A woman in Columbus opens ChatGPT at 11pm because her employer just fired her two weeks after she filed a workers' comp claim, and she wants to know if that's legal before she can sleep. She types the question. The model answers her, in plain language, and then it names a firm. One firm. Not yours, even though you've handled exactly this kind of retaliation case for eleven years and your office is four miles from her apartment.
That conversation happened without you in the room. It happens hundreds of times a day now, across ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews, Grok, and Copilot, and the firms getting named are not always the biggest or the oldest. They're the ones whose websites gave the machine something specific to trust.
The generalist page is invisible on purpose
Walk through most law firm websites and you find the same architecture. A homepage that says "aggressive representation, personalized attention." A "Practice Areas" page that lists nine of them in a tidy grid, each linking to three paragraphs of interchangeable copy. Personal injury, criminal defense, family law, estate planning, business litigation, all present, none developed. It reads like a menu.
To a human skimming, that breadth signals competence. We handle everything, so we can handle you. To an AI assistant, it signals nothing it can use. The model isn't evaluating whether you're a capable lawyer in the abstract. It's trying to answer one narrow question a real person asked, and it's looking for the source that answers that exact question most completely and most credibly. Nine shallow paragraphs each covering a different field give it nine reasons to keep scrolling. A firm three towns over with a single deep page on workers' comp retaliation gives it one reason to stop.
That's the trade the generalist site made without realizing it. It optimized for looking capable at everything and ended up citable for nothing.
Why models trust some pages and skip others
AI systems don't take a firm's website at face value. A page can claim anything. What a model is actually doing when it decides who to name is closer to what a careful referral source does: it looks for information that holds together, that shows up consistently, that answers the real question rather than dancing around it. When a page genuinely resolves the query in front of it, the model grows confident enough to attribute an answer to it. When a page is vague, or contradicts itself, or forces the model to infer what the firm actually does, the model hedges. And a hedging model does the thing that quietly kills your visibility: it recommends someone whose page didn't make it guess.
This is why a directory like Martindale-Hubbell still carries weight in ways that surprise people. It has been publishing attorney information since the 1860s, and its peer-review ratings are exactly the kind of signal a model leans on, because they don't originate from the attorney. A firm can write "highly respected trial lawyer" on its own homepage all day and a model treats it as noise, since every firm writes that. But a rating that comes from other lawyers evaluating each other is a claim the subject didn't author. That independence is the whole point. When a model finds the same picture of a firm reflected in a source the firm can't edit, the claim stops being an assertion and starts being something closer to a fact. A page's job, then, is not to boast. It's to say true, specific, checkable things that line up with what independent sources already show, so the model finds agreement instead of a lone voice insisting on its own excellence.
Notice this has nothing to do with keyword density or how many times you wrote "best personal injury lawyer in Cleveland." It's about whether your content reads like it was produced by someone who actually does this work, in enough depth that a machine can extract a real answer from it and attribute that answer to you.
Depth is a content-architecture decision
So the practice area page stops being a marketing brochure and becomes something else. Each area of law your firm genuinely wants more cases in deserves to be treated as if it were its own website. Not three paragraphs. A page that reads like the definitive local resource on that one thing.
Think about what the Columbus woman actually needed to know. Whether retaliation for filing a comp claim is illegal in Ohio. How long she has to act. What the process looks like from the first call to a resolution. What it costs to find out if she has a case. Whether you've handled situations like hers before. A page that answers those questions directly, in the language she'd use, does two jobs at once. It reassures a frightened person, and it hands an AI assistant clean, extractable material to cite when the next frightened person asks the same thing at midnight.
The FAQ portion of such a page is where this gets concrete. When you write out the genuine question a client asks, and answer it in plain, specific language, and structure that answer so a machine reads it as a discrete question-and-answer pair rather than a wall of prose, you've built exactly the unit an AI Overview lifts and attributes. Someone asks about the statute of limitations for a slip-and-fall in your state. Your page answered that question in those words. The model finds it, uses it, names you. That's not a trick. It's writing the answer to the question people are actually asking, in a form both a human and a machine can read.
State-specific depth matters here more than most firms realize, because so many legal questions are state-modified. "What's the comparative negligence rule in Georgia" is a different query than "what's comparative negligence," and the firm that wrote the Georgia-specific version is the one that gets cited when someone in Atlanta asks. A generalist national-sounding page loses that person to whoever localized their content.
The blog is not a diary
Most attorney blogs are abandoned. The ones that aren't are usually the wrong thing entirely: firm news, a partner's award, a photo from the bar association gala. None of that is what someone types into an AI assistant.
The question people ask is the content. "Can I sue if I signed a waiver?" "How much is a broken wrist worth in a car accident settlement?" "What happens to the house in a divorce if only one name is on the mortgage?" Each of those is a query someone is putting to Perplexity or Gemini right now, and each one that you answer thoroughly becomes another surface where a model can find and cite you. A deep practice area page anchors your authority in a field. The supporting posts underneath it catch the long tail of specific questions that page can't fully absorb. Together they give a model many independent places to encounter you saying credible, consistent things about the same area of law, which is precisely the agreement across sources that turns a model confident.
This is where the discipline pays off. A firm that publishes one genuinely useful answer to a real client question every couple of weeks, tied to a practice area it actually wants to grow, builds a body of work that compounds. A firm that posts nothing, or posts about its softball team, hands that ground to a competitor. The RankCommander AI Visibility Index, which has evaluated thousands of real AI answers across professional verticals, keeps showing the same pattern: the practices that get named are rarely the ones spending the most on advertising. They're the ones whose content most completely answers the question the model was asked.
What you can't see is the part that's costing you
Here's the hard part. You can't watch these conversations happen. When a prospective client asks Claude which estate planning attorney to call in your suburb and gets a name that isn't yours, no notification arrives. No line item on a report. The client you lost never knew you existed, and you never knew they were looking. Your rank-tracking tools and analytics platforms show you Google positions and site traffic. They're silent on whether an AI assistant recommended you or your competitor across seven different platforms, each of which builds its picture of you differently.
That silence is the danger. A firm can look healthy by every traditional metric while quietly going invisible in the place clients increasingly start their search. The attorney down the highway who committed to real practice-area depth two years ago may already be the default name three of these models hand out for the cases you want most, and you'd have no way of knowing.
RankCommander's scan looks at exactly the thing you can't: how your firm actually shows up when real people ask real AI assistants for a lawyer like you. It checks whether your practice-area content gives a model something citable, whether the independent sources describing you agree or contradict, whether your name surfaces at all across all seven assistants for the queries that matter in your city. It reads your visibility the way the machines do, then tells you where you stand.
Start where the cases are
There's no universal first move here, and any strategy that claims one is selling you a template instead of a diagnosis. What actually separates the firms getting named from the ones getting skipped is rarely a single fix. It's a decision about which of your practice areas deserves to become the most complete answer a worried person can find at 11pm, and then the honesty to build it that way before a competitor does.
You've spent years becoming genuinely good at what you do. That work is real, and it's earned you clients the old way for a long time. But the woman in Columbus never saw your name, and the next one won't either, not because your competitor is a better lawyer, but because the machine had a reason to trust their page and nothing to say about yours. That gap is closing in one direction right now, and it isn't yours. See exactly where you stand across every AI assistant your future clients are already asking. Run your scan at /attorneys and find out whose name comes up when the machine gets asked for someone like you.