Measurement & Reporting · 9 min read · July 15, 2026
How Often Does the AI Recommend Your Law Firm? Making Visibility Measurable
When a client today types "good specialist attorney for employment law in Cologne" into ChatGPT, a language model decides which firms it names. Whether yours is among them, you can measure – not guess. This guide shows you how to make your AI visibility as a law firm visible and controllable with fixed prompts, metrics and a simple system.
Why AI visibility suddenly matters for law firms
The client's journey increasingly begins not with Google but with a chat window. People with a legal problem are uncertain, afraid of the costs and first want to understand what it's even about. These are exactly the questions they now ask ChatGPT, Gemini or the AI overview in Google search: 'What does a divorce cost?', 'Do I have to sign the warning letter?', 'Do I need a lawyer for a wrongful-dismissal claim?'. At the end of such dialogues, a sentence like 'For that you should turn to a specialist attorney' often falls – and sometimes concrete names follow.
For you as a firm, this shifts the decisive moment. Formerly the question was: do I rank on page one of Google? Today it is: am I even named in the AI's answer when someone asks for exactly my specialization and my city? This new discipline is called Generative Engine Optimization, GEO for short. It decides not about clicks on position three, but about whether your name is spoken as a recommendation or whether the AI quietly passes over you.
The uncomfortable part: most firms simply don't know whether they are recommended. They have never tested a single question. Whoever doesn't measure relies on a gut feeling – and that is especially deceptive with AI answers, because the same question is answered very differently in Berlin than in a town of 40,000 inhabitants.
What 'visibility' concretely means with an AI
Unlike with Google, there is no ranking list you can conveniently read off in the browser. An AI answer is running text. Your visibility therefore shows itself not in a position, but in three questions: are you named at all? Are you recommended or merely mentioned in passing? And is the context depicted correctly, i.e. your real specialization and your location?
An everyday example. You are a specialist attorney for family law in Freiburg. A test prompt reads: 'I'm getting divorced in Freiburg and am looking for an empathetic attorney for family law. Whom can you recommend?' The AI can name you by name, it can write only 'a specialist firm in the Freiburg area,' or it names three other firms and not you at all. All three cases are measurably different – and only the first actually brings you clients.
It is also important to distinguish between a neutral mention and a genuine recommendation. 'There are several firms in Freiburg such as X, Y and Z' is something different from 'For empathetic advice in family law, the firm X is often named.' When measuring, you should cleanly separate the two, otherwise you talk your situation up to be better than it is.
The first step: an honest baseline measurement
Before you optimize anything, you need a starting value. Sit down with your team and collect the 15 to 25 questions your typical clients really ask. Not phrased in legally correct terms, but the way an insecure person types: 'Boss wants to lay me off for operational reasons, what do I do?', 'Rental deposit not returned, which lawyer helps?', 'Renouncing an inheritance deadline, do I need a notary or lawyer?'. This is exactly the language AI users use too.
Now you ask each question one after another in ChatGPT, in Gemini and in the Google AI overview. For local questions, always add your city or region, because without a local reference the AI rarely recommends concrete firms. For each answer you note three things: were you named? Yes or no. How were you depicted? Recommendation, mention or incorrect. And who was named instead? Those are your real competitors in the AI space, often quite different from the ones you have in mind.
Record everything in a plain table: one row per question, columns for the three AI systems, plus the date. This table is your baseline. At the start it is often sobering, but it is worth gold, because in three months you'll see in black and white whether anything has moved.
The metrics that really matter for your firm
From your table you can derive a few simple metrics that you track permanently. The most important is the mention rate: in what percentage of your test questions does your firm name appear? If you start at 8 percent and reach 35 percent after half a year, that is solid progress you can also prove to your partners.
The second metric is the recommendation rate: how often of that is it a genuine recommendation and not merely a mention in a list? The third is the correctness rate: are the legal field and location right? It does you little good if the AI names you as a traffic-law attorney in Munich even though you do medical law in Hamburg. Such errors cost you mandates and damage your profile.
Track these three rates separately per legal field. Many firms discover that they are already well visible in employment law but completely invisible in inheritance law, even though both belong to their offering. Without this breakdown you would drown in the average and never recognize the actual gap.
Why the AI names other firms of all things
When the models pass you over, that is rarely by chance. Language models rely on what they can find and classify about you in the open web. Firms that are consistently named usually have three things in common: clearly structured expert articles on their own site, unambiguous location and specialization details, and mentions on independent sources such as specialist portals, directories and the local press.
A frequent reason for invisibility is the classic firm site that consists only of phrases: 'competent, committed, personal.' For a language model that is empty of content. It cannot derive from that that you are the attorney who handles operational dismissals in the chemical industry in the Ludwigshafen area. The more concrete and thematically coherent your content is, the more confidently the AI assigns you to the fitting matter.
Timeliness also plays a role. A piece on a fresh ruling, cleanly explained and with your assessment, is a strong signal to the AI that you are active in this topic. Pure advertising texts without professional substance are not.
From measurement to improvement: an example
Take a mid-sized firm with a focus on traffic law in Dortmund. The baseline shows: across 20 test questions it is named only twice, both times as a mere mention. Competitors who repeatedly appear have detailed advisory pages on 'objecting to a fine notice,' 'circumventing the MPU,' 'averting a driving ban' – each with concrete deadlines and local reference.
The firm reacts in a targeted way: it writes a well-founded, well-structured expert article for each of its five most frequent case triggers, clearly names city and jurisdiction in them, and lists itself with a complete profile in two reputable attorney directories. After three months it repeats exactly the same 20 questions. The result: eight mentions, four of them genuine recommendations. The mention rate rises from 10 to 40 percent.
What is decisive here is the repetition with identical questions. Only that way is the comparison clean. Whoever asks different or easier questions in the second run measures nothing but lies to himself. The discipline of always using the same question catalog is the core of a robust measurement system.
A monthly measurement rhythm that works in everyday firm life
You don't need to buy an expensive tool for this. A fixed appointment once a month suffices, at which someone from the team goes through the question catalog and documents the answers. Count on one to two hours for 20 questions across three AI systems. It is important that you work in a freshly opened, non-personalized window, so that the AI doesn't name you merely because you searched for yourself beforehand.
Add a short log to each measurement: what has changed since last time, which new content went online, which competitors have newly appeared? That way you connect the bare number with a cause. If your mention rate rises after the publication of three expert articles, you know the effort paid off.
Keep in mind that AI answers fluctuate. The same question can be answered differently today than tomorrow. That's why the trend over several months counts, not the snapshot of a single day. A single non-mention is no drama, but a continuous downward trend over three measurements is a clear warning sign.
Limits, honesty and the professional-conduct framework
Be realistic about all of this. You can measure and influence AI visibility, but not force it. No reputable provider can guarantee you that ChatGPT will name you on a particular question. Whoever promises that is selling you illusions. What you can steer is the probability – through good, concrete, findable content and a clean profile on the web.
As an attorney you additionally operate within the framework of professional conduct law. Your AI-optimized content must remain just as factual and truthful as any other firm communication. Sensational promises or invented success rates, only so that the AI responds, are not only disreputable but risky under professional-conduct law. Good GEO for law firms is always professionally grounded, never loud and boastful.
And finally: measurement is a means, not an end. The number in your table is only worth as much as it leads to better, more honest content and, in the end, to fitting mandates. If you keep that in view, AI visibility turns from a diffuse gut feeling into a figure you steer like any other metric of your firm.
Common questions
Is it permissible under professional-conduct law to optimize my firm specifically for AI systems?
Yes, as long as you remain factual and truthful. Attorney advertising law permits information about your activity but prohibits misleading or boastful statements. For law firms, GEO means above all presenting your real specializations clearly, concretely and findably. Invented success rates or sensational promises just to appear in AI answers are both disreputable and risky under professional-conduct law.
How many test questions do I need for the measurement to be meaningful?
To start, 15 to 25 questions that cover your typical case triggers suffice. More important than sheer quantity is that you break them down per legal field and use exactly the same phrasings on every repetition. Only that way is the comparison over the months clean. Take the questions the way real clients type them, not in legally correct language.
Why does the AI name other firms, even though I've been in the market for 20 years?
Experience alone is invisible to a language model. It relies on what it can find and thematically classify about you in the open web. Firms that get named usually have concrete expert articles, unambiguous location and specialization details, and mentions on independent portals. A firm site made of pure phrases like competent and committed gives the AI no usable points of reference.
Read on