AEO for law firms: get named when clients ask AI.

AEO for law firmsis the work of getting your firm named when a potential client asks ChatGPT, Gemini, or Perplexity which lawyer to call — and of knowing, day by day, whether the engines answer with your name or a competitor's. Those questions are already being asked in every practice area, and each one ends in a short list of firms. This page covers what the questions sound like, the precise mechanism by which firms lose them, how the work stays inside bar advertising rules, and what actually gets built.

The moment a client asks AI

A person with a legal problem no longer starts by typing two keywords into a search box. They describe their situation to an assistant and ask who to call — and the reply is not ten blue links but two or three firm names with reasons attached. Each practice area has its own version of that moment.

Personal injury

Contingency work is the highest-stakes version: the caller rarely shops past the first credible recommendation, so the firms inside the answer split the docket. The phrasing is almost never two keywords — it is the whole story, with the collision, the city, and the insurance adjuster all in one sentence.

Family law

These get asked in raw moments, often late at night, and a chat with an assistant feels more private than a search history. The firm named in that quiet conversation gets the consultation call in the morning — and the asker usually attaches a qualifier the engine then has to match, like a flat fee or a specific kind of custody posture.

Criminal defense

Defense is the most urgent ask in law: there is no two-week evaluation period, only tonight's answer. Whichever firm the engine names first is the firm whose phone rings, and the caller is frequently a parent or a spouse rather than the client.

Immigration

Immigration clients routinely ask AI in their first language, and the engines answer in kind — with the same short list of firms. A firm invisible in English is invisible in every language the engines speak. Deadlines drive the phrasing here: a visa expiry date usually appears inside the question itself.

Estate planning and probate

Estate clients research discreetly — asking an assistant instead of asking around is the whole point — and they tend to take the first well-reasoned name rather than broadcast the question any further. Probate arrives differently: a death, a dispute, and a fast search.

The questions, as clients actually type them

Practice-area labels are how a firm organizes itself. They are not how a client asks. Below are fifteen question shapes drawn from published legal answer research, written the way people write them — with facts, dates, and worries attached, because that is what the engine gets to work with. Read them as a partner would read a lead sheet: every line is a matter someone is trying to hand to a firm.

Two things are worth noticing before moving on. First, almost every question carries a qualifier the engine must match from evidence it can read — a fee structure, a court, a language, a city, an opposing posture. Second, none of them looks like a keyword. A firm that publishes one page per practice area is answering a question nobody asked, while the answer goes to whoever published something that fits the sentence.

The behavior shift, measured

iLawyerMarketing — a legal marketing firm, so read it as industry research rather than an academic study — surveyed 1,100 consumers in 2026 on how they research and choose attorneys:

The same study found Google's use for attorney research down 15 points, and the share of people who double-check an AI recommendation on Google afterward fell from 94% to 70% — for a growing slice of clients, the AI answer is the entire research journey. Clio, the legal software vendor, reports 28% of consumers now research lawyers with ChatGPT specifically. And this is not the low end of the market: iLawyerMarketing found the heaviest AI users for legal research are 45 to 60 years old with household incomes above $125,000 — the demographic behind the highest-value matters.

None of that would sting if Google strength carried over. It doesn't:

How a firm actually loses the answer

“The engines don't know us” is too vague to fix. In legal answers the loss has a specific shape, and it repeats:

• The answer is assembled from rosters, not from firm websites.In Gorilla Web Tactics' published testing, ChatGPT visibly cited Super Lawyers, Three Best Rated, and Yelp while recommending firms. Those are third-party lists. A firm that is missing from them, or listed on them under a name variant, never enters the pool the answer is drawn from — no matter how good its own site is.
• National advertisers arrive pre-quoted. The contingency brands that buy television have also published hundreds of city-and-injury pages and have been cited across the open web for a decade. When an engine reaches for a name it can support with a citation, that inventory is simply there. A strong local firm with six pages is not competing on merit; it is competing on quotable surface area.
• The firm's best evidence is the evidence it may not broadcast. Verdicts, settlements, and win rates are the proof a partner would reach for — and they are exactly what the advertising rules restrict, and what a matter-management system keeps private anyway. So the engine reads a homepage of adjectives and cites someone whose facts are stated plainly.
• Three name variants split one firm into three strangers.If the bar record says “Whitmore Estate Law, PLLC” (an invented firm, for illustration), Avvo says “Whitmore Law,” and the website footer says something else again, the engines meet three half-strangers where one recognizable firm should stand.
• The loss leaves no record. There is no impression count, no click, no line on a media report. A firm can lose the same question every day for a year and the first symptom is a quiet quarter — which is why measurement, not intuition, is the only way this gets managed.

The mechanism underneath is worth ten minutes of any partner's time: how ChatGPT actually picks businesses explains the two pipelines, and the six reasons a business stays invisible is the diagnostic. The same shift is reaching every trust-bought profession — accountants and CPAs are a chapter behind law, not a different story.

The economics your ad budget already proved

Legal marketing already runs on the most expensive attention in advertising. A firm buying clicks at those prices has settled the question this page asks: being the name in front of a potential client at the deciding moment is worth real money. An AI answer is that same moment — repeated for every asker, with no per-click meter running, for whichever firms the engines have learned to trust.

The spend side has a market too. Marketing retainers commonly sit at $2,000–6,000/month (2026 market research), and a full-service legal marketing agency is the right call more often than a software company likes to admit: if your firm needs intake handled, paid search managed, TV bought, and a partner talked out of a bad billboard, that job takes humans, and a retainer is how you put humans on it without putting them on payroll. The honest case for software is narrower: a firm that wants one specific loop — scan the answers, build the page, verify it, check it again — run every day without a new hire. That loop, described end to end, is what our AEO services page covers; the full price ladder, rung by rung, is in how much AEO costs in 2026.

Where the engines read about lawyers

AI engines don't invent their legal recommendations; they assemble them from a recognizable stack of sources:

• Legal directories: Avvo, Justia, FindLaw, Super Lawyers — the profiles engines treat as third-party evidence of practice areas and standing.
• Review surfaces: Google Business Profile, Yelp, Three Best Rated — where volume, recency, and case-type specifics get read, not just star counts.
• The public record: state bar profiles — the anchor for the detail firms get wrong most often, the exact registered firm name.
• Community threads:Reddit and local forums, where “anyone know a good lawyer for…” threads get quoted into answers — see why AI cites Reddit so much.
• Your own site, if the crawlers are allowed in. OAI-SearchBot is the crawler tied to appearing in ChatGPT Search; GPTBot is primarily training-data collection, and blocking GPTBot does not remove a firm from ChatGPT Search. Anthropic runs the same split with Claude-SearchBot and ClaudeBot. Firms that blanket-blocked every AI user agent in 2024 should revisit that file deliberately.

The stack itself keeps moving — 40–60% of the sources AI engines cite change every month (eMarketer, 2026) — which is why a one-time cleanup was never the job. For the machinery underneath the “near me” questions — Google Business Profile, reviews, NAP, structured data — how AI picks local businesses is the companion page. And a note on markup while we are here: Google states no special structured data is required to appear in AI Overviews or AI Mode. Schema earns its place on a firm's site by stating the facts — registered name, admissions, practice areas, consultation policy — in a form that leaves a machine nothing to guess at, and by driving the classic rich results it always did. It is not a password.

The bar-advertising question, answered up front

Every legal marketing conversation ends at the same desk: the ethics rules. Most states' advertising rules descend from ABA Model Rule 7.1 — no false or misleading communication about a lawyer or the lawyer's services — with state-level additions on testimonials, specialization claims, and promised results. The useful way to evaluate any vendor here is not what it promises to do. It is what it refuses to do. Ours, stated plainly:

• We will not write an outcome claim.No success rates, no “we win,” no settlement figures presented as what a reader can expect. Pages state what the firm handles and how it works. We hold ourselves to the same rule: the citation timelines further down are observations from testing, not promises.
• We will not invent a review, a client, or a testimonial.Not as a placeholder, not as an example, not in schema. Every sentence is built from the firm's own information — practice areas, locations, process, fee policy.
• We will not call an attorney a specialist or an expertwhere the state requires certification for that word. Drafts describe what the firm does; the label stays the firm's call, and its ethics counsel's.
• We will not write a superlative about your firm. No “best,” no “top-rated,” no “#1” — including in the title tags and structured data where superlatives usually sneak back in.
• We will not put anything in schema that the visible page does not say. Markup that contradicts the page is a classic misleading-communication trap; matching the two is a hard rule here, not a style preference.
• We will not publish anything a lawyer has not approved. Drafts arrive as drafts. Whatever review chain your firm already runs — managing partner, ethics counsel, outside compliance — sits exactly where it always has: before publication.

None of this is legal advice, and state rules differ — your ethics counsel gets the last word. The practical point is simpler: each draft arrives with the client question it targets stated at the top, which is precisely the shape a compliance review wants to receive.

What the department builds for a firm

Disclosure, so you can weigh what follows: this page is published by AEO Action, and this is the section where we describe our own product. Everything above is true whoever does the work.

AEO Action runs the loop as software — an autonomous AEO department with an approval gate. For a law firm, the output is not “content.” It maps to the docket above, question by question:

• A practice-area-by-city page for the matters you want. One page, one question, one venue: the motorcycle case in your county, the flat-fee uncontested divorce, the juvenile court you actually appear in — stating what the firm handles, who handles it, and what the first call involves.
• A consultation-and-fee page, because the fee structure is inside the question.“Free consultation,” “contingency, nothing upfront,” “flat fee” — engines can only apply that filter from text they can read. Most firm sites bury it or leave it out.
• A language-served page where the practice warrants it. Immigration and defense questions arrive in more than one language; a firm that states which languages its intake staff actually speak has answered part of the question before it is asked.
• Structured data carrying the registered name, admissions, and practice areas. Not because markup unlocks an AI answer — Google says it does not — but because the bar-registered name, the jurisdictions, and the services should exist in one unambiguous machine-readable statement that matches the page.
• A Reddit research pack for the local threads.The “anyone know a good lawyer for…” conversations in your city, prepared as research a human decides what to do with — never as posts we write for you.

The loop that produces them: daily scans put your buyer questions — regenerated each scan from the firm's profile, practice areas, and cities — to up to six engines: ChatGPT, Gemini, Perplexity, Google AI Overviews, Microsoft Copilot, and Claude. (Claude runs in a labeled model mode — how the model understands your firm long-term — and is never counted as a live answer.) From each scan the orchestrator prepares one complete artifact a day and queues it. Nothing publishes without Approve; on day one of a trial, up to four artifacts are waiting for that review, depending on what the scan finds.

Then the receipt, which is the part a partner can audit. Every approved landing page is registered against the client question it targets and re-checked on each following scan for up to 120 days: shipped on date A, first cited on date B. In published testing, the median wait is about 6.8 days to a first ChatGPT citation once a page is indexed, with roughly 42% of pages cited within 30 days (Semrush / practitioner testing, 2026) — observed medians, not a promise about your firm. Every seven days, a plain-English report a managing partner can actually read: what was asked, what changed, what shipped, and what got cited, dated like everything else a firm files.

The mechanics underneath are stated plainly elsewhere and hold here too: Webflow and Squarespace publish directly, the WordPress plugin publishes approved pages on its hourly sync rather than letting us push into your site, everywhere else the work arrives as paste-ready HTML verified once live — and new pages are submitted through IndexNow within minutes, with Bing deciding what to index. Worth being precise about why we bother: independent testing found roughly 87% of ChatGPT's citations also rank in Bing's top results (Seer Interactive, 2026). That is an overlap measured from the outside. OpenAI documents OAI-SearchBot as the crawler for ChatGPT Search and does not publish which index or indexes sit behind it, so we work the overlap because it is large, not because anyone has documented a pipeline. The Bing and Copilot guide takes that apart in full.

The price sits deliberately against the retainer math above: the Solo plan is $149 a month next to $2,000–6,000/month retainers (2026 market research). If you are also weighing enterprise monitoring platforms, the side-by-side with Profound is the honest version of that comparison. And the first move costs nothing: run the free audit and read the answers yourself — ten questions potential clients ask about firms like yours, put live to ChatGPT, Gemini, and Perplexity, 30 answers scored, the three most revealing quoted word for word with competitor names highlighted. No account, no email, no card; the PDF is yours permanently, and the share link lives seven days.