Concepts

How AI chooses which law firms to name

When a person asks an AI system which law firm to hire, the system builds the list at the moment of asking. Joe Ashta, a GEO / AEO consultant for law firms, describes the selection mechanism here in plain terms: how the question is taken apart, how candidate pages are retrieved and weighted, how firms are resolved from those pages, why the same question can return a different list on a different day, and what that means for a firm that wants to be named.

The question is taken apart first

A buyer's question arrives in plain language: "who is a good personal injury lawyer in Surrey for a car accident, I need someone this week." Before anything is retrieved, the system works out what the buyer needs. Four parts usually come out of that step: the practice area (personal injury, motor vehicle), the place (Surrey, BC), the urgency (a consultation this week), and any credential or quality the buyer asked for (experience with car accident claims, a free first consultation, a good reputation). Each part becomes a constraint the answer has to satisfy.

This matters for a firm because the answer is built against those parts one at a time. A firm that is a strong match on practice area and a vague match on place loses to a firm that matches both.

Candidate pages are retrieved and weighted

The system then searches for pages that bear on each part of the question and pulls a set of them into its working context. That step is grounding. The pages come from wherever the system's search reaches: the firms' own sites, lawyer directories, bar association records, review sites, news coverage, and pages published by other firms.

The retrieved pages do not carry equal weight. Pages that answer the buyer's question directly, come from sites the system treats as reliable, and agree with other retrieved pages are weighted more heavily than pages that are thin, off-topic or contradicted elsewhere. A firm's own practice-area page is one candidate among many, and it competes with third-party pages that describe the same firm.

Firms are resolved from the pages

The retrieved pages mention many firms and many lawyers. The system has to decide which names refer to the same organization and what it knows about each one. This is entity resolution, and it works from the evidence in front of it: whether the firm's name appears in the same form across the pages, whether it is placed in the same city, whether it is described as doing the same work, and whether the same lawyers are attached to it.

A firm whose evidence lines up resolves into one clear candidate with a set of attributes the system trusts. A firm whose evidence is scattered, with two name variants, an old address on a directory listing and a practice description that changes between pages, resolves into a weaker candidate, or into several fragments none of which is strong on its own.

The answer names the most consistent, most corroborated firms

With the candidates resolved, the system writes the answer. It names the firms whose evidence best satisfies the parts of the question and is most consistently stated and most independently confirmed. A firm that says it handles motor vehicle claims in Surrey, and is described the same way by a directory, a bar record and a news story, is a safer firm to name than one whose only evidence is its own website. The system prefers the safer name. The firm's name in the answer is a mention; when the answer tells the buyer to contact the firm, that is a recommendation. Source links shown under the answer are citations. They point to the pages the system read, which may or may not belong to the firms it named.

Why the list changes from day to day

Answers are generated, and the generation runs again each time the question is asked. Three things vary between runs. The search behind the grounding step can return a different set of pages. The weighting of those pages can shift as the pages change, are updated, or are joined by new ones. And the writing step has some randomness in it by design, so two runs over the same evidence can phrase the answer differently, name firms in a different order, or drop the last name on the list.

The firms that survive that variation are the ones with the most consistent evidence. A firm named because of one strong page is exposed each time that page is left out of the retrieved set. A firm named because a dozen independent sources agree about it keeps its place when any one of them is missing. The BigLaw AEO Leaderboard, which ranks 118 US law firms by AI search citations reported by Ahrefs, shows one consequence of this at the scale of large firms: citation counts vary widely between firms of similar size and standing.

What this means for a firm

The mechanism rewards the clearest, most consistently described, most independently confirmed candidate for a specific question. General fame helps less than firms expect, because the system is answering a specific question and fame is a general property. A well-known firm with a scattered public record can lose a Surrey car accident question to a smaller firm whose record on exactly that question is clean.

For your firm, that sets the work. Decide which specific questions you want to be named for: practice area by place, with the credential the buyer asks about. Give each of those questions a page on your site that answers it directly. Then state the firm's name, place, practice and lawyers in one form across the site, the lawyer profiles, the directories and the bar records, so that the resolution step produces one strong candidate. The step-by-step version of this mechanism is on AI search visibility, and the work of being named is AEO. The canonical statement of who Joe Ashta is and what he measures is the entity record.