Law firms appear in AI assistant answers mainly when someone asks who to hire or how to choose a lawyer. On questions about what the law says, assistants in our testing cited laws, ministries, the OECD and large accounting firms instead, so a firm gets named through crawlable practice-area pages, directory listings and clear local signals, not through legal explainers alone.
What does an assistant do with a legal question?
First, it decides whether to search at all. In a test we ran on 23 September 2026 with 20 German business questions, the model answered 10 without searching and cited nobody. Those answers came from what the model already knew, so no firm website had any chance of being cited.
Where it did search, it cited primary sources on 9 of 10 questions: laws, ministries, the OECD and large accounting firms. Think of a question like "what notice period applies when I dismiss an employee?" The assistant wants the statute or the ministry page, not a firm's blog post summarising it.
Firms were named only on "who should I hire" and "how do I choose" questions. For a law firm, this split matters more than anything else about AI answers. There are two kinds of question, and they work differently.
When does an assistant name a law firm?
The table below separates what we measured from what we infer. The first row comes from the 23 September 2026 test. The other two rows draw on a local measurement for a medical practice (described in the next section), which is the closest thing we have to a law-firm measurement, so treat those rows as an extrapolation to legal services.
| Question type | Example | What the sources show | Are firms named? |
|---|---|---|---|
| What the law says | "Is a non-compete clause enforceable?" | In our 23 September 2026 test, searched answers cited laws, ministries, the OECD and large accounting firms on 9 of 10 questions | Not on these questions in that test |
| Costs, risks, process | "What happens at a first consultation?" | We did not record which sources were cited for this type | In the comparable medical-practice measurement, questions about costs, risks and treatments named no providers |
| Who to hire, how to choose | "Which employment lawyer in Hamburg should I contact?" | Not recorded in the 23 September test; in the medical-practice measurement, a directory and competing practices' own websites were cited most. We expect a similar pattern for law firms, but have not measured it | Yes, this is the only type where firms were named |
One more thing from the same test: two runs of the same question named two different sets of firms. Being named once is not a fixed position. Assistants sample, and the list shifts from ask to ask. We treat a business as named for a question if any of three asks names it, and we only compare weeks on the same questions with the same number of asks.
The informational pages still have a job. They show the assistant, and the person reading, what your firm actually handles, and they give a retrieval agent something to fetch when the question turns from "what is" to "who can help".
Which sources do assistants use when recommending a firm?
On 20 October 2026 we put 20 patient questions about an orthopaedic practice in Bonn to four assistants. The practice was named in 12 of 80 answers (Claude 5 of 20, Gemini 4, ChatGPT 2, Perplexity 1).
Two patterns stood out. The directory jameda.de was cited in 12 answers. And the most cited sources overall were the websites of competing practices. Directories mattered, but practices' own sites mattered more.
For a law firm, the equivalent is probably the legal directories used in your market plus the websites of competing firms. That gives you two concrete jobs:
- Directory profiles. Make sure your firm exists in the directories that come up when you search your practice area and city, with the same firm name, address, practice areas and lawyers listed everywhere.
- Your own site. If competing firms' websites are what assistants cite, your site has to be a page that deserves to be cited for "who handles X in Y".
Local signals tie these together. An assistant answering "employment lawyer in Hamburg" searches with the city in the query. Pages that state the practice area and the location together, with an address, the courts or regions you cover and the languages you work in, match that search. A page that only says "we advise on all areas of law" matches nothing in particular. Our article on how local businesses get recommended by AI assistants covers these signals in more detail.
What should a law firm publish on its own site?
Practice-area pages
One page per practice area, not one page listing ten. Each should say who it is for, the situations you typically handle, how a matter usually runs, how you approach fees, and where you work. That is the material an assistant needs to answer "who should I hire for a contested will in Munich?"
Pages that answer pre-hiring questions
Clients ask a lot before they pick a lawyer: how to choose a specialist, what to bring to a first meeting, when a matter needs a lawyer at all, what to ask about fees. These are close to the "how to choose" questions where firms get named. Answer them plainly, one question per page, within the professional conduct rules that apply to you.
Explainers that cite the law themselves
Since assistants prefer primary sources on informational questions, do not compete with the statute. Link to it, quote the relevant provision, and add what the statute does not give: what this means in practice, common mistakes, when to get advice. That practical layer is what a firm can offer and a ministry page cannot.
Volume and markup
Markup helps machines parse a page, but it does not create visibility on its own. We have seen a site with complete markup (server rendering, Article, FAQPage and Organization JSON-LD, a sitemap) but only 75 URLs stay absent from answers on its own article's exact topic, while a site with about 190,000 crawlable URLs was crawled constantly. A firm site with a home page, an about page and a contact form gives assistants very little to retrieve. More on this in does schema markup help with AI search.
Content changes do move results. The peer-reviewed GEO paper from Princeton (KDD 2024), which coined the term, reports that its optimization framework increased content visibility in generative engine responses by as much as 40% on its benchmark of real queries.
Can assistants actually reach and find your pages?
This is where many firm sites fail before content even matters. We made the mistake ourselves. On our own site, a robots.txt rule meant to hide a single redirect page (/b) was written as a prefix. Because robots.txt rules match prefixes, from 11 August to 21 October 2026 it also closed /blog, /business and every article in our business blog to all crawlers. Bing Webmaster Tools listed our business page as "discovered but not crawled" from 11 August 2026, and Google had none of the business pages indexed. We changed the rule to /b$ on 21 October 2026. Check your own robots.txt line by line.
Being in Google is not enough either. On 19 October 2026 we asked assistants for the exact titles of 18 articles, each older than three weeks, on three client sites. ChatGPT found 1 of 18; Claude found 7 of 18. Google had indexed 29 of 30 articles of one of those sites. Being in Google's index did not mean an assistant could find the page. Submitting to Bing and using IndexNow is part of the work; see does ChatGPT use Bing.
And being fetched is not the same as being cited. OpenAI's web search documentation describes inline citations, the URLs actually referenced in the answer, and a separate, larger sources field listing every URL the model consulted. A firm's page can be read and still not appear in the answer.
Live retrieval agents such as ChatGPT-User, OAI-SearchBot and Perplexity-User fetch pages while an assistant is answering a real person. In one two-week period of logs, a client blog of a few dozen articles received 56 such fetches across 24 different articles. Those fetches show up in your server logs and are a sign that assistants are using your pages to answer real people.
How do you know whether it is working?
Not by traffic. In about two weeks of logs from three of our course websites, human visitors arriving from AI assistants totalled one from Copilot and zero from ChatGPT, Perplexity, Gemini, Claude and Meta AI. Assistants mostly answer in place. A prospective client may read your firm's name in an answer and call you directly, and your analytics will never show where they came from.
The measure is whether you are named. Put the questions your clients ask to each assistant, three times each, every week, and count the answers that name you and the sites they cite. It is cheap to run: with API prices of October 2026, a report of 20 questions across 4 assistants with 3 asks each costs us about $5.40. Our guide on how to measure AI visibility over time explains the method.
Watch the cited sources as well as the names. One observation from our own work: in our weekly measurement of 20 October 2026, Perplexity cited an article we had published on a client's site, a German company-formation advisory, which is a field close to legal services.
To see where your firm stands today, the free check generates five client questions from your homepage and shows in 25 to 40 seconds whether ChatGPT and Perplexity name you. ItGrows also does this work as a service: articles on your own domain, crawler and indexing fixes, and a weekly count across ChatGPT, Claude, Gemini and Perplexity, for $499 a month (plus a one-time $499 setup, free with 3 months or a year). Nobody can guarantee that an assistant will name a firm, and we say so up front.
Frequently Asked Questions
Do AI assistants recommend specific law firms?
Yes, but mainly on "who should I hire" and "how do I choose" questions. In our 23 September 2026 test, firms were named only on those questions, while informational questions cited laws, ministries, the OECD and large accounting firms.
Does a legal directory profile matter for AI answers?
Probably, though we have not measured it for law firms. In our local measurement for a medical practice in Bonn on 20 October 2026, a directory was cited in 12 of 80 answers, while competing practices' own websites were cited most. Keep directory profiles complete and consistent, and treat your own site as the main asset.
Will schema markup get my firm cited by ChatGPT?
Not on its own. We have seen a site with complete markup but only 75 URLs stay absent from answers on its own topic. Markup helps parsing; pages that answer real client questions, and assistants being able to reach them, matter more.
Why does the assistant name a different firm each time I ask?
Assistants sample their answers, so the same question asked twice can name two different sets of firms, as it did in our 23 September 2026 test. That is why we ask each question three times and compare weeks only on the same questions.