AEO for Law Firms: How AI Assistants Decide Which Attorney to Name
A referral channel that never rings, never emails, and never tells you it passed you over.
The Question Before The Question
Very few people ask an AI assistant to name a lawyer as their first move. They ask something upstream of that. "Do I have a case if I was fired two weeks after reporting harassment." "How long do I have to file after a car accident in North Carolina." "What does an uncontested divorce actually cost."
The assistant answers, and somewhere in that answer or the follow-up it starts naming firms. By the time a name appears, the person has already been given a framework for thinking about their problem, and they are predisposed to trust whoever the assistant treats as authoritative.
This is the same dynamic as ranking for informational content and converting downstream, which firms have understood for fifteen years. What is different is that there is no click in the middle, so your analytics show nothing.
Practice Area Beats Firm Size
The good news for small practices is that assistants are matching on specificity, and firm size buys little. A four-attorney firm that has published clearly on North Carolina workers compensation is more likely to be named for a workers compensation question than a large general-practice firm with a bigger ad budget.
Breadth hurts here. A homepage that lists fourteen practice areas gives a model nothing to latch onto. A page that covers one area in depth, for one jurisdiction, is legible.
If you practice in three areas, you want three substantial bodies of content; one page each will not register. The firms getting named consistently have depth in a lane.
The Compliance Problem Nobody Mentions
Legal marketing carries advertising rules that most AEO advice ignores completely, and they do not stop applying because the reader is a language model.
State bar rules on advertising, testimonials, claims of specialization and required disclaimers still govern what you publish. Content written to be quoted by an assistant is still content you published. If your state restricts describing yourself as a specialist without certification, that restriction covers the sentence a model might lift out of your site.
There is a second-order version of this that is worth thinking about. If an assistant summarizes your page and drops your disclaimer, you did not make that claim, but the page it summarized is yours. The practical answer is to write the qualifying language into the sentences themselves rather than isolating it in a footer, so that anything extracted carries its own limits. "Outcomes depend on the facts of each case" inside the paragraph survives extraction. A disclaimer at the bottom of the page does not.
Have your own compliance read on this. Nothing here is legal advice about your obligations, and the rules differ meaningfully by state.
Directories Carry Unusual Weight
Legal has denser third-party infrastructure than almost any other industry, and assistants lean on it heavily. Avvo, Justia, FindLaw, Martindale, Super Lawyers and your state bar directory are all read, and they are trusted more than your own site because they are harder to self-serve.
That makes directory hygiene a real task rather than a nice-to-have. Every profile should carry the same firm name, the same address, the same phone number and the same practice areas. Stale profiles from a prior firm name or a former office are actively damaging, because they make a model less confident about any claim regarding you.
An hour spent correcting five directory profiles will usually do more than a month of blogging.
Write The Answer First
The content that gets quoted answers a question directly in the first two sentences and then explains. The content that gets ignored opens with the firm's founding year and a paragraph about commitment to clients.
Take the statute of limitations question. A page that opens "In North Carolina you generally have three years from the date of the accident to file a personal injury claim, with important exceptions" is extractable. A page that opens "Navigating the aftermath of an accident can be overwhelming" is not.
The pattern is unglamorous and it works. Lead with the answer. Add the exceptions. Explain when someone needs an attorney rather than a form. Note what varies by jurisdiction.
What Not To Bother With
Stuffing your pages with the phrase "best personal injury lawyer in Charlotte" does nothing, and it may trip advertising rules on top of doing nothing. Assistants are not counting keyword density.
Paying for a service that promises to get you into AI Overviews is money set on fire. Nobody can guarantee that, and anyone selling the guarantee is either confused or lying.
Publishing thirty thin posts to cover thirty long-tail questions is worse than publishing three good ones. Thin content is the thing models most reliably skip.
Measuring It
Set up a segment in GA4 for referrals from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com, and watch it monthly. Expect the numbers to be small and to undercount, because a lot of AI-influenced traffic arrives as direct.
The more useful measurement is manual. Once a month, ask the assistants the ten questions your intake team hears most, and record whether your firm is named and whether the description is right.
Then add one question to your intake script. "How did you hear about us." When the answer starts being "ChatGPT said to call you," you have your attribution, and it will show up in that conversation months before it shows up in analytics.