Why Your Firm Does Not Appear When Someone Asks an AI for a Lawyer

Something sells well right now: a package that promises to get your firm named when a client asks an assistant for a lawyer. The pitch usually involves schema markup and a file called llms.txt.

Before buying any of it, two things are worth checking, and both take about a minute.

Rule out the thing you can check in a minute

A crawler that cannot read your site cannot quote it. That is the one failure in this whole subject that is absolute rather than probabilistic, and it is the first thing to eliminate.

Open your own robots.txt. Type your domain, add /robots.txt, and read what comes back. You are looking for any line that names one of these and disallows it.

A line reading Disallow: / under any of those names is a closed door, and no amount of content fixes it. A crawler your file never mentions is allowed by default, which is the normal and usually correct state.

Worth knowing before you go looking: this is rarely the problem. Of 15 law firm sites checked, one blocked GPTBot and ClaudeBot. Every one of the other fourteen was readable, and all 15 declared a sitemap.

llms.txt is not the lever it is sold as

An llms.txt file sits at the root of a site and lists what it contains, in a form meant to be easy for a language model to read. Vendors sell it as the fix.

Seven of those same 15 firms already publish one. If llms.txt were the deciding factor, nearly half the firms checked would already be winning, and the firms asking this question would not be asking it.

No major assistant has committed publicly to reading the file, and none of the guides on this subject makes a measured claim that it changes anything. It is cheap and harmless, so there is no strong argument against having one. There is also no evidence it is why a firm does or does not get named.

What actually comes back when somebody asks for a lawyer

This is the part that answers the question, and it is checkable.

Six ordinary lawyer-seeking searches were run across different practice areas and cities. They were phrased the way a client types them. Three of the six: “who is the best car accident lawyer near me”, “I need a divorce attorney in Atlanta”, “find an estate planning attorney in Dallas”. Every result was then sorted by what kind of page it was.

Those directory results are worth understanding before you react to them, because most legal directories for lawyers share a single owner. Two of the six searches returned no individual firm site anywhere in the results. Both were the most client-sounding phrasings, the ones asking for a recommendation rather than naming a firm type. Avvo alone accounted for ten results and LawInfo for seven.

The worst case was the highest-intent query of the lot. “Who is the best car accident lawyer near me” returned almost nothing useful. The results included a hardware supplier’s corporate blog, a city government blog, a Swedish insurance subdomain and a university wiki page, all carrying “best car accident lawyer” content. One real firm appeared. The rest was junk sitting on borrowed domains.

Why that is the answer

An assistant answering “who should I call about my car accident” does not know your firm. It runs a search, reads what comes back, and writes a summary of that.

So the pool it draws from is the pool measured above. When a third of that pool is directory pages and another sixth is roundups, the summary names directories and roundups. Not because the model prefers them, but because they are what was there.

That reframes the problem usefully. The question is not how to talk to the model. It is how to be present in what the model finds. That is the same search layer firms have competed in for twenty years, with the directories sitting on top of it.

What follows from that

• 
  Claim the directory profiles, because they are what gets returnedAvvo and LawInfo together produced seventeen of the fifty-six results. Claiming a profile is free, and an accurate claimed profile is a cited mention of your firm in a page the assistant is likely to read. This is the least glamorous item on the list and the most directly connected to the measurement.
• 
  Answer the question your clients actually askThe queries that returned no firm sites were phrased as requests for a recommendation. Pages built to answer a real question in plain words get quoted because they are quotable. A practice area page that only describes your firm has nothing in it for a summary to lift.
• 
  Put the facts where they can be readName, address, phone, practice areas, bar admissions and jurisdictions belong in text on the page. Not inside an image, and not in a graphic header. An assistant summarizing your firm can only use what is written down.
• 
  Add the schema, but hold it in proportionEvery one of the eleven guides checked recommends structured data, and it is sound practice that costs little. Treat it as hygiene rather than as the thing that will change the outcome, because no page recommending it offers evidence that it does.

The one thing worth building for

If the summary is written from whatever the search returns, then the page that gets quoted is the page that answers the question in words a summary can lift.

Most law firm practice area pages are not written that way. They open with how long the firm has practiced, how committed it is, and how free the consultation is. None of that answers what a worried person typed, so there is nothing in it for a summary to lift. The directory page listing fifty lawyers in the city answers the question better than that does.

The fix is unglamorous. Take the five questions clients ask on the phone before they hire you, and answer each one on the page in plain sentences. They are usually some version of these five. “What is this going to cost”, “how long does it take”, “do I have a case”, “what happens first”, “what do I need to bring”.

Pages written that way get quoted because they are quotable. The law firm website pages that manage it are usually the plain question-and-answer ones, not the practice area pages. They also convert better for the humans who read them, which is the part that pays whatever the assistants decide to do next. A page built only for machines is a bet on a system nobody outside these companies can see.

What nobody can tell you

Of eleven guides on this subject, not one admits a limit on what is known. That silence is worth noticing, because the limits here are substantial.

Nobody outside these companies knows how an assistant decides which sources to name. The systems change without notice, the same question can produce different answers on different days, and there is no equivalent of a rank tracker that tells you where you stand. Any firm selling a guaranteed position in an AI answer is selling something it cannot control, and the reasoning is the same as for a guaranteed Google ranking.

What is knowable is the input side, which is what this page measured. You can confirm your site is readable, you can see what currently comes back for the questions your clients ask, and you can be present in those results. That is less satisfying than a lever, and it is what there is.

The useful test is time. Ask the assistants the questions your clients would ask, write down what comes back, and repeat it in three months. Three months is early in any law firm SEO timeline, so treat the first record as a baseline rather than a verdict. Your own record, on your own practice areas, will tell you more than any vendor’s claim. It also belongs in the same review as the rest of your law firm marketing budget, rather than in a category of its own.

Common Questions

One practical note on doing that. Answer the cost question with a real range and the reason it varies, rather than routing it to a call. A page that genuinely answers it will probably be the only one in your market that does. That is the whole of the opportunity here, and also why most firms will not take it.

Where these figures come from

The crawler figures come from reading the robots.txt file of law firm sites found ranking for ordinary legal queries across several practice areas and metros, in October 2026. Twenty domains were attempted and 15 returned a readable file; the rest timed out or refused the request, and are excluded rather than counted as permissive.

The results figures come from six lawyer-seeking searches run the same month, producing 56 results. Each was sorted by whether it was a firm site, a directory, a roundup or something else. These are web search results rather than assistant transcripts, which is stated plainly above because the distinction matters.

The observation that every guide recommends schema and none states a limit comes from reading eleven pages that rank for this question. Both counts are small samples taken on one day, and anyone can repeat either in an afternoon.

AttorneyPlot builds websites for law firms, and much of what is sold in this area is sold by companies like it. That is a reason to check the robots.txt claim yourself rather than take this page on trust.

1. OpenAI — GPTBot, OAI-SearchBot and ChatGPT-User
2. Anthropic — ClaudeBot and how to block it
3. Perplexity — PerplexityBot
4. Google — crawlers, including Google-Extended
5. Apple — Applebot and Applebot-Extended
6. Google Search Central — no one can guarantee a ranking
7. Google Search Central — creating helpful, people-first content
8. The Robots Exclusion Protocol
9. The llms.txt proposal
10. American Bar Association — 2023 Websites & Marketing TechReport
11. Clio — 2025 Legal Trends Report, methodology