Key findings
ChatGPT read Avvo in 18.8% of answers and cited it in 4.9%.
Gemini read Super Lawyers in 9.3% of answers and cited it in 8.2%.
A directory can shape an answer without being credited in it.
Being read is not the same as being cited
A lawyer directory can shape an AI answer without ever appearing in it. In our New York tracker, ChatGPT read Justia in 14% of its answers about car accident lawyers and cited it in fewer than 1%. The engine used the page to check a firm’s name, city and practice area, then credited a different source.
We measured this for every major directory. For 30 days to 7 November 2026 we ran 111 questions that people ask after a crash in New York, every day, on ChatGPT, Gemini and Google AI Overviews, across 14 competing firms. For each answer we recorded two things per source. Whether the engine read the page while it wrote, and whether it showed that page to the reader as a citation.
The gap between the two is the finding. A firm that judges its directory profiles by where it sees citations will stop maintaining the profiles the engines lean on most.
ChatGPT reads seven directories and credits two
ChatGPT is the engine that reads directories. These are the shares of its answers that read each one, and the shares that cited it.
Avvo. Read in 18.8% of answers, cited in 4.9%.
Justia. Read in 13.9%, cited in 0.6%.
Expertise. Read in 12.3%, cited in 3.4%.
Martindale. Read in 11.9%, cited in 0.4%.
lawyers.com. Read in 10.9%, cited in 0.9%.
FindLaw. Read in 7.5%, cited in 0.5%.
Super Lawyers. Read in 6.5%, cited in 0.8%.
Avvo and Expertise are the two a reader sees. The other five work in the background. ChatGPT opens them, compares what they say about a firm with what the firm’s own site says, and writes the answer. When the profiles disagree with each other, the answer hedges or names another firm. The state bar sits in the same group. It was read in 18.1% of answers and cited in 3.6%.
The practical point is that Martindale, lawyers.com and FindLaw matter for a firm even though they will almost never show up as a source. They are read roughly one answer in ten.
Gemini and AI Overviews behave differently
Gemini reads one directory and cites it heavily. Super Lawyers was read in 9.3% of Gemini answers and cited in 8.2%, about twice per answer when it appeared. Every other directory was below 2.5%. Gemini builds its shortlists from law firm websites, and Super Lawyers is the one outside opinion it checks.
Google AI Overviews cite everything they read, and read little. Justia was read and cited in 4.8% of answers, Super Lawyers in 3.9%, Avvo in 3.7%. FindLaw and Yelp were near 1.3%. On this engine the directories are minor sources. YouTube was cited in 25.6% of the same answers.
So the same directory plays three roles. Avvo is a visible source on ChatGPT, a small one on AI Overviews and close to absent on Gemini. Super Lawyers is the reverse. It is a background check on ChatGPT and the leading directory on Gemini.
What a firm should do with this, and sources
Keep all seven profiles current, not only the ones you see cited. Practice areas, office cities, fee model, consultation terms and attorney names should read the same on every profile as on the firm’s own pages. The profiles that are read silently are the ones most often left stale.
Put the effort where the citation is. For ChatGPT that is Avvo and Expertise. For Gemini it is Super Lawyers. A complete profile with reviews on those three does more than a thin presence on ten.
Measure both numbers, by engine. A citation report alone would have told this firm that Justia and Martindale do not matter. The retrieval record shows they are read more often than Super Lawyers.
How we run this for a firm is on our law firms page, the documented result is on the results page, and every figure we have published is listed with its sample on the numbers page. The definitions of retrieval and citation are in the glossary.
Sources
SourceNode tracking: 111 questions, 14 New York motor vehicle accident firms, ChatGPT, Gemini and Google AI Overviews, daily, 9 October to 7 November 2026. About 3,330 answers each from ChatGPT and Gemini and 2,390 from Google AI Overviews.
Related: Who do AI engines trust when someone asks for a car accident lawyer in New York? and How a law firm gets named in ChatGPT.
The researcher
Odin Lund is co-founder and COO of SourceNode and leads its data analytics and research, including the daily tracking behind this work across ChatGPT, Gemini and Google AI Overviews. Odin founded First Query, the AI search practice that became SourceNode, and co-founded The Pixel Paper.
Meet the research team
About this research
Published by SourceNode
SourceNode is a data and research-first AI search firm. We record what AI models say about a business every day, trace each claim to its source, and use that data to get businesses named in AI answers. Every figure we publish carries its method and its sample.
Share Blog
Related research
AI search
AI search statistics for 2026: the numbers we trust, with sources
Read the research
AI search
What McKinsey’s AI search research says, and what to do about it
Read the research
