AI search for law firms: get named in AI answers, measured honestly

AI search at Probative measures how often AI answers name your firm, in English and Spanish, and works on what those answers read, so your firm can be named more often. We report the rate every month with an honest error range, because AI answers change from one run to the next.

Why it matters now

• Google’s AI answers appear on 23.6% of legal searches, and they cut clicks on the top result by about 58% (Ahrefs; Ahrefs).
• In Pew’s data, people clicked a result on 8% of visits when an AI summary appeared, 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% (vendor survey).

Fewer people click through to a website. So being named in the answer itself matters, not only being the link underneath it.

What AI answers read

Each engine draws on different sources, so a firm that looks strong in one can be missing from another.

Directories carry weight everywhere. In a 2026 benchmark of AI answers about lawyers, Avvo appeared in 45% of answers (benchmark). An incomplete or wrong directory profile can cost a firm a mention, or put a wrong phone number into an answer.

How we measure: the prompt panel

We measure the way a careful survey would, not with one screenshot.

1. The panel. We write 40 to 120 questions people actually ask, across four kinds: finding a lawyer (“truck accident lawyer in [city]”), deciding whether they need one (“do I need a lawyer after a minor crash in [state]”), solving a problem (“what to do after a car accident in [city]”), and checking a firm (“[your firm] reviews”). Each has a Spanish version where you serve Spanish speakers, written by a native writer, not machine-translated.
2. The runs. Each prompt runs at least 5 times a month on each engine, from points in your service area, logged out, so personal history doesn’t color the answer.
3. The rate, with its range. For each engine, practice, city and language, we count the answers that name your firm and divide by the answers sampled. Each rate carries a 90% range, computed with the Wilson method (method). We say a rate has moved only when two months’ ranges don’t overlap.
4. Against a fixed list. Your share of mentions is measured against the same named competitors every month, chosen with you at the start.
5. Sources and errors. We record which sites each answer cites, and log every factual error about your firm: a wrong phone number, a wrong practice area, an old address, a claim that isn’t true.

ILLUSTRATIVE. A fictional firm, to show how we report. In month 1, the panel’s Spanish injury prompts on one engine named the firm in 14 of 60 answers: 23%, with a 90% range of 16% to 33%. In month 3, it was named in 23 of 60: 38%, with a range of 29% to 49%. The ranges still overlap, so the report says “higher, but not yet a proven change,” and the work continues.

That example is the point. A jump from 23% to 38% looks like a win, and many reports would call it one. At 60 answers, chance alone could explain much of it. We’d rather tell you that than sell you a screenshot. AI answers are known to vary from run to run (MediaPost).

What a panel looks like

ILLUSTRATIVE. Ten of the questions a panel for a fictional injury firm might hold. A real panel holds 40 to 120, chosen with you.

Each question runs at least 5 times a month on each engine, from points in your service area. The Spanish questions are written by a native writer as people actually ask them, not translated from the English ones.

What we need from you

• Your firm’s facts, confirmed by you: names, offices, phones, hours, practice areas, languages your team really speaks, and anything an AI answer should never say about you.
• Your cities and your competitors. The places you want to be named in, and the firms you compete with, so the comparison is fixed from the start.
• Manager access to your Business Profile and the listings we’ll fix, granted by you and removable by you.
• A quarterly half hour to review the panel and the fix list with us.

What the monthly report shows

The report fits on two pages. Page one is the table below, for each practice and city you choose. Page two lists the fixes shipped, the fixes next, and every factual error we found.

ILLUSTRATIVE. Invented numbers for a fictional injury firm, to show the layout only.

The Spanish column is usually where the gap is largest, and where a fix list can move fastest, because fewer firms have done the work in Spanish.

Why firms go missing from AI answers

These are the causes we find most often, each with its fix.

The fix list

Measurement finds the gaps. The fix list closes them, in order of likely effect.

• Your Google Business Profile. Correct categories, services, hours, practice descriptions in both languages, photos, and a steady, honest review process.
• Reviews, done by the rules. We help your firm ask every client the same way, at set moments. Google bans offering incentives for reviews and selectively asking happy clients (Google), and the FTC bans fake reviews and testimonials (FTC). Some states add their own testimonial rules; your attorney checks them.
• Directories and listings. Accurate, consistent profiles on the directories AI answers cite, plus Bing Places, Apple Business Connect and Yelp.
• Answer-first pages. Pages that answer the panel’s questions plainly, in English and Spanish, with the facts an engine can quote: practice areas, languages spoken, fees explained, office locations. See law firm SEO.
• Crawler access. Your site lets the search and AI crawlers in, unless you choose otherwise.
• Error corrections. When an answer gets a fact wrong, we trace where it came from and fix it at the source.

How it connects to signed cases

A mention is a means, not the result. We tie AI search to the one number that matters:

• AI assistants get their own line in the Monday report. Visits referred by ChatGPT, Perplexity, Gemini or Copilot are tracked to calls, forms and signed cases.
• Intake asks. “How did you hear about us?” includes “an AI assistant,” so people who read an answer and then searched your name still get counted.
• Honest about size. On one vendor’s law-firm sites, AI referrals were 0.47% of visits (LawFuel). Small now, and growing; we report it as it is.

A month of AI search work

Five checks for any AI-visibility report

Many agencies now sell “AI visibility.” Use these checks on any report, ours included:

1. How many runs sit behind each number? One run per prompt is a screenshot, not a measurement.
2. Is there a range? A rate without a range can’t tell you whether a change is real.
3. Are Spanish questions included? If your clients ask in Spanish, an English-only panel misses them.
4. Is the competitor list fixed? Changing the comparison set each month can make any trend look good.
5. Can you see the raw answers? You should be able to read what each engine actually said, with the date and place it was asked.

The honest limits

We put these in writing at the start, and in the first report:

• Answers don’t repeat exactly. They vary by run, by user and by place. That’s why we report rates and ranges.
• The panel is a sample. No engine publishes how often a question is asked, so the panel stands in for demand. We choose it with you and keep it fixed, so months compare fairly.
• Sources shift. When an engine signs a new data deal or changes how it searches, the sources it cites can change overnight. The report dates every change we see.
• No placement for sale. No engine sells a spot in its answers. Paid ads in ChatGPT exist, and they are a separate, clearly labeled channel: see ads in ChatGPT.

Ethics notes

• What an AI engine repeats about your firm should be true, and much of it comes from your own pages and profiles. We keep claims inside your state’s advertising rules: no superlatives, and no “expert” or “specialist” unless your state allows the exact wording and it is true.
• We never write fake reviews, post as clients, or ask only happy clients. Review requests go to every client, the same way.
• No client or prospect information goes into any AI tool for this work. The prompts are public questions.

How it changes by practice

• Personal injury: most questions name a crash type and a city, and the Spanish answers can name different firms than the English ones, so we measure both.
• Criminal defense: many questions are urgent and asked by a relative, such as how to get someone out of jail.
• Family law: decision questions about custody and costs come before any search for a firm.
• Estate planning: long, careful questions about wills, trusts and probate, often by adult children for a parent.
• Employment law: questions about rights and deadlines, in English and Spanish.
• Lemon law: questions naming a make, model and defect.
• Immigration: questions in many languages, where wrong answers can cost people their case.

Want to see a sample first? The free AI check runs a small, labeled version of the panel for your firm.