Brand & Positioning · 9 min read · July 15, 2026
Why specialist lawyers get named in AI answers
Ask ChatGPT or Gemini for a lawyer and you get a paragraph naming two or three firms, not ten links. Which names appear depends on how narrowly you are known.
How clients actually search for a lawyer now
The search itself has changed shape. A client used to type 'employment lawyer Munich' into Google and click one of the first three results. Now they type the whole situation: 'I was dismissed three weeks after coming back from parental leave, who handles that and what does it cost?' Questions that long and that specific increasingly get typed into ChatGPT, Perplexity, Google Gemini or the AI Overview that sits above the ten blue links. What comes back is a paragraph of prose. Sometimes it names firms. Often it lists criteria instead and leaves the client to go looking again.
The difference matters more than it sounds. With ten blue links, a client who skips you at least scrolls past you. In an AI answer there is one paragraph and perhaps two or three names, and everything else is simply absent. The traffic side of this is now measured rather than argued about: Pew Research Center, watching the real browsing of 900 US adults across 68,879 Google searches, found people clicked through to a website in 8% of visits where an AI summary appeared, against 15% of visits where none did, and clicked a link inside the summary itself in about 1% of visits. Being named in the paragraph is increasingly the whole prize. The term for the work of getting into it is GEO: Generative Engine Optimization, a phrase from a 2024 ACM SIGKDD paper by researchers at Princeton and IIT Delhi that Andreessen Horowitz later pushed into mainstream use.
Be clear about the mechanism, because a lot of nonsense is sold around it. Nobody is paying for these mentions, and no engine publishes the formula. OpenAI, Perplexity and Google all answer questions about businesses by retrieving pages and writing over them, and none of them documents how a retrieved page becomes a cited one, so any vendor quoting you weighted signals is guessing. What the outside evidence does point at is repetition across independent sources. An Ahrefs analysis of roughly 75,000 brands found that how often a brand is mentioned around the web correlates with its AI citation rate at about 0.664, roughly three times the correlation found for backlinks at 0.218. For a law firm, that volume of third-party mention is almost always earned in one field, not eight.
Why retrieval favours a narrow profile
A language model does not weigh reputations. It matches patterns, and the pattern it can act on is a name that keeps turning up in the same narrow context: on the firm's own site, in the bar directory, in bylined commentary, in the trade press, always as 'specialist in road traffic law, accident settlement after personal injury'. Repeated that consistently, the association is easy to retrieve and easy to restate.
The general practice sends the opposite signal. Family, tenancy, criminal, probate and commercial law, all on one page, all weighted the same. Nothing in that profile makes the firm the most obvious answer to any single question, so it tends to get retrieved for nothing in particular. Breadth that read as reassurance in a printed directory reads as noise to a system that has to pick two names.
This is not unfair. It is the same thing humans do. Ask a colleague at the bar for a referral and you hear 'call Dr Berg, she does nothing but employment law' far more often than 'try the firm that covers everything'. The machine is copying that habit at scale, from the written record instead of from memory.
A niche is a magnifying glass, not a sacrifice
The standard objection is that narrowing your stated focus costs you instructions. That is a fair worry about a brochure. As a description of how AI visibility works, it has things backwards. A firm presented as a specialist in medical law with a focus on birth-injury claims does not stop taking the straightforward matters that walk through the door. It adds the difficult, better-paying ones, because the person with a birth-injury claim is exactly the person typing a long, precise question into an assistant.
The advantage scales with how specific the question is. Ask an assistant for 'a good lawyer' and the candidate pool is everyone, so the answer stays generic and hands back criteria instead of names. Ask it about 'withdrawing from a car finance agreement after the auto-loan ruling' and the pool collapses to the few firms that have actually written about it. Same firm, same credentials, completely different odds of being named.
None of this requires restructuring the practice. It requires that every lawyer and every legal field has one written focus, specific enough to say in a single sentence, and that the same sentence turns up everywhere the firm is described.
Three client questions, worked through
Take three questions of the kind that reach an inbox most weeks. First: 'My employer dismissed me while I was pregnant, what can I do?' A firm described everywhere as a specialist in employment law with a focus on unfair dismissal and maternity protection is an easy match for that. A firm listing eight legal fields is a match for none of them in particular, even where the same lawyer would handle the case just as well.
Second: 'I got a warning letter over file-sharing, do I need a lawyer?' Here the firm that wins is the one that has published plain answers to that exact question, in enough places that a retrieval step trips over them. Ahrefs, looking at 1.4 million real ChatGPT prompts and the 23.4 million URLs they pulled in, found the median cited page was roughly 500 days old, so this is work that compounds rather than work that spikes. Third: 'Siblings fighting over an estate, the house has to be sold.' A probate specialist with written-up experience of co-heir disputes is the obvious name to hand back.
In none of the three does firm size, advertising budget or domain age settle it on its own. What settles it is how unambiguously and how often your name is already attached to that specific problem, across every source the system can reach.
What these systems read before they name you
These systems do not read only your website. They pull in bar directories, legal portals, review platforms, press coverage, bylined articles, interviews and sometimes forums, and the answer gets assembled from whichever of those came back. That is why inconsistency is so expensive. If your own site says road traffic law, a directory entry says general civil litigation and a newspaper quote frames you as a criminal defence lawyer, the three cancel each other out and no clear association survives.
On the technical side, be careful what you are sold. Google's own guidance is explicit that no special markup, schema type, llms.txt or AI-specific file is required for AI Overviews or AI Mode, that its generative features run on the ordinary Search ranking and quality systems, and that a page only has to be indexed and eligible to show with a snippet to be usable as a supporting link. Schema.org markup still earns its place, but as an extractability aid: it makes your name, title, jurisdiction and contact details harder to garble, and Google has said since 2018 that it is not a direct ranking factor. The rest of the work is editorial. Headings that use the words a client would use, an unambiguous statement of which specialist titles you actually hold, the case types you take, and answers written the way the question was asked. 'Holistic legal advice at the highest level' gives a model nothing it can quote.
The question-and-answer shape does real work here. Put the sentence 'What does an unfair dismissal claim cost?' on the page as a heading and answer it in plain figures and conditions, and you have written the paragraph the model needs. Google's caution is aimed at pages written separately for machines, which it treats as scaled content abuse. Answering a real client question properly is the opposite of that, and it is the difference between being named and being quoted.
Authority signals in a high-stakes field
Law is a field where a wrong answer costs somebody money or a deadline, and these systems are not yet careful in proportion. The Tow Center at Columbia Journalism Review ran 1,600 queries across eight AI search tools, asking each to identify the source, headline, date and URL of a news article. More than 60% of the responses were wrong, from 37% for the best performer to 94% for the worst, and ChatGPT misidentified 134 of 200 articles while signalling any uncertainty only 15 times and never declining to answer. Assume the same confident wrongness can attach to your title, your legal field and your address unless you have made them trivially easy to get right. The specialist-lawyer title is the strongest asset you have here, because it is regulated and checkable. Put it where it is visible on every page that describes you, not only in the imprint.
The other signals worth building are the ones a third party can verify: bylined articles in legal publications, conference papers, teaching posts, being quoted by outlets that check things, a people page with real names and real biographies, and reviews that are genuine. This is not a volume exercise. The brand-mention finding above cuts the same way as the link data, which is that being written about by other people outweighs anything you can do on your own pages. A firm with actual substance can win that. A competitor without it mostly cannot.
There is a useful side effect. The client does not stop at the answer, they check the name, and the things that made you legible to the machine are the same things that reassure a person landing on your site after an AI recommendation. Reputation work and GEO are not separate budgets.
Staying inside the professional conduct rules
None of this suspends professional conduct rules. Lawyers' advertising has to stay factual, promises of a particular outcome are not permitted, and a misleading self-description can cost more than it ever earned. That constraint sits with GEO rather than against it. The content these systems can actually use is specific, checkable and unexciting in the good sense, and language written to satisfy a bar association tends to be language a model can quote without inventing anything.
Get the labels right. The German title Fachanwalt, specialist lawyer, is protected and may be used only where it has actually been conferred; the softer formulations for a field of activity or a field of interest carry rules of their own. Precision pays twice. It keeps you on the right side of the professional rules, and it hands the machine terms that attach to something definite. A stretched self-description is both a professional risk and a weaker signal.
When in doubt, claim one legal field fewer and prove that one properly. That is the version that survives both a bar association reading it and a retrieval system trying to work out what you do.
Where to start, in order
Start by looking. Type the questions your target clients would actually type into ChatGPT, Gemini and Perplexity, and read what comes back. Write down which firms get named instead of you and what about their profile is easier to read. Do it more than once, with the wording varied, because these answers move between runs. That list is your baseline, and it is usually more specific than any audit you could buy.
Then write the sentence. One per lawyer and one per legal field, carrying the focus, the typical case types and the region, and put that exact sentence everywhere the firm appears: your own site, the bar directory, review profiles, LinkedIn, press biographies. Keep the title, the jurisdiction and the contact details identical across all of them. Build pages that quote real client questions and answer them. Then go and earn mentions you do not control, which is the slow part and the part that moves things.
And treat it as maintenance, not a project. Models get retrained, retrieval shifts, competitors tighten their own positioning, and new areas of law arrive before anyone has written about them. The reason it deserves a standing slot in the week is scale: ChatGPT reported 900 million weekly users in late February 2026, and Google has put AI Overviews in front of more than two billion monthly users across 200-plus countries and 40 languages. The firm that keeps one sharp profile fed with precise answers stays the obvious name to hand back.
Common questions
Will narrowing my stated focus cost me instructions?
No. A narrow focus does not turn away the straightforward matters that come to you anyway. It adds the ones where somebody is searching deliberately, and those are usually the better-paying files. The more precisely a client phrases a legal question to ChatGPT or Perplexity, the smaller the pool of firms that plausibly match it, and the more a specialist stands out. You can run several focuses. Describe each one sharply and separately rather than blending them into one all-purpose paragraph.
Is GEO compatible with lawyers' advertising rules?
Yes, within the ordinary limits on lawyers' advertising. Content has to stay factual, must not promise an outcome, and may use the protected title Fachanwalt only where it has actually been conferred. Those limits point the same way as the work itself, because specific, checkable, unsensational content is what these systems can quote, while an inflated claim is both a professional risk and a weak signal. In doubt, claim one legal field fewer and back that one up properly.
How do I find out whether ChatGPT or AI Overviews name my firm?
Test it yourself, and test it more than once. Ask each of the main assistants the questions your clients would ask, in their words, something like 'good unfair dismissal lawyer in Frankfurt'. Note whether you appear at all, which firms appear instead, and how their focus gets described back to you. Vary the phrasing and repeat, because answers are not stable between runs or between engines. One Ahrefs study found only about 6 to 8% of the URLs ChatGPT cites overlap with Google's top ten for the same query, so your search rankings tell you very little about this.
Read on