How a law firm gets named in ChatGPT
Written by
Odin Lund
Category
Law firms
Date
Share this article
What does ChatGPT read before it names a lawyer?
Being named in ChatGPT means that when a person asks it who to call after a crash, the written answer includes the firm by name, usually as one of two or three firms, with a line on what it does and a link to the page the name came from.
ChatGPT does not pick those firms from memory. It has been able to search the web since OpenAI added search to it in October 2024, and for a local legal question it uses that ability. It runs a search, fetches a small set of pages, reads them, and writes an answer from what those pages say. We see three steps in every answer our tracker records. Retrieval is the engine fetching pages. Selection is the engine deciding which of the firms on those pages to name. Synthesis is the engine writing the description, the fee line, the office cities and the reasons to call. A firm cannot be selected at step two unless it appears on a page fetched at step one, and it cannot be described correctly at step three unless those pages agree with each other.
Which pages ChatGPT fetches is not a mystery. Between 18 September and 17 October 2026 our tracker ran 111 New York car accident questions every day on ChatGPT, Gemini and Google AI Overviews for William Mattar Law Offices, the firm we manage in that category, against 13 competing firms. That produced 9,049 answers, 3,330 of them from ChatGPT, and for each one we recorded every page the engine retrieved and cited. ChatGPT's diet has a clear order. It reads the state first, then the directories, then the firm's own site. The sources below are ranked by the share of ChatGPT answers in which each one was retrieved. One note on the words. Retrieved means the engine fetched the page while writing the answer. Cited means it attached that page to a line of the answer. The two are not the same, and we record both. The shares below are retrieval shares, because retrieval is the step a firm has to clear first.
nycourts.gov, 28% of answers. Court procedure, filing deadlines and the forms behind a lawsuit. This is where ChatGPT learns how a case moves.
ny.gov, 27%. The state's no-fault rules, accident reporting and insurance requirements. The source behind most of the "what do I do now" answers.
nysenate.gov, 15%. The statute text itself, including the no-fault and serious injury provisions people ask about by name.
The firm's own site, 15%. William Mattar's practice pages and process answers. Retrieved more often than any competitor's site, and when it was retrieved ChatGPT cited it four times on average.
Avvo, 15%. Attorney profiles with practice areas, office cities, ratings and reviews. This is where ChatGPT turns a legal question into a list of names.
Justia, 11%. Firm listings by city and practice area, with attorney profiles behind them.
Expertise, 9%. Curated city lists of car accident lawyers, which ChatGPT treats as a ready-made shortlist.
The state bar, lawyers.com and Super Lawyers, 9% each. Three more places where the firm's facts are read and repeated, and three more places where they can disagree with the site.
What this means is simple. ChatGPT learns the law from the state and learns the firm names from the directories and the firm's own pages. A firm that is thin on the directories, and whose site does not answer the process questions, is absent at the exact step where the names are chosen.
Which questions put a firm in the answer?
The 111 questions we run fall into five groups, which we published in full in our companion post. Fifty-one ask the engine to find or recommend a lawyer. Twenty-six ask how a claim works in New York. Ten ask whether hiring a lawyer is worth it and what it costs. Eight ask the engine to compare firms. Sixteen name one firm and ask about it. ChatGPT treats these groups differently, and the difference decides where a firm can be named.
The finding questions are where ChatGPT names firms. Most of them carry a city, and ChatGPT answers them with two or three firms drawn from the directory pages it retrieved, with a line on each. This is the group where a firm either appears or does not, and it is the largest group by a wide margin. The cities in our set are Buffalo, Rochester, Syracuse, Albany, Long Island and New York City, and ChatGPT answers each city question with a different short list, so a firm with five offices is in five separate contests. The comparison questions also return named firms. ChatGPT answers them from directory profiles, review counts and whatever results a firm has published that can be checked, and it tends to be careful, listing firms side by side rather than picking one.
The process questions are answered from the state. For how no-fault works, what counts as a serious injury and how long a person has to file, ChatGPT reads ny.gov, nycourts.gov and nysenate.gov and writes an explanation. A firm is named in these answers only when one of its own pages was retrieved next to the official source, and in our tracking that was often the moment the engine first connected the question to the firm. The whether-to-hire questions sit between the two. ChatGPT explains contingency fees and free consultations, and it often closes with a firm or two whose profiles state those terms plainly. The firm-name questions return a description of that one firm, assembled from its site and its directory profiles, so accuracy on those sources decides whether the description helps or hurts.
In the customer's words, the questions sound like this.
Who should I call after a car accident in New York City?
I got rear-ended on the Thruway near Albany and my back hurts. Do I need a lawyer or will my insurance handle it?
How does no-fault work in New York if the other driver had no insurance?
Is it worth hiring a lawyer for a minor car accident in New York?
Which car accident lawyers in Rochester have the best client reviews?
Does William Mattar take motorcycle cases in Syracuse?
Two of those six are taken from our published list. The rest are the kind of phrasing that arrives in intake calls, and they are what we write into a firm's question set alongside the published ones. In our earlier analysis the same firm was named far more often in the whether-to-hire answers than in the comparison answers, and that gap, not a blended number, told us where to work next. The point of grouping the questions is to find the group you are losing.
What a firm changes to get named
The work for ChatGPT follows its diet. Each item below is a change to a source the engine reads, in the order it has paid off for us.
Make every directory profile state the same facts as the site. Avvo, Justia, Expertise, lawyers.com and Super Lawyers are where ChatGPT finds firm names. Practice areas, office cities, fee model, consultation terms and published case results must read the same on every profile as on the firm's own pages. Where they disagree, ChatGPT hedges or picks another firm. For William Mattar we agreed 16 facts with the firm and made them identical everywhere.
Answer the process questions on your own pages, next to the court and government sources. One page per question, in plain language, with the statute and the numbers. A page that answers how no-fault works as clearly as ny.gov does gets retrieved alongside it, and that is often the first time ChatGPT connects the question to a firm name.
Write for the city questions. Most finding questions carry a city. Each office needs a page that states the city, the cases taken there and the same facts as the rest of the site, and each office needs its own directory presence. A firm that is not named for its own city in these answers is not in the running.
Publish results that can be checked. ChatGPT answers the comparison questions from published settlements and verdicts. Case results the firm has already published, stated the same way on the site and on the profiles, give the engine something it can repeat without hedging. Everything we publish about a firm is a verifiable fact, and every page carries attorney approval before it goes live, which keeps the work inside the advertising rules of the firm's state.
Re-check the answers by engine, every day. ChatGPT re-reads its sources and competitors keep publishing. In our tracking, a blended visibility figure hid a gap of more than 30 points between ChatGPT and AI Overviews for the same firm on the same questions. The per-engine number is what tells you where to work.
Benchmark where you stand before any of this. Take twenty of the questions your clients ask, in their words and with your cities, and run them on ChatGPT from a logged-out browser. For each answer, note which firms are named, in what order, and which pages are cited. Use a logged-out browser because a signed-in account carries history that shapes the answer, and run each question more than once, because the list changes from day to day as the sources change. After a week you will know whether you are missing from the directories, from the process answers, or from both, and you will have a list of the pages ChatGPT trusted instead of yours. If you would rather we did it, request a free preview audit at audit.sourcenode.co and we will send back what ChatGPT says about your firm today and where it comes from.
This is what the work looked like for William Mattar Law Offices. When we started, the firm held 4.6% share of voice and was named in 2.4% of relevant answers, fourth of 14 firms. In month one we rewrote the main practice pages so each answered one customer question with the same set of facts, and share of voice reached 29.5%. In month two we published 31 new pages against the questions where no firm was yet named and worked the directories and publications the engines cited most, and share of voice reached 43.8%, with 54.7% on ChatGPT specifically. Month three was spent holding the position, and share of voice closed at 47.2% with visibility at 23.3%, first of 14. In the most recent 30 days the firm holds 54.6% share of voice on ChatGPT, is named in 22.3% of its answers, and sits at an average position of 2.0 when named. The full method and its limits are on our results page, and our law firm page describes how an engagement runs week by week.
Common questions
Does ChatGPT use our Google rankings? Not directly. ChatGPT runs its own searches and fetches its own pages, and in our data the pages it fetched for New York car accident questions were the state sources and the legal directories, not a copy of the first page of Google. A firm that ranks well on Google and is thin on Avvo and Justia can still be missing from ChatGPT's answers. The test is to ask the question and read the citations.
Why does ChatGPT name a firm we have never heard of? Because a page it trusts listed that firm. A curated city list on Expertise or a well-filled Justia profile is enough to put a small firm into the answer for its city. The engine is not judging the firm's record. It is repeating what the directories agree on, which is why the work starts on those profiles rather than on the answer itself.
How long does it take to show up? First movement usually shows on our tracking within weeks, because ChatGPT re-reads directories and firm pages often. For William Mattar, share of voice went from 4.6% to 29.5% in the first month on the strength of the firm's own pages and the agreed facts, and the firm reached first place in the category by the end of month three. We show the baseline first so the firm can judge the movement itself.
Can you guarantee that ChatGPT recommends us? No, and anyone who does is selling something else. We do not control the engine. What we commit to is the measurement, the work on the sources ChatGPT reads, and a published account of what moved and what did not. One documented engagement is one engagement, and categories differ in how many firms contest the same answers.
Do we need to replace our SEO agency to do this? No. The work sits on top of what you already run. We change the sources the engines read, not your rankings or your ads, and we hand page changes to whoever maintains your site. Most of our law firm clients keep their existing agency. Where we overlap is directory hygiene, and we coordinate that rather than duplicate it.
Sources. SourceNode, Who do AI engines trust when someone asks for a car accident lawyer in New York? 9,049 answers, three source diets, October 18, 2026. SourceNode, 111 questions people ask AI before they call a car accident lawyer in New York, October 18, 2026. SourceNode, Measured results, sourcenode.co/results, data to 17 September 2026. Our tracker, 14 New York motor vehicle accident firms, 30 days to mid-October 2026. OpenAI, Introducing ChatGPT search, October 31, 2024.
Share Blog
