Two family law firms run the same playbook. Both publish jurisdiction-specific pages, both structure answers cleanly, both get their statutes right. One gets named when a searcher asks who to hire, and the other does not.

Lexicon Legal Content asked ChatGPT, Claude and Gemini the same six California divorce questions on the same day in August 2026. The three engines did not disagree at the margins. They behaved like three different products.

How this was tested

What did each AI engine cite for California divorce questions?

Each engine produced a different result on identical questions. The Lexicon Three-Engine Family Law Test recorded the following across six California divorce questions in August 2026.

Engine	Cited law firm sites	Named a firm in the answer
ChatGPT	0 of 6	0 of 6
Claude	4 of 6	0 of 6
Gemini	6 of 6	6 of 6

ChatGPT produced 29 citations across the six questions and sent every one to the California courts self-help site or the State Bar. Claude searched on four questions and law firm content dominated what it found, reaching 6 of 8 sources on the divorce cost question. Gemini grounded all six answers and attributed claims to firms by name in the visible text.

What is the difference between being cited and being recommended by AI?

A citation is a link to a page in the sources panel. A recommendation is a firm named inside the answer as somewhere to look, which can happen with or without a link.

The Lexicon Three-Engine Family Law Test found these are separate outcomes rather than degrees of the same one. Across 12 ChatGPT and Claude answers, zero law firms were named in the answer text. Across 6 Gemini answers, a firm was named every time.

Firms are being sold answer engine optimization packages that promise the second outcome and perform work that produces the first. The work is real. The promise runs ahead of it.

Does Claude search the web for family law questions?

Not always. Claude answered 2 of 6 questions in the Lexicon Three-Engine Family Law Test without searching at all: property division and child custody.

Both answers were accurate, citing California Family Code section 3011 from memory, with no sources panel and nothing linked. No page could have been cited on those questions regardless of how well it was built, because the engine never went looking.

Gemini grounded on those same two questions, so this reflects how one engine handles settled state law rather than a general property of AI search. The practical consequence stands either way: part of what an engine says about a practice area is fixed before a firm’s content enters the picture.

Which law firms did AI engines cite most often?

Four firms appeared on two engines independently in the Lexicon Three-Engine Family Law Test: Griffith Young, Moshtael Family Law, Provinziano & Associates, and Erica Bloom Law.

Two engines with separate retrieval systems, run separately, surfaced the same firms. That is not one product’s ranking quirk. Those firms have published content on the specific questions people ask, at a level of detail that survives being summarized.

Cross-engine repetition is the most useful pattern in the dataset, because it is the part that does not depend on any single engine’s behavior. It is what the signals engines use look like in practice rather than in theory.

Do AI engines recommend law firms to hire?

Rarely, and never straightforwardly. On the hiring question, all three engines in the Lexicon Three-Engine Family Law Test routed to institutions rather than to law firm marketing.

ChatGPT cited the California courts and the State Bar. Claude cited seven sources and all seven were bar referral services. Gemini named a firm, then wrapped it in the State Bar attorney search, county bar referral panels and third-party directories, and told the reader not to choose off sponsored ads.

The most commercially valuable query produced the most consistent refusal. A firm competing for that answer is not competing with other firms, it is competing with the State Bar, and the route in is being the content the institutional answer draws on.

How should a family law firm measure AI visibility?

Run the same questions on more than one engine and record two things per answer instead of one. Which domains were cited, and whether any firm was named in the answer text.

The second column is the one almost nobody records, and it is the only way to tell a citation problem from a recommendation problem. Our own California audit took under two hours.

Three outcomes, three different responses. If firms are cited but yours is not, that is a content problem and content work moves it. If the engine answers without searching, no content changes that query. If firms appear on one engine and nowhere on another, that is the engines being different products rather than a failure, and the fix is measuring more than one.

Most family law firms have never checked which of the three they face. The urgent, high-stakes questions are worth testing first, because that is where a searcher decides whether to call anyone at all.

Frequently Asked Questions

Does ChatGPT recommend law firms?

No, not in the Lexicon Three-Engine Family Law Test (August 2026). All 29 ChatGPT citations across six California divorce questions went to the California courts self-help site or the State Bar, and no law firm was named in any answer. That included the question asking directly how to find a divorce lawyer.

Does Claude cite law firm websites?

Yes, on informational questions. In the Lexicon Three-Engine Family Law Test, Claude searched on 4 of 6 California divorce questions and law firm content made up most of what it found, including 6 of 8 sources on the divorce cost question. On the hiring question, Claude cited seven sources and all seven were bar referral services.

Does Gemini name law firms in its answers?

Yes. In the Lexicon Three-Engine Family Law Test, Gemini named a law firm inside the answer text on 6 of 6 California divorce questions, including Szeto-Wong Law, Provinziano & Associates, Moshtael Family Law and Candelaria LLP. ChatGPT and Claude named zero firms across their twelve combined answers.

What is the difference between being cited and being recommended by AI?

A citation is a link to a page in the sources panel, produced when an engine searches and finds it. A recommendation is a firm named inside the answer as somewhere to look. The Lexicon Three-Engine Family Law Test found 12 ChatGPT and Claude answers naming zero firms while 6 Gemini answers named one every time, which makes these separate outcomes that need separate measurement.

How should a family law firm measure AI visibility?

Pick six questions clients actually ask, name the state in each, and run them on at least two engines. The Lexicon Three-Engine Family Law Test recorded two columns per answer: which domains were cited, and whether any firm was named in the answer text. There is no analytics report for the second one, because a firm named without a link produces no referral traffic, so it has to be observed directly rather than estimated.

Before You Buy an AI Visibility Package

Any vendor quoting a single AI visibility score is averaging across engines that behaved nothing alike on identical questions. That number cannot identify which of the three problems a firm has, and the three need different work.

Every piece Lexicon delivers is reviewed by a licensed attorney before it reaches the client. Call 877-486-8123 with the six questions clients ask most, or start with the legal content writing. You can also contact us with what you are seeing.

---

David Arato, JD is the founder of Lexicon Legal Content, an attorney-owned legal content marketing agency serving law firms since 2012. Lexicon tests AI engines directly and records whether a firm is named in the answer, not just which pages get linked, because those turn out to be different results. David is a frequent contributor to Attorney at Law Magazine and Attorney at Work and a recurring guest on legal marketing podcasts.