By Bryan Osima
With every passing day the gatekeeping powers of the AI search engines — to direct and control web traffic — continue to grow at a rapid pace.
Increasingly, potential legal clients searching for help online are not typing simple keywords and seeing a list of search results. They are interacting directly with agents like ChatGPT, Perplexity, Google’s AI Overviews and AI Mode, Gemini, and Copilot. They are asking questions, in plain language, and those agents are providing direct answers or pointing them to specific lawyers.
Only a small number of sources get cited in those answers now. An even smaller number of law firms get recommended for a search query. The question now is whether your firm is the one getting cited or recommended, because the other paths to visibility are shrinking.
This guide is about how to be the answer. The discipline goes by several names, and I’ll untangle those in a moment, but the work underneath is consistent — making your firm something AI systems can recognize, trust, and confidently recommend.
We’ll dive into how the machines make decisions, what entities are and why they now matter more than keywords, the signals that build machine trust, how to structure content AI can use, and the important shift you need to internalize — moving away from targeting simple keywords to covering the client problems behind the prompts people actually type.
It’s a deep dive and if you want the condensed version first, my earlier writing on SEO for lawyers covers the foundation this builds on.
AEO, GEO, AIO: The Alphabet Soup, Untangled
The marketing industry has produced a pile of acronyms for this work, and the first useful thing to know is that they largely describe one discipline.
AEO, Answer Engine Optimization, grew out of optimizing for direct answers like featured snippets, People Also Ask boxes, voice assistants, and now AI answer boxes.
GEO, Generative Engine Optimization, is the newer term aimed at generative systems that write full answers, like ChatGPT, Perplexity, Gemini, Google’s AI Overviews.
AIO, AI Optimization, is the catch-all some marketers use for all of it.
You will sometimes see people draw fine distinctions between these. The distinctions matter far less than the shared goal — when an AI system composes an answer to a question your future client asked, your firm’s content is understood, trusted, and cited, and where the question calls for a lawyer, your firm is the one named.
In this guide I’ll mostly say AI search and GEO. Whatever your vendor calls it, the underlying signals are the same, which is convenient, because it means you don’t need three strategies. You need one, done properly.
How AI Search Differs From Traditional SEO
Traditional SEO is a ranking contest. Ten links, per page, ordered by relevance and authority, and your goal was to be as high on the list as possible. Whoever was searching got many options from which to make an independent choice.
AI search is a selection contest. The system reads many sources, synthesizes one answer, and cites or names a handful. The search user often sees no list at all.
For a large and growing share of searches, and legal questions trigger AI answers at among the highest rates of any industry, the answer is the whole experience.
That changes the math in several ways worth understanding.
Position one on a search engine results page is no longer the prize.
Research on AI citations keeps finding that a meaningful share of cited sources do not rank in the traditional top ten for the query.
The AI is not reading the rankings off the page; it is choosing sources it can extract a confident answer from.
A page can rank first and be absent from the answer, and a page can rank fifteenth and be the answer’s backbone.
One question becomes many. When you ask an AI assistant something, it often runs multiple searches behind the scenes, rephrasing your question several ways, then synthesizes across everything it found.
Search marketers call this query fan-out. The practical effect is that you are no longer optimizing for a keyword. You are trying to be relevant and present across a neighborhood of related questions.
In the answer box, you are not just competing with other firms. You are competing with bar association pages, legal encyclopedias, directories, and news sources for citations, and with a synthesized answer that may satisfy the person before they click on anything.
The winnable ground is the part of the answer where expertise, locality, and recommendation live.
And none of this replaces traditional SEO.
AI systems find your content through crawling and search indexes. A site that is technically unsound, slow, or thin on content fails AI search for the same reasons it fails Google.
Everything in my SEO guide for law firms still applies. GEO is not SEO’s replacement. It is SEO’s harder exam.
How AI Assistants Actually Choose Who to Recommend
To optimize for these systems, it helps to know, in plain terms, how they produce an answer. Three mechanisms do most of the work.
The first is training. These models have read an enormous amount of the public web, and what they absorb shapes what they “know” — which firms exist, what they’re known for, how often and in what contexts they were mentioned. You cannot edit training data after the fact, but you can influence what the next round absorbs, and the consistent, widely corroborated presence this guide teaches is exactly what sticks.
The second is retrieval. For current or specific questions, the assistant searches the live web, pulls candidate pages, and grounds its answer in what it retrieved. This is where structure matters enormously. The system skims fast, extracts passages, and favors pages where the answer is clear, self-contained, and attributable.
The third is corroboration. Before naming a firm in a recommendation, these systems effectively cross-reference information across various sources. Does the website, the Google Business Profile, the directories, the reviews, and the coverage all tell the same story about who this firm is, where it practices, and what it handles? Agreement builds confidence. Contradiction kills it. An AI assistant is cautious about recommendations in high-stakes areas, and law is about as high-stakes as it gets, so the recommendation goes to the firm the whole web agrees on.
It’s important to keep those three factors in mind: what the machines absorb, what they retrieve, and what they can corroborate. Every tactic in the rest of this guide feeds one of them.
Entities: How AI Thinks About Your Firm
There is an important concept that separates a real AI search strategy from a list of tips: entities.
An entity is a distinct thing the machine can identify — a person, an organization, a place, a practice area.
As an example, let’s take the string of letters “Smith Law”. That text string could be a hundred different businesses. But Smith Law, the personal injury firm at a specific address in a specific city, founded by a specific attorney with a specific bar number, is an entity.
Search engines have organized knowledge this way for years through knowledge graphs. Language models think in the same shapes. They connect entities to attributes and to other entities.
Your firm is an entity. Each attorney is an entity. Your city, your practice areas, the courts you appear in, are all entities.
The machine’s understanding of your firm is, in effect, a web of these connections.
When someone asks “who handles contested custody cases in Fort Worth,” the system is not matching those words against pages. It is looking for entities of type law firm, connected to the practice area family law and the attribute custody, connected to the place Fort Worth, with enough corroborated trust attached.
If your firm exists in the machine’s understanding as exactly that entity, you are a candidate for the answer. If you exist as an ambiguous text string with inconsistent details scattered across the web, you won’t be a candidate, no matter what your homepage says.
So the foundational GEO work is entity work, and it looks like this:
Give every entity a home
Your firm’s entity home is your website, and each attorney should have a substantial bio page that serves as that person’s canonical source. It should contain their full name, credentials, bar admissions, practice focus, history, photo, etc. There must be an authoritative page the whole web can agree points at the real you.
Make the entity unambiguous
The firm’s name, address, and phone number should be identical everywhere they appear, which readers of my local SEO guide for law firms will recognize as NAP (Name, Address, Phone number) consistency doing double duty.
Similarly, an attorney’s name should be identical everywhere it appears online, because “Robert J. Smith,” “Bob Smith,” and “R. Smith, Esq.” fragment the entity and cause confusion.
Connect the entity to its other homes
Structured data and schema markup — code you add to your web pages that help the search engines understand what your content is all about — is how you make this connection in language the machines understand easily. Learn more about schema markup for lawyers.
There are different types of schemas that represent different types of entities. Use LegalService or Attorney schema on the site, Person schema on bios, and the sameAs property pointing to the firm’s and attorneys’ profiles — the Google Business Profile, the state bar listing, the major directories, LinkedIn, etc. Each connection is a thread that helps the machine resolve every mention across the web into one entity.
Then build the entity’s attributes deliberately. This means that every practice area page, every location page, every case result, every piece of coverage attaches attributes to the entity and demonstrates who handles what, serves where, and achieved what.
As the entity grows richer, the machine’s confidence grows with it. This is why thin, duplicated pages do nothing. They add no attributes, and as I covered in my previous digital landfill post, they increasingly don’t even get indexed.
The Trust Signals: Links, Mentions, Citations, and Authority
Once the machine knows who you are, the question becomes whether it trusts you enough to cite or recommend you.
Trust is built from signals, and the signal mix has shifted in ways lawyers should understand.
Links still matter, but differently. In classic SEO, backlinks were the currency of authority. They still count. Links get your pages crawled, ranked, and into the retrieval pool. But the generative systems also weigh something links were always a proxy for, and can now read directly.
Unlinked mentions now carry real weight. When your firm is named in a news story, a legal publication, a “best firms in” roundup, a podcast transcript, or a community page, the machine reads the mention, the context, and the sentiment, link or no link.
Every credible mention is corroboration that this firm exists, practices here, is spoken of in this way. A firm mentioned across dozens of independent, credible sources reads as established. A firm the web is silent about reads as a question mark, and question marks don’t get recommended for legal matters.
Citations, in the AI sense, are the prize — being the source an answer quotes. You earn them by publishing content worth quoting, which is covered below, and by being the kind of source machines prefer to quote: named, credentialed, specific, and consistent.
What this means in practice is that digital PR has quietly become core GEO infrastructure. The following all now heavily influence your standing with the AI engines:
- Commentary in legal and local press
- Contributions to bar publications
- Community involvement that produces coverage
- Directory profiles on Avvo, Justia, FindLaw, Martindale, completed properly and kept consistent
- Reviews, which the machines read as independent human corroboration of what you claim about yourself, and which influence recommendation prompts heavily
None of this is new advice. What’s new is the mechanism. You are no longer just building reputation with humans who might hire you. You are building the body of evidence the machines consult before they say your name.
Author Entity Signals: Why the Person Behind the Content Matters
Legal questions are what search engineers call YMYL, your money or your life content, answers that can truly expose someone to harm if they’re wrong. The systems answering them apply extra scrutiny, and one of the strongest scrutiny signals is whether content is connected to an accountable, credentialed human being.
This is where solo and small firm lawyers hold an advantage most never use. You are a real, specific, verifiable expert. The machines can check. The byline on the article, connected to the bio page, connected to the state bar listing, connected to the LinkedIn profile, connected to the directory profiles, all agreeing that this person is a licensed attorney in this state practicing this kind of law. Content carrying that chain of accountability is structurally more trustworthy than content published by nobody specific, or just a general firm admin account, and the machines treat it accordingly.
Building the author entity is concrete work and the following standards must be maintained:
- Every substantive piece of content carries a real attorney byline. Not “Staff” or, worse still, the marketing agency’s name. An individual attorney should be mentioned by name.
- Every byline links to a substantial bio page with credentials, bar admissions with years, education, practice focus, notable work, professional memberships, and a real photo.
- The bio page carries Person schema with sameAs links to the attorney’s bar profile, LinkedIn, and directory listings, welding the on-site author to the verifiable off-site identity.
- The attorney’s name appears in one consistent form everywhere — website, bar associations, directories, LinkedIn, publications, etc.
Where the attorney publishes or is quoted elsewhere, those pieces reference back, thickening the entity from the outside.
One more implication worth stating plainly, because it connects to what I previously wrote about in a post on marketing ethics and accountability: putting your name on content means owning what it says.
That is a compliance obligation, but in AI search it is also the strategy. The accountability the ethics rules demand and the accountability the machines reward are the same. Firms that publish carefully, under real names, with verifiable claims, are simultaneously safer and more visible. That alignment is rare in marketing. Take advantage of it.
Content Structure AI Can Actually Use
When a generative system retrieves your page, it doesn’t experience your design. It parses text, fast, looking for extractable answers it can attribute. The structure of your content decides whether it finds them. Applying the following rules to your content will be very helpful:
Lead with the answer
For any page built around a question, answer it in the first sentences, clearly and completely enough to stand alone, then go deep for the human who keeps reading. The buried answer, warmed up by four paragraphs of throat-clearing, loses to the page that answers immediately.
Make headings do real work
Use question-shaped H2s and H3s html heading tags that match how people actually ask, with each section answering its heading fully and standing on its own. Generative systems lift passages, and a self-contained section is a liftable passage. A clever heading that hides what the section covers is invisible to a machine scanning for the answer to a specific question.
Stick to one problem per page, covered properly
A page that thoroughly handles one client question, with the state-specific and local specifics only you know, beats a page that touches on ten questions. Depth on a narrow question is exactly what wins citations, because the machine wants the best source for this answer, not the biggest page.
Use plain language
State the essentials in plain, easy to read and understand, text. Who you are, what you handle, where you practice, should be stated in words on the page, not locked in images, sliders, or scripts. Your jurisdiction should be named explicitly, because legal answers are jurisdiction-bound and the machine needs to know your content is about Georgia, specifically, as an example.
Use structured data where relevant
Schema markup for entities like LegalService, Attorney, Person, FAQPage where real questions and answers exist, LocalBusiness details and more, exists. But schema doesn’t rank you; it simply removes doubt about what your pages mean.
Keep your content visibly current
That means using real dates on your content, making updates when laws change, and removing any stale claims. Retrieval favors fresh, up to date content.
Produce substantive, authoritative content
Original, substantive, legal content, written by or with someone who actually knows what they’re talking about is what the engines favor. Structure gets the machine to the answer but only substance makes the answer worth citing.
Moving From Keywords to Client Problems
There is a mental shift that this whole discipline turns on, and it’s one that many firms haven’t made or properly understood.
Keyword-era thinking asks: what search terms have volume, and how do I rank for them? “Car accident lawyer Atlanta.” “Divorce attorney near me.” Those are short strings, guessable, finite, trackable.
Prompt-era thinking asks a different question: what problems do the clients I want actually have, and what do they type when those problems are in front of them at 11pm?
Because that’s what prompts look like.
It’s no longer just about “car accident lawyer Atlanta” but “I was rear-ended on I-285 last week, the other driver’s insurance offered me $3,000 and my neck still hurts, should I take it or talk to a lawyer first?”
Prompts are long, situational, specific, and written in the language of the problem, not the language of legal services. The person often doesn’t know they’re shopping for a lawyer yet. They’re trying to understand their situation, and the AI’s answer is where they learn whether they need one, what kind, and, increasingly, who.
You cannot keyword-target that. The exact phrasing is unrepeatable, and the assistant fans it out into multiple searches anyway. What you can do is own the problem space it belongs to. That rear-end prompt lives in a family — low insurance offers, delayed injury symptoms, fault disputes, whether to involve a lawyer, what a case is worth.
A firm that has covered that family thoroughly, each question answered properly with local specifics, under a credentialed byline, is present across the whole neighborhood of prompts, however any individual one is phrased.
So the method looks like this:
- Define the client problems you want to be found for, which is really the market-definition work from my small firm marketing guide: the case types you want, the clients who bring them, the geography.
- Mine your intake conversations. The questions clients actually ask, the misconceptions you correct weekly, what they wish they’d known sooner. Write them down for two weeks and you’ll have the raw prompt research most agencies fake with tools.
- Ask the assistants yourself. Put your clients’ real situations to ChatGPT, Perplexity, and Google’s AI features, phrased the way clients phrase them. Read what comes back, who gets cited, who gets named. That is your competitive research now.
- Build content that covers problem families, not keywords: one real question per page, answered fully, with jurisdiction and local texture, connected by internal links into a topic the machine can recognize you own.
Notice the de-emphasis on exact-match keyword thinking. The words still matter, because retrieval still starts with search and it has to be relevant. But coverage of the problem, in the client’s language, with real depth, is what gets you into answers across a thousand phrasings you never targeted.
How Different Prompts Change Who Gets Recommended
Not all prompts are the same kind of question, and the type of prompt changes which signals decide the answer. This is worth understanding, because it tells you where each investment pays off.
Informational prompts
“What happens if I die without a will in Texas?” No firm needs to be named at all; the contest is for citation. Winners here are the clearest, most authoritative, most extractable answers. That is informed by your content structure, author signals, and depth. This is where your content earns its way into the conversation before the person knows they need you.
Recommendation prompts
“Who’s a good estate planning attorney in Plano?” Now the machine is naming firms, and it leans hard on corroboration — the Business Profile, reviews, directories, mentions, consistency, locality. Your entity work and reputation evidence decide this one. Content alone won’t.
Comparison and validation prompts
“Is [firm] legit?” “Reviews of [firm].” The machine assembles what the web says about you specifically, assessing reviews, coverage, ratings, complaints, your own site. This is where reputation management shows up in AI answers verbatim, and where a thin or contradictory footprint reads as a warning.
Situational prompts
Situational prompts that describe an event that has occurred or a situation a user is experiencing fall under this category. An example would be someone who has just had a rear-end collision.
These types of prompts blend everything and often result in longer conversations and follow up questions and answers between the searcher and the agent. The answer educates, then often ends with whether and how to get help. In these contexts, locality plus corroborated expertise determines who gets suggested. These are the highest-value prompts in law, because they catch people at the decision moment, and they reward firms strong across all of it — content, entity, authority, reviews.
Two more dynamics greatly shape recommendations. Locality is the first. Assistants increasingly account for where the person asking the question is located, which routes the answer through your local entity signals, the same infrastructure local SEO built. The second is consensus. when the machine must pick names in a high-stakes category, it favors the firms the corpus agrees on, mentioned credibly, reviewed consistently, described the same way everywhere. There is no trick that substitutes for being truly, verifiably established in your market.
The good news for smaller firms is that being established in your defined market is achievable, and the machine doesn’t care about your headcount. It cares about the evidence.
Advertising on AI Search: Where That Stands
A fair question at this point is whether you can simply pay your way into these answers. The answer is that while advertising inside AI search now exists, it is moving fast, and it is not yet a mature channel for law firms. Here is where things stand as of this writing.
Google is the most practical entry point. Ads now appear in and around AI Overviews and AI Mode, and you don’t buy those placements directly — eligible campaigns from your existing Google Ads account are extended into the AI surfaces when you opt into Google’s AI-powered campaign types. For a firm already running search campaigns, that means some AI visibility is already reachable through tools you know, with familiar billing and reporting. Local Services Ads also continue to operate alongside all of this, unchanged.
OpenAI began piloting ads in ChatGPT in early 2026, and they have rolled out quickly since — clearly labeled sponsored placements that appear alongside answers, kept separate from the organic response itself. The formats so far lean toward consumer products and shopping, pricing has favored large advertisers, and there is no established playbook for legal services yet. Microsoft sells AI placements in Copilot through its regular ads platform. Perplexity, notably, tested ads and then walked away from them entirely, betting that an answer engine’s value is trust.
Two things are worth understanding before spending a dollar here. First, what you are buying is a labeled ad next to the answer — not the answer. No platform sells the recommendation itself, and everything in this guide about earning citations and recommendations still applies with ads running or not. Second, an ad inside an AI answer is still lawyer advertising. The identification requirements, the claim restrictions, and the filing regimes in states that have them apply to a sponsored AI placement the same way they apply to a billboard, and the formats are new enough that your compliance review has to be, too.
The sensible posture for most firms right now could be to let your existing, well-managed Google campaigns extend into the AI surfaces and watch what they produce, while treating the newer platforms as experiments to revisit rather than budgets to commit, and keep the real investment in the earned visibility that this guide is all about. Paid placement in AI search will mature. The firms it will reward most are the ones the machines already trust.
Measuring AI Visibility, and What Not to Do
Measurement here is younger and messier than with traditional SEO reporting, so it helps to be clear-eyed about what’s currently trackable.
You can test prompts directly and track your presence over time, but always keep in mind that answers vary by user and session.
You can watch referral traffic from AI platforms, which is growing and increasingly visible in analytics. You can watch branded search and direct traffic lift, which often follows AI mentions since many people verify a recommendation by searching the name. And you can ask every new client how they found you and actually record the answer.
As for what not to do? Every dark art from old SEO has an AI-era imitation, and they fail the same way now.
Mass-produced AI content pointed at prompt lists now consistently fails. Search engines decline to index it and generative systems have nothing distinctive to cite.
Fake reviews and purchased mentions are corroboration fraud aimed at systems built to cross-check, and they now carry federal, platform, and bar consequences stacked together.
Hidden text and prompt-injection tricks aimed at manipulating AI answers are the new keyword stuffing, except the audience being tricked is a machine that gets better at spotting tricks every quarter, attached to platforms that penalize them. In AI search, the manipulation-resistant signals are the ones that are expensive to fake because they’re real — consistent identity, credentialed authorship, original depth, genuine reviews, and earned mentions.
Summary
AI engines now firmly control the search experience. It is now the path through which a large share of your potential clients pass through, and the systems answering them choose sources and firms by evidence: a clear entity, consistent everywhere; credentialed humans behind the content; structure a machine can extract; depth on the problems your clients actually have; and a web-wide record that corroborates all of it.
None of that is a hack, which is the point. The firms winning AI recommendations are the ones that built the evidence, and the ones losing are often bigger, louder, and invisible to the machines because their footprint is generic, inconsistent, or anonymous. This game rewards care and discipline over a large budget.
If you want help working out what this looks like for your firm, your market, and your practice areas, you can book a strategy session here. A straight conversation about where you stand in AI search today, and what it would take to be the answer the engines return.