Across 200 Legal AI Answers, Avvo Was Named 60 Times. Gemini Named a Firm in 53% of Them; Claude in 12%.

Avvo was named 60 times across 200 legal AI answers. Martindale-Hubbell 41. Super Lawyers 37.

The question is not whether the assistant praises your firm. It is whether it hands the client a firm at all, or hands them a directory. In February 2026 Constellate put 50 legal-service queries to four models — ChatGPT, Claude, Gemini and Grok — and collected 200 responses. Directory mentions across those answers: Avvo 60, Martindale-Hubbell 41, Super Lawyers 37, Yelp 18, Best Lawyers 17, Justia 17, Nolo 14, Lawyers.com 5, FindLaw 5 (Constellate). Constellate is a legal-marketing vendor, so a study concluding that firms need help getting named is a study that sells help getting named. The ordering is the useful part.

Sixty mentions across 200 responses puts Avvo in roughly one legal answer in three. The study publishes no per-firm count, so we will not claim a firm-versus-Avvo ratio nobody measured. What it does establish is where an assistant reaches when a firm's own site gives it nothing to stand behind: a directory profile, written by someone else and updated by nobody. Avvo is not outranking your site on merit. It is answering, in fielded text, a question your site left unanswered.

So a client asking who to call after a crash can end the conversation holding a link to Avvo, not a phone number for your firm. Your competitor inside that answer is not only the firm across town. It is the half-filled profile carrying your name, created by someone else and updated by nobody.

Gemini named a firm in 53% of answers and Claude in 12%: the engines disagree with each other more than they disagree with you

The same 200 responses, split by model, show that whether AI recommends lawyers at all depends mostly on which AI you ask. Gemini named specific firms in 53% of legal-query responses, Grok 52%, ChatGPT 26% and Claude 12% (Constellate, February 28 2026; 50 queries across 4 models; legal-marketing vendor).

A partner who checks ChatGPT alone, gets a careful non-answer and concludes AI does not recommend lawyers has learned a fact about ChatGPT — not about Gemini, where firms are named in half the answers.

Assistants do sometimes decline to name a lawyer and answer with criteria instead: check the state bar, find someone who handles this case type, ask how the contingency is structured. Whether that hedge produced Claude's 12% is untested — Constellate measured how often firms were named, not why the low-naming models declined. Either way, a criteria-style answer still quotes somebody's text, and the firm that published those criteria in readable words is the firm it can quote (how to get recommended by ChatGPT).

34% for personal injury, 23% for DUI: the assistant answers by case type, so a firm-name check measures almost nothing

In April 2026 LocalPicks.ai ran 150 legal queries across ChatGPT, Perplexity, Google AI Overview and Claude. AI named at least three specific firms in 34% of personal injury queries and 23% of DUI and criminal-defense queries (LocalPicks.ai, May 16 2026). LocalPicks.ai sells AI-visibility services.

The figure to carry away is not 34. It is the eleven-point gap between two practice areas inside one study. Personal injury and DUI are both urgent, both local, both fed by the same directories — and the assistant still behaves differently. A firm's AI visibility is not a property of the firm. It is a property of the question, and the question is a case type. Only personal injury and DUI have published numbers; a vendor quoting you a divorce or immigration figure is quoting a number that does not exist.

• Personal injury — measured: at least three firms named in 34% of queries (LocalPicks.ai, 150 queries, vendor). The densest directory competition and the most crowded answer.
• DUI and criminal defense — measured: three or more firms in 23% of queries, eleven points below personal injury at identical urgency. That gap is the clearest evidence case type moves the answer.
• Divorce and custody — unmeasured. Clients rarely type "best divorce lawyer"; they type what they should look for, and the assistant answers with criteria quoted from someone's page.
• Employment and wrongful termination — unmeasured. A client who reads three pages before dialing.
• Immigration — unmeasured, and frequently procedural ("what does a spouse green card need"). The firm that publishes the procedure in readable text is the one the answer can name.
• Probate and contested estates — unmeasured. County-level, slow, rarely urgent: the case type nobody tested.

Your own site is where a fact about your firm can exist at all. Everything below corroborates it — ranked by the only counts anyone has published

One study has counted legal sources by name, and its publisher sells legal marketing. Counts below are Constellate's, from 200 responses to 50 legal-service queries across four models, February 2026 (Constellate), unless labelled otherwise. None of it is a proven ranking factor for law firms. And read the order for what it is: the engine reads text, not the state bar's register and not your matter-management system. A fact stated on your own pages and repeated identically by a directory is the maximum verification available to a model — which is why the firm's own site leads this list and the counted directories follow it.

The nearest primary evidence is our own, and it comes from other verticals. In the three markets we have measured ourselves — dental, med spa and plastic surgery in Dallas, July 2026, across ChatGPT, Gemini and Perplexity — 73% to 79% of every source cited behind those recommendations was a business's own website rather than a directory (the Dallas benchmark). Law firms were not in that sample, so treat it as direction, not as your number. Dallas, July 2026 — one market, one day, and a primary measurement: the raw answers are stored, the method is published, and the run repeats against any city, including yours.

And the two studies that touch legal sources contradict each other. Constellate counted FindLaw 5 times in 200 responses; BrightLocal's July 2025 test of 20 queries across 10 verticals found ChatGPT and Perplexity leaning heavily on Superlawyers.com and FindLaw.com for legal queries (BrightLocal; BrightLocal sells local SEO and reputation tools). Anyone selling you a ranked directory strategy is picking a study. And some of the answer comes from text no firm can edit: ChatGPT cites 15.4 sources per answer, leaning on Reddit and Wikipedia, against Gemini's 3.3 (Semrush, 126 million prompts, January to April 2026; sells AI-visibility tracking).

• 1. Your own firm website — the only source on this list where a fact about your firm can originate rather than be repeated: the case types you take, the courts you are admitted before, how the contingency is structured, whether intake answers at 11pm. Never counted for legal specifically. In BrightLocal's July 2025 test the business's own site was the majority of sources for every engine and vertical, and ChatGPT used it 58% of the time in December 2024 (BrightLocal). Our own Dallas run points the same way in adjacent verticals: 73% to 79% of cited sources were businesses' own sites across dental, med spa and plastic surgery (the Dallas benchmark; our primary measurement — one market, one day, no legal cut). Yext's October 2025 analysis of 6.8 million citations across 1.6 million AI answers put Gemini at 52.15% own-site citations and ChatGPT at 48.73% third-party listings (Yext); Yext sells listings management, so a finding that listings decide everything sells listings management — and its own numbers still land more than half of Gemini's citations on brands' own sites. Do this: one crawlable page per case type, in text. Every source below either repeats what that page says or contradicts it.
• 2. Avvo — 60 mentions, the highest count in the study (Constellate; legal-marketing vendor). Do this: claim it; complete practice areas, bar admissions, jurisdictions, years in practice. An unclaimed profile is still what the assistant hands your prospect.
• 3. Martindale-Hubbell — 41; Lawyers.com — 5 (Constellate, a legal-marketing vendor). Two sites in one network scoring 41 against 5: these counts measure what models cite, not what a network wants credited. Do this: complete the Martindale profile and its peer-review fields.
• 4. Super Lawyers — 37; Best Lawyers — 17 (Constellate, a legal-marketing vendor) — the one tier corroborated outside legal marketing, since BrightLocal found ChatGPT and Perplexity leaning heavily on Superlawyers.com (BrightLocal, 20 queries, July 2025). Do this: make the profile carry your case types in the words a client uses. What you may say about a selection on your own site is a state advertising question — below.
• 5. Justia — 17; Nolo — 14; FindLaw — 5 (Constellate, a legal-marketing vendor), against BrightLocal's finding that FindLaw.com is heavily cited. Do this: complete the free profiles, and log them as a hypothesis, not a fix.
• 6. Google Business Profile, and Yelp at 18 mentions. BrightLocal found Google Business Profile strongly preferred by Google's own models (BrightLocal). Yelp is minor in the legal counts and large elsewhere: 512,680 citations across 28 million AI answers in Q4 2025, more than BBB, Angi, Thumbtack, HomeAdvisor and Nextdoor combined (Foundation Inc. with AirOps, a content agency and an AI-search vendor, no legal breakout). Do this: address matching your bar registration, hours, case types listed as services.

The loudest signal in local search is a substantive review reply, and ABA Formal Opinion 496 is why a law firm's is thinner than everyone else's

Reviews with sentences in them are what an assistant reads and quotes, and in most local trades the owner's written reply is half of that text — the half the owner controls. A law firm can still reply. What it cannot do is reply the way a restaurant or a plumber does: with the specifics of what actually happened.

ABA Formal Opinion 496, issued January 13 2021, states: "A negative online review, alone, does not meet the requirements for permissible disclosure under Model Rule 1.6(b)(5)" (ABA ethics opinions; full text). Rule 1.6(b)(5) is the self-defense exception to client confidentiality, and the Opinion's position is that a bad review does not open it. A lawyer may ask the platform or the client to remove the post, and may respond generally without revealing confidential information — for instance, by stating that professional obligations prevent a fuller response. What a lawyer may not do is reveal information relating to the representation in order to rebut the reviewer.

Look at what that leaves you writing. A contractor answering a one-star review posts dates, what was installed, what was offered: crawlable, specific, quotable text. Your compliant equivalent is "We take every client concern seriously. Our professional obligations prevent us from discussing any matter publicly." Correct, nearly content-free, and the sentence every other firm in your city has posted. The specifics have to come from the one place you are allowed to be specific — your own pages.

Formal Opinion 496 interprets the ABA Model Rules; the rules binding your firm are your state's, and they vary. Confirm what applies to you with your own bar or ethics counsel. Nothing here is ethics advice.

The specifics an engine rewards — results, certifications, numbers — are the specifics Rules 7.1 to 7.4 police

Generative engines reward the kind of writing attorney advertising rules restrict. The peer-reviewed GEO study (Aggarwal et al., ACM SIGKDD 2024, 10,000 queries) found citing sources lifted a page's visibility in generated answers by up to 115% for low-ranked content, statistics by 41% and quotations by 28%, while keyword stuffing produced no gain (arXiv). For a law firm, the most tempting specifics — verdicts, settlements, "specialist" — are the ones a bar polices.

The Model Rules, which most states adapt rather than adopt verbatim: Rule 7.1 — "A lawyer shall not make a false or misleading communication about the lawyer or the lawyer's services" (ABA). Rule 7.2 permits communicating about legal services through any media, subject to 7.1 and 7.3; Rule 7.3 restricts live person-to-person solicitation where a significant motive is pecuniary gain. Rule 7.4 governs fields of practice: a lawyer may say what they practice, while "certified specialist" claims require the proper certification context.

A truthful certification claim is constitutionally protected: in Peel v. Attorney Registration, 496 U.S. 91 (1990), the Supreme Court held that a lawyer's truthful statement that he was a certified civil trial specialist was protected commercial speech (Justia). That matters because a certification is verifiable and repeatable — the exact shape of claim an engine restates. What a state may require alongside such a statement is state law, and differs.

Past results are where the divergence between states is sharpest. Two examples, not a checklist:

• Florida: Rule 4-7.13(b)(2) requires past results to be factually verifiable and not misleading (The Florida Bar handbook). Applies in Florida.
• New York: Rule 7.1(e) requires the disclaimer "Prior results do not guarantee a similar outcome" (22 NYCRR Part 1200). Applies in New York.
• There is no single federal disclaimer rule for testimonials or past results. Requirements vary by state; the two above illustrate that variance, they do not instruct your firm. Confirm what your state requires with your own bar.
• What an engine can actually use is the checkable, non-comparative fact: bar admissions and their year; the courts you are admitted before; the case types you take; your fee structure, including the contingency percentage; languages spoken; whether intake answers at 11pm. None of these requires a superlative, and each is verifiable against a public record — which is what a model can repeat without judging you. Advertising rules still vary by state, and a fee or credential statement is still a communication about your services, so have your own bar's rules read against the page before it goes up.

Seven questions a client types in the first hour after something goes wrong. Ask the engines what they answer

Any firm can measure this free, this week, with no tool and no vendor. Open ChatGPT, Gemini, Perplexity and Claude, use a fresh chat per question, paste the seven below with your city or county in the brackets, and record two things: every firm named, and every directory named. That is 28 answers, and the count that matters is not only how often your firm appears — it is how often a directory appears instead.

Sort the misses by case type: a firm named on personal injury and absent on probate has an addressable gap, while "we scored badly" is a mood. Then add one prompt — "Is [your firm name] in [city] any good, and what do they handle?" — to see what the engines already believe about you.

You are reading what the engines already say, so put no client detail and no matter facts into the prompt — ask them exactly as a stranger in your city would. One pass is one moment, though: ask twice and the firms change, and a notepad cannot track a competitor's share across a quarter. Run it anyway; a rough snapshot is still the first finding your firm will have.

• "I was hit by a drunk driver in [city] — who should I call?"
• "DUI lawyer near me available today, first offense, license suspension."
• "Divorce lawyer for a custody dispute in [county] — what should I look for?"
• "Employment lawyer for wrongful termination near me."
• "Immigration attorney for a spouse green card in [city]."
• "Do I need a lawyer after a minor car accident with back pain?"
• "Best probate attorney for a contested estate in [county]."

What managing partners say when we show them this, including the objections that cut against us

These come up in nearly every conversation with a firm owner, and several are correct.

• "Our clients come from referrals." Largely true, and the assistant is entering the referral rather than replacing it. LLM Listed surveyed 871 US adults who had recently used or expected to use legal services: 74% use AI to research a firm's reputation before hiring (LLM Listed, Q2 2026; covered by the ABA Journal, July 10 2026). LLM Listed sells AI-visibility services, so that is a sales argument as well as a finding. But read it: the referral still happens, and then the referred client checks.
• "I need qualified cases, not leads." Fair, and nothing here promises cases. What is measurable is whether the firm is named at all: 62% of those same 871 respondents use AI during discovery to identify reputable firms in their area (LLM Listed, vendor). No study measures whether AI-sourced inquiries convert better or worse than any other channel, and nobody — us included — can tell you what a matter from this channel is worth.
• "People in a crisis call. They don't sit in ChatGPT." Partly right, and the data cuts both ways: AI named at least three firms in 34% of personal injury queries and 23% of DUI and criminal-defense queries (LocalPicks.ai, 150 queries, vendor). The person still trapped in the car calls. The person choosing which of three numbers to dial at 11pm, three days later, with back pain that has not gone away — that person types.
• "Every number here comes from a company selling AI visibility." Mostly, and it pays to know who profits from which finding: Constellate sells legal marketing; LocalPicks.ai and LLM Listed sell AI-visibility services; no independent ABA, Clio or Thomson Reuters study of firm-versus-directory citation shares exists. What separates a claim from a measurement is whether you can check it. Ours you can: the raw engine answers are stored and the method is published. And the fifteen-minute check above costs nothing and runs on your firm, in your market.
• "Can we just claim the Avvo profile and be done?" Unknown, and we will not pretend otherwise. Sixty mentions across 200 responses tells you Avvo sits in the source pool; it does not tell you a complete profile gets your firm named. Sitting in the pool is not being picked out of it. And a profile can only repeat facts that exist somewhere first — the case types, the admissions, the fee structure — which is why the profile is hygiene and the pages carrying those facts are the lever. Claim it, and log it as a hypothesis.

No independent study of law-firm AI citations exists, which is why the only number worth having is your own

The hole in the middle of this subject should be named. Nobody at the ABA, Clio or Thomson Reuters has published a study of how often AI names a specific firm rather than a directory. The Constellate counts, the LocalPicks percentages and the LLM Listed figures are all we have, and all three publishers sell into the problem they measured. The market-level answer a firm would like to buy does not exist. The only number that survives the scrutiny lawyers apply professionally is one measured on your firm, for your case types.

RecoSignal's free AI Visibility Snapshot puts client-style questions to ChatGPT, Gemini and Perplexity and reports how often your firm is named, which firms are named instead, and which sources those answers lean on. It reads public answers and involves no client information. For what moves the odds, see how to improve your AI visibility; for why a strong Google position does not carry over, AI visibility versus local SEO. It shifts probability, not certainty.

No one can promise an assistant will name your firm, and there is no #1 in ChatGPT to buy. A vendor offering either is promising a certainty that does not exist — the kind of claim you would not put on your own website.