How can a law firm get recommended by AI assistants?

Google classifies legal questions as YMYL, or Your Money or Your Life, and assistants handle them with particular caution. For information about the law, they tend to cite courts, bar associations and established legal publishers. A firm’s blog post about a limitation period is rarely the main source. Firms have more opportunity on questions about hiring a lawyer: which practice, in which city, for which matter and at what cost.

The people asking are clients at the moment of choice, not law students. They name a city, a practice area, often a language and a budget, and they ask the assistant for a shortlist. Google’s guide to its generative search features describes how such a question is split into sub-queries — the practice, the location, the reviews — and answered from several sources at once, which is why a directory profile, a Google Business Profile and the firm’s own page are retrieved together for one prompt. We do not publish a market-wide figure for how many clients start with an assistant; the number that matters to a firm is its own — how often the assistant names it on the prompts its clients ask — and the measurement section below shows how we get it.

Not all of them, and the honest split is the first thing an agency should show you. A prompt about the law itself — “what is the statute of limitations for medical malpractice in Texas” — is answered from statutes, court sites and publishers like Nolo or FindLaw; a firm can appear as a secondary source at best. A prompt about hiring — “best immigration lawyer in Houston”, “recommend a personal injury firm in Miami that speaks Spanish”, “how much does a divorce lawyer cost in Ohio” — is answered from directory profiles and firm websites, and that is where the competition is open. On 22 September 2026 we put twelve hiring prompts (personal injury, immigration, divorce and startup law in Houston, Miami and London) to ChatGPT, Perplexity and Google AI Mode, twice each, and published the study of all 72 answers. ChatGPT named firms in every answer, 5.6 on average, and in 11 of its 12 first-run answers those names came with a map and a row of business cards. The wording of the prompt mattered too. Asked how to choose an immigration lawyer, Perplexity named no firm in any of its six answers and AI Mode in four of six; both explained how to vet a lawyer instead. Only two prompts produced a firm that appeared in all six answers, so read any single run as a snapshot, not a ranking.

The unit of competition is therefore practice area × jurisdiction × language, not the firm as a whole. A firm with one page listing twelve practice areas gives the model nothing to match against “Houston immigration lawyer”; a firm with a page for immigration law in Houston, with named attorneys, fees and a Spanish version, gives it a candidate. For a US firm the Spanish-language layer deserves a separate check: a client who asks in Spanish gets an answer in Spanish, assembled from the pages and reviews that exist in Spanish — usually a much shorter list.

For a law firm, useful citation material consists of facts an assistant can check against another source and quote directly. State those facts clearly near the top of the page.

Firms often overlook the sources that can confirm their own claims. Assistants consult Avvo, Justia, FindLaw, Martindale-Hubbell, Super Lawyers and state bar registers alongside the firm’s website. Practice areas, office addresses and attorney names need to agree across these sources so the firm can be identified reliably. Google’s guidance on its generative features names Google Business Profile as a source of local business details in AI responses. Our lawyer study shows how much that listing layer weighs: in the 11 answers where ChatGPT showed a map, its cards carried 43 of the 63 firms it named, while only 8 of its 71 cited sources were firm websites, and 41 of Google AI Mode’s 68 sources were firm listing cards. OpenAI says ChatGPT gets nearby business listings from third-party providers without naming them, so we cannot tell which database feeds its map. The practical answer is the same either way: a complete Google Business Profile and identical details wherever the firm is listed.

What does not earn the citation is volume. Forty articles a month about legal topics compete with courts and publishers on the informational layer and add nothing to the commercial one. Google’s guide calls such material commodity content and says its systems favor first-hand, non-commodity pages instead.

Every US state regulates lawyer advertising, and a page written for an assistant is still advertising. ABA Model Rule 7.1, on which most state rules are based, prohibits false or misleading communications about a lawyer’s services, and its comments extend that to truthful statements that create unjustified expectations — including advertising past results without context, and stating that a fee is contingent without disclosing that the client may owe court costs. This guide is not legal advice; check your state bar’s rules before publishing fee pages, results or testimonials.

The constraint is, for GEO, an advantage. What a bar association forbids — “the best injury lawyer in Miami”, guaranteed outcomes, unqualified results — is also what an assistant discounts, because it cannot verify it. What the rules permit — admissions, practice areas, fee structure with disclosures, verifiable results with context — is exactly what a model can cite. A page that would pass a bar review is usually the page that gets named.

We start with a prompt set, not a keyword list: 30–50 prompts per language built from practice area × city × constraint — “immigration lawyer Houston”, “divorce lawyer cost Texas”, “personal injury Miami Spanish contingency” — phrased the way a client asks. We run them in clean sessions on ChatGPT, Perplexity and Gemini, with repeats, and log every answer with date and screenshot. The screenshot starts at the top of the answer, and the log records whether a map or business cards appeared, because on local prompts that is where most of the names come from.

Two metrics come out of it. Citation rate is the share of prompts where the assistant cites one of your pages. Share of voice is the share of firm mentions that are yours among the firms the assistant names on the same prompts. The baseline is re-run every two to four weeks in a pilot and monthly on a retainer. If neither metric has moved after 90 days we say so first, and the prompt set, logs and reports stay with the firm whatever it decides. The free audit covers ten prompts in one language where the firm is currently invisible — for a US firm the Spanish prompt set is often the more revealing place to start.