How does ChatGPT decide which lawyers to recommend?

The short answer: ChatGPT composes its recommendation from two sources: what it learned in training, and what it reads on the live web when it searches. When it searches, it skims a handful of pages, directories, rankings, and law firm websites, and names the firms that show up consistently for the specific place and problem in the question. There is no form to fill in and no ad product that buys an organic ChatGPT recommendation. The only lever a firm has is what those sources say.

How we know

Ayo asks ChatGPT, Gemini, and Google AI Overviews the questions New Yorkers type when they need a lawyer, every day, many times per question, and records every firm each answer names and every page each answer cites. We ask the consumer products themselves, not the vendor APIs (why that distinction matters). This article reports what that dataset shows. Where we state a number, it is measured, not guessed.

Four things the data shows

1. The engines do not agree with each other

In our July 2026 Brooklyn personal injury tracking, ChatGPT, Gemini, and Google AI Overviews each named a different firm most often. Across the same questions, the three engines surfaced 121 different firms at least once. Optimizing for "AI" as one thing is a category error: each engine has its own sources and its own taste. The leaderboard is public, and we wrote up the full disagreement data story.

2. ChatGPT rewards local specificity

ChatGPT leans hyper-local. Ask about a crash in Williamsburg and it reaches for firms and pages anchored to that place, not just the biggest citywide brand. Gemini leans the other way, toward large established names. A firm with strong neighborhood-level presence can outrank a much bigger competitor on ChatGPT while staying invisible on Gemini, and both readings are stable patterns in our samples, not one-off quirks.

3. Different engines trust different pages

When ChatGPT cites sources for a lawyer recommendation, the citations skew toward law firm websites: practice pages, case results, attorney profiles. Google AI Overviews skews toward directories and review sites instead. That means the fix differs by engine: for ChatGPT, publish substantive pages on your own site that answer the client's actual question; for AI Overviews, be present and complete on the directories it already trusts. (And check that an AI Overview appears for the search at all: some lawyer searches never get one.)

4. One answer is noise; the pattern is signal

Ask ChatGPT the same question five times and you get overlapping but different lists. Change the phrasing and the leader can change too: when we ran the same Brooklyn car accident question in three phrasings, the most-named firm flipped between phrasings on ChatGPT. Any single screenshot of an AI answer proves nothing. Share of many sampled answers is the number that means something, and even that number moves: the engines change their favorites within weeks.

What a firm can actually do

• Answer the real questions on your own site. Pages that directly answer "rear-ended by a truck in Brooklyn, who do I call" style questions are the pages ChatGPT cites.
• Be findable at the neighborhood level. Borough and neighborhood specificity is a ChatGPT advantage smaller firms can win.
• Complete the directory profiles AI Overviews trusts. Claimed, consistent, complete.
• Measure before and after. Recommendation share moves month to month, and it moved for real in our data: one month's batch saw the market leader change on all three engines at once. Without tracking you cannot tell whether anything you did worked.