the other two.
- Is this the same data as the pillar page?
- No — the pillar page's chart covers personal injury and commercial debt recovery, ten questions each. This is a separate sweep: 50 questions about general Southampton solicitors, run 17 July 2026. We're publishing both because the underlying pattern — named is not the same as recommended, and it varies sharply by engine — held independently in two different markets swept two weeks apart.
- Are these real AI answers?
- Yes, verbatim quotes from the sweep, lightly trimmed for length where marked. Every quote is attributed to its engine and the question that produced it.
AI can't recommend what it doesn't know.
Start free. No commitment.