AI search for law firms: how firms get named in AI answers

AI assistants don’t discover law firms. They repeat what the web already says about them, and for lawyers the web mostly speaks through directories, Business Profiles and reviews. So the honest version of “AI search” is unglamorous: make every source an engine reads accurate, complete and quotable, in each language your clients use, and then measure how often you’re named as a rate with a range. A single screenshot of your firm in ChatGPT is an anecdote, not a result.

What “AI search” means for a law firm

AI search is the set of answer engines people now ask instead of, or before, a search results page: Google’s AI Overviews and AI Mode, ChatGPT with search, Gemini, Microsoft Copilot, Perplexity and Claude. Each one writes an answer and sometimes names firms or cites pages. For a law firm, “AI search” work has one goal: that when someone asks a question your firm can answer, the engine’s sources say true, useful things about your firm, and the engine names you more often as a result.

The shift is real. Google’s AI answers appear on 23.6% of legal queries (Ahrefs), and people click a result on 8% of visits when a summary appears against 15% without one (Pew Research Center). In one 2026 consumer survey, 41.9% said they would use ChatGPT to research a lawyer, up from 28.1% the year before (vendor survey).

The traffic is still small. On one vendor’s law-firm sites, AI referrals were 0.47% of sessions (LawFuel). That is not a reason to ignore AI answers. It is a reason to measure them as influence on decisions, not only as clicks.

Which index each engine reads

This is the single most useful thing to understand, and most AI-visibility pitches skip it. An engine can only name what its sources contain, so a firm that is strong in Google can be missing from Copilot or Claude.

Two practical rules follow. First, Bing matters more than its search share suggests, because it feeds Copilot and part of ChatGPT. IndexNow lets a site tell Bing and other engines about new or changed pages the moment they publish (IndexNow). Second, decide your crawler policy on purpose. Many firm sites block AI crawlers through a security default nobody chose. Allowing the search crawlers and deciding separately about training-only crawlers is a reasonable middle path.

Who AI answers actually name

When people ask an AI assistant about lawyers, the answers lean on directories and aggregators, not on firm websites.

• In a study of 359 AI answers about injury lawyers, Super Lawyers was 33% of cited sources, Justia 20% and Reddit 8% (Citorian, secondary).
• In a 2026 benchmark, Avvo appeared in 45% of answers (benchmark).
• In an October 2026 benchmark across ChatGPT, Gemini, Perplexity and Grok, no individual firm made the top 20 cited sources (GlobeNewswire).
• For local questions in general, the Google Business Profile was the top source in BrightLocal’s study, at 28.63% (BrightLocal).

The lesson isn’t “buy directory listings”. It is that a firm usually reaches an AI answer through the sources the engine trusts. A complete, accurate profile on the directories that matter in your practice, the same facts on your Business Profile, and a site that confirms them, give an engine three agreeing sources instead of one.

What a firm controls

Most of what decides an AI mention sits in eight places. Work them in this order; the first three move fastest.

1. Your Google Business Profile. Correct primary category, every service in plain words, real hours, real photos, a description that states practice areas and languages. Local AI answers lean on it.
2. Directory profiles. Claim and complete the legal directories cited in your practice. Same name, address, phone, practice areas and languages as everywhere else.
3. Consistent facts. One fact sheet (names, offices, phones, hours, practice areas, languages, bar admissions) applied everywhere. Engines that see two phone numbers may print the wrong one.
4. Reviews, asked the right way. Ask every client the same way at set moments. Google bans review incentives and selective asking (Google); the FTC bans fake and bought reviews (FTC).
5. Bing Places and Apple’s business listing. They feed Copilot, ChatGPT and Apple Maps answers, and few firms keep them current.
6. Answer-first pages. A direct answer in the first two sentences under a question heading, a sourced and dated number, one-sentence definitions, a table where comparison helps. Engines quote pages that answer plainly.
7. Spanish pages written by native writers. Spanish questions surface Spanish sources. Machine-translated pages rarely earn the citation and can misstate the law.
8. Crawler access and indexing. Search Console, Bing Webmaster Tools, IndexNow and a deliberate robots.txt.

What a firm doesn’t control: how often a question is asked, how an engine weighs sources this month, and new data deals that change citations overnight. That is why measurement has to be repeated, not one-off.

Measuring it honestly: rates, ranges and runs

AI answers change from one run to the next (MediaPost). Ask the same question twice and you can get two different lists of firms. So the only meaningful measure is a mention rate: answers that name your firm divided by answers sampled, for one engine, practice, city and language, with a range that shows how much it could move by chance.

The range matters more than most reports admit. Here is what a “20% mention rate” really tells you at different sample sizes, using the Wilson method (method).

At 5 answers, a “20%” rate is compatible with almost anything from rarely named to named half the time. That is why a report built on one run per prompt can show a “doubling” that is pure chance. Our rule, and a good rule for anyone: claim movement only when two months’ ranges don’t overlap.

How to build a panel

1. Write 40 to 120 real questions across four kinds: finding a lawyer, deciding whether you need one, solving a problem, and checking a firm by name.
2. Write the Spanish versions fresh, with a native speaker, as people actually ask them. Don’t translate the English.
3. Fix the competitor list at the start: the same 5 to 10 firms every month, so share of mentions is comparable.
4. Run each prompt at least 5 times a month on each engine, logged out, from locations in your service area.
5. Record three things per answer: whether your firm is named, which competitors are named, and which sources are cited. Log any factual error about your firm.
6. Keep the panel fixed for a quarter. New questions join only at the start of a quarter.

Five checks for any AI-visibility report, ours included

1. How many runs sit behind each number?
2. Is there a range on each rate?
3. Are Spanish questions included, if your clients ask in Spanish?
4. Is the competitor list fixed from month to month?
5. Can you read the raw answers, with date and place?

A starter prompt library, by practice

Two warnings on the immigration rows. Never let any page or profile describe a non-lawyer as a “notario”: in many states that wording is restricted because it misleads Spanish speakers. And wrong AI answers in immigration can cost people their case, so log errors about your firm and fix them at the source quickly.

What doesn’t work, and what to avoid

• Special files as a shortcut. Google says its AI features need no special markup or files (Google). An llms.txt file is fine hygiene; it is not a ranking lever.
• Content written to game engines. Mass-produced pages stuffed with “best lawyer” phrasing are scaled content under Google’s spam policies, and the bar may see them as misleading advertising.
• Fake or gated reviews. Banned by the FTC and by Google. Asking only happy clients is gating.
• One score. A single “AI visibility score” hides the engine, the language, the sample size and the range. Ask for the rates behind it.
• Paying for placement. No engine sells a spot in its answers. Ads in ChatGPT exist and are labeled; they are a separate channel. See ads in ChatGPT.

Connecting AI answers to signed cases

A mention is a means. The result is a signed case. Three steps tie them together:

1. Track AI referrers. Visits from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com get their own line, followed to calls, forms and signed cases.
2. Ask at intake. Add “an AI assistant” to the “How did you hear about us?” list. People who read an answer and then search your name otherwise look like brand searches.
3. Report it under The Probative Standard: AI assistants are their own channel line in the Monday report, and no cost per signed case is shown until at least 5 cases sit behind it.

What this means for your firm

• Start with the sources, not the engines. Business Profile, directories, consistent facts and reviews come first; they feed every engine at once.
• Get into Bing. Bing Webmaster Tools and IndexNow cost nothing and reach Copilot and part of ChatGPT.
• Do Spanish properly if your clients speak it. The Spanish column is usually where the gap is widest.
• Measure with ranges. Ask for at least 5 runs per prompt per engine, a fixed competitor list and a 90% range, or treat the report as anecdote.
• Refuse guarantees. Nobody controls AI answers. Pay for work and honest measurement.

Probative is AI-first SEO and ads for law firms: firms get named in AI answers because of the work on their sources, measured monthly with ranges. See our AI search service, try the free AI check, and read how the same work plays by practice on the personal injury SEO and immigration SEO pages.

Marketing information, not legal advice. Your profile, pages and reviews are lawyer advertising under your state’s rules; check with your bar and your own counsel.