AI Search Visibility for Law Firms

What is known, and how little of it is legal-specific

When somebody with a legal problem asks an AI assistant a question — whether a landlord can evict them, how long they have to sue after a collision, whether they need a lawyer at all — the assistant answers in prose and usually cites a handful of sources. Occasionally it names firms. AI search for legal services is the question of which sources those are, whether a firm can influence the outcome, and what happens when an assistant says something about a firm that is not true.

Start with the part that gets skipped. Nothing in the retrieval mechanism is legal-specific. Google publishes no legal-vertical guidance, and a vertical here means an industry segment treated as its own market — legal, healthcare, insurance — that vendors report on separately. Google's general position applies unchanged: there are no additional requirements to appear in AI Overviews or AI Mode, and eligibility requires only that a page be indexed and eligible to be shown in Google Search with a snippet. Google's July 2024 explainer names the categories where AI Overviews are restricted — highly sensitive, explicit or dangerous topics, most election queries, hard news, and data voids — and legal advice is not among them. OpenAI, Perplexity, Microsoft and Anthropic publish nothing about legal queries either.

What is legal-specific is the competitive structure of the source pool, and the state of the evidence about it.

The measurement gap, and who fills it

The legal vertical is one of the worst-evidenced in AI search, and the gap is documented by absence. BrightEdge tracked AI Overview citation behavior across nine verticals from February 2025 to February 2026 — healthcare, business technology, education, insurance, entertainment, travel, ecommerce, finance and restaurants. Legal is not one of the nine, and neither Pew's click study nor Ahrefs' citation studies break it out.

That vacuum is filled by parties selling to law firms. The main legal-vertical study in circulation is the Legal AI Visibility Index 2026, published in April 2026 by the Haute Lawyer Network with 5W Public Relations, a firm that sells AI-visibility audits to law firms. Its sample is 50 to 100 queries across four AI systems, eight practice areas and three large firms. From that sample comes the widely repeated claim that 23.6% of legal queries trigger an AI Overview — published with no keyword set, no sampling frame, no location and no date range, and replicated nowhere independent. A vertical prevalence figure drawn from a hundred queries is an anecdote with a decimal point in it. The report's framing that firms are "six quarters behind" is not a measurement either: no baseline, no metric, no comparison group.

This is why the verdict on this page is Unsupported. It is not a judgment that AI search does not matter to law firms. It is a judgment that almost every legal-specific claim being sold to them rests on one small study by an interested party.

Two query types that behave differently

Conflating two kinds of query produces most of the bad advice in this space, and the two are easy to separate once named.

Informational legal questions — what comparative negligence means, how long a statute of limitations runs — are exactly the query type AI Overviews trigger on. Ahrefs found 97.7% of AI Overview keywords to be informational across 55.8 million overviews in May 2025. This is where legal content marketing has lived for fifteen years, and it is the traffic most exposed to being answered on the results page.

Lawyer-finding queries — a practice area plus a city, or "near me" — are local and transactional, historically served by the local pack and by directories. Whether they trigger AI answers at comparable rates is not established by any independent study located.

The consequence is uncomfortable for the way most legal marketing budgets are drawn. The measured click damage lands on the informational library, the practice-area explainers and the blog. The lawyer-finding query, where the money is, is the one nobody has measured. Firms are routinely sold work aimed at the second while the loss is accruing against the first.

Why directories sit in the answer layer

The one legal-specific finding that is probably true is also the least surprising. Avvo, FindLaw, Justia, Martindale-Hubbell, Super Lawyers, Best Lawyers, Chambers and Legal 500 have dominated organic legal search for two decades, because a site aggregating every firm matches almost any query about firms. The same index reports seven of those directories dominating AI citations in April 2026.

Consider the mechanism — the causal path by which one thing produces another, as distinct from the correlation you can see. Generative systems retrieve from the same index that serves ordinary results. No documented step in that pipeline disadvantages an aggregator, and nothing in it advantages a single-firm site over a page that already ranks for the same query. The finding is credible precisely because it requires no new mechanism: it is what organic legal search already looked like, carried forward.

That cuts two ways, and the second way is usually left out. It is a reason to believe the direction of the finding. It is also a reason to stop reporting it as news, and to be careful with the conclusion drawn from it. A directory profile is a plausible route into an answer. It is also a property whose content the firm controls only partly, whose ranking the firm does not control at all, and whose incentives are not the firm's.

What query fan-out does to a page-one ranking

The mechanical change that matters most to a firm is one Google documents. Query fan-out is Google's term for issuing multiple related searches across subtopics and data sources before an answer is composed. The page cited in an answer about a legal question therefore need not rank for that question at all.

Ahrefs measured the share of AI Overview citations coming from pages ranking in the top ten at 76.10% in July 2025 and 37.9% in March 2026, disclosing that it improved its citation parsing between the two studies — so the exact size of the drop is soft while the direction holds. For a firm, the reading is that ranking on page one for a city-plus-practice-area term predicts an AI citation less well than it did a year ago, and that no surface shows which sub-queries built a given answer.

This is the point at which generative engine optimization — the practice, marketed as GEO, of writing and marking up pages in the hope of being quoted by AI answer systems — is usually sold. It is worth knowing what the offer does not include. No AI operator has a submission mechanism, so a firm cannot be listed with, or registered to, ChatGPT or Google's AI. Structured data has not been shown to produce citations: the one controlled test, on 1,885 pages in May 2026, found no uplift on any surface. And Google has stated in writing that Google Search ignores llms.txt files, so building one for a firm's site is not AI-search work.

What the consumer research actually measured

The most-quoted consumer figure in legal AI marketing comes from Clio's 2025 Legal Trends Report, released 16 October 2025, on a survey of 1,000 US adults weighted to census statistics. Three caveats travel with it: Clio sells practice management software and publishes the report as marketing; the base for the AI sub-questions is an unstated subset of the thousand; and the headline is a compound measure.

That last point is the one that matters. "More than half of consumers have used or would consider using AI to answer a legal question" mixes behavior with stated willingness. It is not a usage rate, and it is quoted as one constantly.

Two figures underneath the headline are more interesting and almost never quoted. Among consumers who did use AI for a legal question, 28% were directed by the AI to contact a lawyer — AI as a referral step. And in 12% of cases, consumers were convinced by an AI that their legal problems weren't worth pursuing. That second number is the most consequential figure in the vertical and the least repeated. It describes an assistant suppressing demand rather than routing it, a mechanism with no real analogue in classic search, where somebody who reads a page still has to decide for themselves. Both figures come from the same survey, they point in opposite directions, and neither sample is large enough to settle which effect dominates.

The advertising rules already apply

Content written to be quoted by an AI assistant is still a communication about a lawyer's services, and the governing rule does not care about the channel. ABA Model Rule 7.1 states that "A lawyer shall not make a false or misleading communication about the lawyer or the lawyer's services," and defines such a communication as one containing "a material misrepresentation of fact or law, or omits a fact necessary to make the statement considered as a whole not materially misleading." The rule text is published by the ABA, and state analogues vary. The obvious exposure is the superlative written to be quotable: a page engineered so an assistant will repeat "the best personal injury firm in Houston" is a Rule 7.1 problem in a GEO costume, and a machine repeating it does not change who published it.

One state has addressed AI in advertising directly. Florida Bar Ethics Opinion 24-1, issued 19 January 2024, holds that a lawyer using a generative AI chatbot for advertising and intake must inform prospective clients they are communicating with an AI program rather than a person, that communications carry a clear and conspicuous disclaimer, that the use must comply with the state's rule on deceptive and inherently misleading advertising, and that the lawyer is "ultimately responsible in the event the chatbot provides misleading information to prospective clients." It adds that a lawyer cannot claim their generative AI is superior to another firm's unless the claim is objectively verifiable. The opinion is published in full. Note its scope: Florida only, and it governs the firm's own chatbot rather than an assistant's description of the firm.

Two corrections to what circulates. ABA Formal Opinion 512, from July 2024, is cited constantly as marketing guidance; it addresses competence, confidentiality, supervision, fees and communication with existing clients, and mentions Rule 7.1 in a footnote. That footnote is the whole of it. And most states have addressed none of this: Justia's fifty-state survey identifies Florida as the only jurisdiction with explicit AI-advertising guidance, while an agency compilation counts five states with some AI-specific guidance and roughly forty-five without. They disagree because they define the category differently; the direction is safe.

The open question nobody has answered

Here is the question a marketing partner will eventually be asked and cannot currently answer: what is a lawyer's obligation when an AI system, unprompted, tells a prospective client something false about the firm — invents a practice area, attributes a result the firm never obtained, or calls it the best in a city?

Three situations are distinguishable and only two have answers. If the firm published the claim, Rule 7.1 plainly applies and the assistant merely repeated it. If the firm paid for it, the rules governing payment for recommendations come into view, and no authority located has addressed whether an AI recommendation is a recommendation within their meaning. If nobody published it and the model generated it, there is no authority at all as of August 2026, in either direction. Do not let anyone tell you it is settled.

What makes the exposure hard is structural. An assistant describing a firm is assembling text from directory profiles, review sites, news coverage, court listings and the firm's own site. The firm wrote one of those inputs, cannot see the output, and has no correction mechanism to appeal to: no operator provides one for a false statement about a business, beyond the general opt-out that removes the site entirely. That leaves contemporaneous documentation as the only record a firm will have — the prompt, the date, the system and version, and the response exactly as it was returned. The discipline of documenting and verifying AI output so that it holds up later as a professional record is treated at length at The Expert Record.

What is worth doing anyway

Strip out the legal-specific claims that do not survive scrutiny and a short list remains. None of it is legal-specific, which is the finding rather than a disappointment.

• Do the ordinary ranking work, because AI features are grounded on the ordinary index and eligibility is ordinary Search eligibility. Whatever gets a page crawled, indexed and ranked determines whether it can be selected at all.
• Audit the directory profiles the firm can edit. If the source pool is genuinely dominated by aggregators, an accurate, complete profile on the ones a firm already pays attention to is the cheapest plausible route into an answer available — and it is plausible rather than proven.
• Write the informational library so each answer is self-contained, since a retrieved passage is read without the rest of the page around it. That is an observed pattern with a sound mechanism and no platform confirmation, and should be labeled that way rather than sold as a citation lever.
• Read every superlative on the site as if a machine will repeat it, because one may.
• Report AI impressions as impressions. Google's generative AI performance report gives impressions only, merges AI Overviews with AI Mode, and carries no query dimension and no clicks. It is not traffic and cannot be converted into traffic.

The honest position is that this is a subject to watch closely and spend on cautiously, and that anyone offering a legal-vertical AI-visibility number should be asked for the keyword set, the sample size and the date before anything else is discussed.

Frequently asked questions