GEO for Law Firms: How to Get Recommended by ChatGPT, Perplexity and Google AI Mode

“I want to know what happens when a prospective client skips the traditional search and asks ChatGPT who they should hire.” Matthew Khorsandi’s question is the whole of GEO for law firms, and it stopped being hypothetical some time ago. When Gunderson Dettmer rolled out Perplexity Enterprise, more than half the partners logged in on the first day; the firm reports 80% of lawyers active and 35,000+ queries a month, and one attorney says: “I now use it for all my internet searches.” If lawyers search that way, so do their clients.

An answer engine does not return ten links. It returns two names and a reason, and whether one is yours depends on signals most law-firm websites were never built to send.

Why this matters: clients ask ChatGPT first

Clio’s 2025 Legal Trends Report found that 28% of consumers had been told by their AI to contact a lawyer and 12% had been talked out of a claim by it. The LMA/RubyLaw 2025 technology study notes ChatGPT is now the fifth most visited website globally and Google’s search share has slipped below 90%, and Brendan Chard cites Pew data that 57% of US adults use chatbots several times a week.

Divorce.law reports 158,000+ monthly visitors and “107,000+ AI crawler visits daily” for a site whose chatbot routes leads to participating firms; vendor figures, but the direction is plain. Family and estate work, where the client is a consumer with a phone, is exposed first; the family law and estate planning guides cover the practice side.

How answer engines pick a firm

Marcel Zirkel’s German-language analysis puts it in one sentence: AI systems evaluate “nicht primär Keywords und Backlinks, sondern Klarheit, Struktur und nachgewiesene Expertise”, not keywords and backlinks but clarity, structure and demonstrated expertise. The goal of GEO, in his words, is that the AI systems use your content as their source material.

Mechanically, an answer engine retrieves a handful of pages it trusts, reads them and synthesises. It can only name a firm it can retrieve, describe in a sentence and corroborate elsewhere. So a bio that takes four paragraphs to say what you do is invisible; a firm under three names across directories, LinkedIn and the bar register is three weak entities; and the engines have preferred sources. Perplexity’s June 2026 launch post for Computer for Counsel lists Midpage, Deel and LegalZoom among its premium sources, and JD Supra’s 2026 LinkedIn guide observes that “AI tools increasingly pull from LinkedIn to identify experts”. If you are not in the corpus the engine trusts, you are not in the answer.

The Khorsandi test: run it on yourself

Khorsandi’s method takes ten minutes: ask the bottom-of-funnel question a client would ask, then ask the engine to show its sources. “If the same competitors keep showing up, now I have something to investigate.”

I am a [type of client, e.g. the founder of a 40-person software company / a parent going through a divorce] in [city]. I need a [practice-area] lawyer. Who should I consider, and why? Name specific firms or lawyers.
After answering, list every source you relied on, and for each source say what in it made you include that firm.

Run it in ChatGPT, Perplexity, Gemini and Claude, with and without browsing where the tool allows, and log engine, date, firms named and sources cited. Do it again in a week.

Signals: entity, citations, reviews, schema, freshness

Zirkel’s six signals and LaFleur’s July 2026 GEO guide overlap heavily, which is reassuring: two practitioners in two markets reached much the same list.

Jessica Aries reports an Ohio estate-planning firm that improved its AI visibility through location-specific language and Google Business Profile updates, and found the same changes “boosted traditional SEO performance”. GEO and SEO are one discipline with two audiences.

Five audit prompts

Aries’ AI Visibility Audit prompt takes your firm URL, competitor URLs and practice areas and rates your presence across Perplexity, ChatGPT, Gemini and Claude. The Khorsandi test is the first prompt here; four more turn the answers into work.

Here are the visibility-test results from four assistants: <results>. My firm's site is <url>; the competitors that recurred are <urls>. For each recurring competitor, identify the signals they have that we lack: structured data, FAQ pages with first-sentence answers, named-author bios, directory and bar listings, citations in legal publications, content published in the last 90 days. Output a prioritised list of ten fixes, each naming the page on our site it applies to and the competitor page that shows the pattern. Mark anything you cannot verify from the pages themselves as INFERENCE.

Read this lawyer bio: <bio>. (1) If a client asked "who handles [practice] in [city]", which sentence would you quote to justify naming this lawyer? If none, say so. (2) What does the bio say about the type, size, location and industry of clients served, and where? (3) Rewrite the first three sentences so that a reader, or an engine, knows practice, location and client type immediately, using only facts already in the bio.

Here is a FAQ page from our site: <faq>. For each question, check whether the first sentence of the answer answers it. Where it does not, rewrite so that sentence one is the direct answer and the qualifications, exceptions and jurisdiction notes follow. Do not change any legal statement; if an answer is legally incomplete, flag it [LAWYER TO REVIEW] rather than fixing it.

What do you know about [firm name] in [city]? List the firm's practice areas, office locations, named lawyers, notable matters, awards and rankings as you understand them, and for each item say where the information comes from. Mark anything you are unsure of.

The last prompt finds what the engines are saying about you; the bio audit in the advertising-rules guide fixes what you say about yourself.

Content that gets cited: specific, sourced, structured

Passle’s Tom Elgar contrasts two numbers in Attorney at Work: The Economist’s finding that search visits to news and reference sites fell 15% by July 2025, and a 59% rise in direct traffic to legal thought leadership on Passle’s platform over the same period. “AI thrives on yesterday’s internet. Lawyers who are winning the content game write about tomorrow’s legal risk.” His advice: a content culture rather than campaigns, publishing within hours of a regulatory change, and niche sector expertise that is “harder for AI to replicate”.

Jay Harrington’s principle points the same way: “The best content doesn’t make the lawyer look smart. It makes the reader feel smart.” An engine hunting for a source wants what a GC wants: a specific answer to a specific question, dated, with a name on it.

Two cautions: Originality.AI classified 81.2% of 5,000 July 2026 LinkedIn posts as “Likely AI”, and generic text is exactly what an engine has no reason to cite; the LinkedIn guide covers keeping your voice. And Brendan Chard’s 2026 review of Google’s position: “Google does not ban AI generated content”, but E-E-A-T scrutiny keeps rising and about 40% of Google’s updates since 2021 have targeted spam.

Google AI Mode and the UKUT fake-authority warning

The GEO lesson is uncomfortable. The engine that may recommend your firm invented a Court of Appeal bench, and it will attribute to you, with equal confidence, a ranking you never held, an office you closed or a competitor’s “notable matter”. Hence the fourth audit prompt runs monthly, errors are corrected at source (bio, directory, bar register) rather than argued with the chatbot, and nothing an answer engine says about the law goes into a filing. The European hallucination cases guide has the rest of that story.

The Divorce.law funnel, and what it means for you

Divorce.law’s consumer chatbot “Victoria” routes leads to participating family-law firms at $155 a month at its founding rate, according to its own 2026 post. Whatever the true traffic, a third party with better GEO than you is having the first conversation with your client and selling you the introduction. You can rent that position or build your own from the six signals above; either way the chatbot disclosure rules apply to any bot that speaks in your name.

A 30-day GEO plan

• Week one, baseline. Khorsandi test in four engines on three days; the “what do you know about us” prompt; log everything; correct factual errors at source.
• Week two, entity hygiene. One name for the firm and each lawyer everywhere; bios whose first two sentences state practice, location and client type; Google Business Profile complete; schema for lawyers, offices and FAQs.
• Week three, one page that answers a question. Take the question clients ask most, write the page under a named author with the answer in the first sentence, date it, cite your sources.
• Week four, re-test and comply. Repeat the baseline; run the competitor-gap prompt on what changed; put every new page through the pre-publication checklist, because an engine that cites you will cite the award you invented too.

The business-development module of AI Lab for Lawyers covers the LinkedIn and outreach workflows hands-on; run the Khorsandi test on your own firm before the session and you arrive with a fix list rather than a hunch. The five prompts above are in the prompt library.

Where to go next: the business-development pillar for how GEO fits the rest of the BD rhythm, the Perplexity guide for the engine your clients may be using first, and client communication for what happens after the engine sends them to you. The full cluster is in the business-development hub.