We asked the AI to find a lawyer. 92 times.
Before your next client calls a professional, they ask an AI. We ran the questions they ask, across three regions and five AI tools, and logged every firm every answer named. Here's what the machines are actually recommending.
On 26 August 2026 we ran 92 real client questions through the AI tools people now ask first. Google's AI Mode, ChatGPT, Perplexity, Claude and Gemini. "Best family lawyer in the Sutherland Shire." "My dad died without a will in Newcastle, what lawyer do I need." The questions people actually type.
We covered thirteen areas across Greater Sydney, the Central Coast and the Hunter, four practice areas each. Every answer was logged: which tool, which question, which firms, in order.
We expected gaps. We didn't expect what we found.
The machines have already picked winners
88 answers came back. 239 different firms were named. And the biggest surprise: Google's AI Mode named specific firms in all 52 of its answers. Not "here's how to choose a lawyer". Actual shortlists, with star ratings, years in business, and individual practitioners recommended by name.
No firm dominates state-wide. The game is patch by patch, and patch monopolies are already forming. In one Hunter town, a single firm appeared in all four practice areas. In Sydney's Inner West, one firm took a slot in every category we tested.
When the same client question was put to Claude with no web access, it named zero firms in 14 of 20 runs. The models' memories hold maybe a dozen suburban firms. To the tool most buyers actually use, everyone else gets ranked live, from signals the machines can read today.
The ghosts
The real story is who's missing. One Hunter Valley firm has practised continuously since 1896. Around 130 years of service, five-star client feedback, one of the oldest practices in its region. Named in zero of 92 answers, while a competitor a few hundred metres away appeared in every practice area.
A beachside firm with fifty years of history is recommended by name in local community groups. The AI named it once, and never for the work the locals actually send people there for. A Manly practice holding specialist accreditation in two fields, the exact credential the AI treats as proof, appeared nowhere, because that credential isn't written anywhere the machines read.
These aren't struggling firms. They're good, established practices whose reputations live in referrals, handshakes and community threads. Word of mouth built them. The machines can't hear it.
What actually gets a firm named
The pattern was consistent across every tool. Review volume wins slots, and reviews that praise individual practitioners get those people recommended personally. The "years in business" field on a Google profile gets quoted verbatim. Accredited Specialist status is treated as proof: in one thin market, a firm ranked first for wills with a single review, off accreditation alone. Peer directories carry weight. And the AI reads community threads too, quoting recommendations word for word.
None of that is clever. That's the point. Every signal on that list is boring, buildable and measurable. The firms winning aren't running sophisticated campaigns. They're simply legible to machines.
How we work it
This study is now how we start every visibility engagement. Two fires, applied.
Positioning, credentials, story and reviews rebuilt so what makes a firm good is written where the machines read. Thirty years of brand craft pointed at a new referee.
The same logged questions, re-run at day 0 and day 90. Clients watch themselves appear in the answers, or see exactly why not. Measured, not promised.
We don't control the AI, so we never guarantee a ranking. We build every signal it reads and measure in the open. AI answers change week to week. Everything above is a dated snapshot, and every number in it is logged.
The question for your business
This study happened to be about law firms. The mechanics apply to every local service business: the accountants, the physios, the builders. Somewhere on your patch, an AI is already answering the question your next customer asks before they call anyone.
The question isn't whether it recommends someone. It's whether that someone is you. We'll run your patch and show you, on one page, no charge. If the answer stings, it's fixable. The fixes are boring. That's the good news.