Short answer: GEO for law firms means making sure that when a prospective client asks ChatGPT, Perplexity, Gemini, Claude or Grok to recommend a lawyer for a specific matter in a specific city, your firm is one of the names in the answer, and the answer describes you correctly. In our Chicago sample of 20 legal questions asked to five assistants on 7 September 2026, every answer to a direct request for a lawyer named at least one firm, while questions about trust, credentials and comparison almost never did. The firms that were named got there by two routes: a practice-area page on their own site that the assistant cited, or directories and best-of lists that carried their name. This article goes through the data question by question and turns it into a playbook.
Our law firms industry page publishes the raw tables: who was named, by which assistant, and which domains were cited. Here we go further: why the numbers look the way they do, what they mean for a firm deciding where to spend the next six months, and the specific pages, profiles and checks that follow from them.
- Intent decides everything. All 33 completed answers to the seven “recommend a lawyer” questions named firms. Only 15 of the 65 answers to the other thirteen questions did.
- The city in the question matters as much as the intent. Questions that mentioned Chicago produced named firms in 40 of 48 answers; questions without a city did so in 8 of 50.
- There is no stable top three. 183 distinct providers were named. The most-named firm appeared in 7 of the 98 answers, a 7.1% share, and most names appeared in a single answer from a single assistant.
- Your practice-area page is the most direct lever. In at least four of every ten times a firm was named, the same answer cited a page on that firm’s own site, usually a matter-specific practice page, not the homepage.
- Directories are the second door. Avvo, Super Lawyers, Justia, FindLaw, Best Lawyers and “Top 10 lawyers in Chicago” list sites supplied roughly one citation in seven, and one firm was named by all five assistants without any page of its own being cited.
- The trust questions are an open slot. Asked how to verify a lawyer, the assistants cited bar authorities from California, Florida, Pennsylvania and elsewhere; the Illinois lawyer search was cited once.
What we measured, and what a single-city sample can tell you
The benchmark behind this article is a dated snapshot, and it is worth being precise about its shape before reading anything into it. We generated 20 questions that a prospective client might ask an AI assistant about hiring a lawyer. None of them contained a firm name, so every name in the answers was the assistant’s own choice. Each question was sent to five assistants: ChatGPT (gpt-5-mini), Perplexity (sonar), Gemini (gemini-3.5-flash), Claude (claude-haiku-4-5) and Grok (grok-4.20-non-reasoning), all with web search available. Chicago, Illinois was the sample city for questions that needed a location. The run completed on 7 September 2026.
That gives 100 question-and-assistant cells. 98 returned an answer; two Gemini cells failed, both on recommendation questions (a custody lawyer and an estate planning lawyer). We exclude failed cells from every percentage instead of scoring them as zero, because a timeout says nothing about whether a firm would have been named. A separate parser read the full text of each answer and extracted the providers it named, whether it explicitly recommended them, the form of the answer and every source cited inside the answer text. Retrieval candidates that a model looked at but did not cite are not counted.
What the sample can support
It can support statements about how these assistants handle this kind of question: which intents produce names, how much the assistants agree with each other, what kinds of sources they lean on and how those sources relate to the firms that get named. These patterns come from 98 answers and more than a thousand citations, and they are consistent enough across practice areas to act on.
What it cannot support
It cannot tell you that a particular Chicago firm “ranks first in AI”, and you should be suspicious of anyone who tells you that from a sample like this. Each named firm here appears in a handful of answers at most. AI answers change between runs, between days and between users, so a firm that appeared in four answers on 7 September might appear in two or six the next week. It also cannot tell you how a firm in Houston or Phoenix would fare: the citation landscape, the directories and the local competition differ by city, and the same 20 questions asked with a different sample city would surface a different set of firms and possibly a different source mix. Treat the specific names below as worked examples of mechanisms, not as a league table.
We name firms below only as the data shows them, with the assistant and the question. Being named in this sample is not an endorsement, and not being named says nothing about the quality of a firm’s legal work.
Finding 1: assistants name firms when the client asks for a lawyer, and almost never otherwise
The single most useful fact for anyone planning GEO for law firms is how sharply the answers split by intent. We tagged each question with the intent behind it: a direct request for a recommendation (“who would you recommend for…”), evaluation (“is it worth hiring a specialist…”), research (“what should I look for…”), comparison (“solo practitioner or large firm?”) and trust (“how do I verify a lawyer’s credentials?”).
Every completed answer to the seven recommendation questions named at least one firm, and most named several: the 33 answers contained more than 230 names between them. Across the other thirteen questions, 15 of 65 answers named anyone at all. The trust questions were the starkest: two of 20 answers named a provider, and in both cases the “provider” was a service (Super Lawyers once, CheckMyAttorney once), not a law firm. Every one of the 20 trust answers was a guide with no list of firms. Every answer to the questions about red flags, licensing and what to expect at a first meeting named nobody.
Why the assistants behave this way
A recommendation question carries an implicit instruction to produce names. A trust or research question carries the opposite: the safest, most helpful answer to “how do I verify a lawyer?” is a procedure that works for any lawyer, and naming firms would look like an advertisement. The assistants are not withholding names they know; the question does not call for them. That is good news for planning, because it means you can predict which questions are worth competing on for a mention and which are worth competing on for a citation.
The city changes the answer as much as the intent
Ten of the 20 questions mentioned Chicago. Answers to those named firms in 40 of 48 cases. The other ten questions were generic (“How do I choose between a retainer agreement and paying hourly?”), and answers to them named a provider in 8 of 50 cases. The effect holds even when the intent is not a recommendation. Three research and comparison questions included the city, and their answers named firms in 7 of 15 cases, against a handful in the generic set. A question such as “What should I look for when choosing a lawyer in Chicago for my small business incorporation?” was answered by four of five assistants with a checklist followed by a short list of local firms; ChatGPT, for example, listed five firms “you can call for initial consultations” after its checklist.
What this means for where you spend effort
The questions that produce names are narrow and specific: a matter type, a city and, often, urgency (“closing in two weeks”, “threatening eviction”, “first-time DUI”). Those are the questions to build your measurement set around and the ones your practice-area pages must answer. The general questions about choosing and vetting lawyers are not worthless. They produce no names, but they do produce citations, and the sources cited there shape what a prospective client believes good looks like before they ask for a recommendation. We come back to that opening in Finding 5.
| Question type (Chicago sample) | What the answer looked like | What a firm can win | Asset that wins it |
|---|---|---|---|
| Recommendation, with city and matter | Ranked or unranked list of firms; 33 of 33 answers named firms | A mention, ideally an explicit recommendation | Matter-and-city practice page, consistent directory profiles |
| Research or evaluation, with city | Checklist, then often a short local list; 7 of 15 answers named firms | A mention after the checklist, or a citation | Practice page with process and fees; local explainer |
| Comparison, generic | Pros and cons; names rare | A citation for the explanation | Honest comparison article (e.g. retainer vs hourly) with Illinois specifics |
| Trust and verification, generic | Step-by-step guide, no firms named in 20 of 20 | A citation, and the framing of what to check | Verification page linking to the state lawyer search and your own record |
Finding 2: the list is long, shallow and different in every assistant
If you imagine AI answers as a new version of the top three map results, the data corrects that quickly. Across 98 answers, 183 distinct providers were named. The most-named, Regas & Dallas P.C., appeared in 7 answers across two questions, a 7.1% share of all answers. Behind it, four providers appeared in 5 answers each, and a long tail followed: most names in the data appeared exactly once, in one answer from one assistant. Depending on how strictly you merge spelling variants (“The Toney Law Firm” and “The Toney Law Firm, LLC”), between roughly two-thirds and three-quarters of names were one-offs.
How much the assistants agree
For each question, we compared the sets of names that each pair of assistants produced, using the share of names they had in common. On a typical question, two assistants shared roughly one name in ten. The closest pair, ChatGPT and Claude, shared about one name in five. The most different pair, Perplexity and Grok, shared about one name in twenty. Even on the question where the assistants agreed most, the commercial litigation question, 33 distinct names came back, and 13 of them were named by more than one assistant.
Five providers were named by all five assistants: Regas & Dallas P.C., Gurney Law Group, Peck Ritchey, The Toney Law Firm and Rentervention, a tenant help service run by the Lawyers’ Committee for Better Housing. Each of the five owes its consensus to one or two questions, not to general prominence. Gurney Law Group appeared in exactly one question, the real estate closing, and was named there by every assistant.
Why agreement is so low
The assistants do not share an index. Each one runs its own web search with its own query rewriting, reads a different set of pages and decides for itself how many names to give. Perplexity cited 19.05 sources per answer on average and Gemini 5.56, so they are working from very different evidence. Answer style differs too: Gemini named providers in 11 of its 18 completed answers, Perplexity in 7 of 20. The answer form varies as well: on the recommendation questions, 14 answers were ranked lists and 19 were unranked lists, so “position” often does not exist.
What to measure instead of rank
Because the overlap is low and positions are often absent, a single “AI rank” for a law firm is not a meaningful number. Three measures hold up better:
- Mention rate per assistant and per practice area. Out of the recommendation questions for estate planning, how many answers from each assistant named you? Keep the assistants separate; averaging them hides that one may name you every time and another never.
- Recommendation rate. Being listed is not the same as being recommended. In the published table, Legal Aid Chicago and Loftus & Eisenberg were each named in 4 answers and explicitly recommended in none; Cook County Legal Aid for Housing and Debt was named in 4 and recommended in all 4.
- Source attribution. For each answer that names you, which page was cited: yours, a directory or nothing? That tells you which lever is actually working, and it is the subject of the next two findings.
Our guide to measuring share of voice in AI answers covers how to count competitors symmetrically so that these rates mean the same thing for every firm in the set.
Finding 3: the firm’s own practice-area page is the most common source
For each answer that named a firm, we checked whether the same answer cited a page on that firm’s own website. Matching names to domains is loose. Some firms use abbreviations, and we only counted a match when the firm’s distinctive name appeared in the domain. Even with that conservative rule, in at least four of every ten times a firm was named, its own site was among the sources. On the recommendation questions, the ratio was the same.
The pages cited were rarely homepages. They were matter-specific practice pages with the city in the title, written to answer the question a client would ask. Three examples from the data show the pattern.
Gurney Law Group and the real estate closing question
The question was “I need help with a real estate transaction closing in two weeks near downtown Chicago. Who would you recommend?” All five assistants named Gurney Law Group, and three of them recommended it explicitly. Its site was cited 10 times across those answers: the homepage, titled “Real Estate Attorneys”, five times, a page titled “Real Estate Closing Attorney Chicago” four times, and a page aimed at realtors once. The firm’s pages say what the question asks, in the words the question uses.
Regas & Dallas and the business questions
Regas & Dallas P.C. was the most-named provider in the sample: 7 answers, all from two questions, one about choosing a lawyer for a small business incorporation and one asking for commercial litigation specialists. Its domain, rfd-law.com, does not contain the name the assistants used, which is why an automatic name-to-domain match misses it, and why the benchmark’s automatic domain classifier filed it under “other” instead of “brand site”. Its pages were cited 8 times: “Incorporation Lawyers in Chicago”, “Commercial Litigation Attorney in Chicago”, “Small Business Attorney Chicago” and the homepage. The domain classifier missing it is a useful warning for your own measurement. If your firm’s domain is an abbreviation, a tool that matches on name will under-report how often your site is the source.
Combs Waterkotte and the first-time DUI question
Three assistants named Combs Waterkotte for “Which criminal defense lawyers in Chicago would be best for a first-time DUI charge?” Its page titled “First DUI Lawyer Chicago, Illinois” was cited 4 times, and a second DUI page twice. One more citation went to a chapter of an online guide on the firm’s site about how to choose a criminal defense lawyer. That is a firm whose content answers both the recommendation question and the research question around it.
What these pages have in common
None of this requires a special format for AI. The cited pages share plain characteristics. One matter per page. The city in the title. A first paragraph that says who the service is for. A description of what happens and roughly what it involves. Enough text for an assistant to quote without inventing anything. The alternative most firms run is a single “Practice Areas” page listing twelve matter types in one sentence each. That page gives an assistant nothing to cite for “first-time DUI in Chicago” and nothing to distinguish you from the next firm.
| Element | Typical “Practice Areas” page | Matter-and-city page that gets cited |
|---|---|---|
| Title | Practice Areas | Firm Name | First DUI Lawyer, Chicago, Illinois | Firm Name |
| Opening | “We offer a full range of legal services to individuals and businesses.” | Who this is for, what the first steps are, what the first consultation covers, in three or four sentences |
| Process | Absent | Numbered steps from arrest or engagement to resolution, with the local court or agency named |
| Fees | “Contact us” | How fees are structured (flat, hourly, retainer) and what changes them; ranges only where your rules allow |
| Credentials | “Experienced attorneys” | Named attorney, year and state of admission, a link to the state lawyer search |
| Local specifics | None | The county, the court, the deadlines and terms a local client will hear |
| Questions | None, or generic | Five to eight real questions clients ask, each answered in a self-contained paragraph |
If you want to see how passages get lifted into answers, our article on extractable writing covers sentence-level structure: self-contained claims, no pronouns pointing three paragraphs back, numbers next to the thing they measure.
Finding 4: directories and best-of lists are the second door
The other route to being named runs through third parties. Look at the most-cited domains across all 1,005 citations and the pattern is clear: directories, ratings sites and “Top 10 lawyers in Chicago” pages sit alongside the American Bar Association and Reddit at the top.
Adding up the lawyer directories, peer-rating lists and best-of list sites we could identify (Avvo, both Super Lawyers domains, Justia, FindLaw, Best Lawyers, Lawyers.com, LawFirmSquare, OnToplist, LawyerLegion, 101Attorney, Expertise.com, Legal 500, Chambers and a few one-offs), they supplied roughly one citation in seven. On the recommendation questions they matter more. Avvo’s “Top Rated Chicago, IL DUI Lawyers” page was cited 4 times on the DUI question, and LawFirmSquare’s “Top 10 Contract Lawyers in Chicago (2026)” 5 times on the contract review rate question.
The firm that was named without its own site
On the DUI question, all five assistants named The Toney Law Firm. None of those answers cited a page on the firm’s own site. What they cited were ranking and directory pages: Avvo, OnToplist (“Top Chicago DUI Lawyers”, 4 citations across the answers), Super Lawyers, LawFirmSquare and a site called chicagodefensefirm.com. The name reached the answer through other people’s lists. That works, but it means the description of the firm in the answer comes from those lists, and if a list is out of date, wrong about a practice area or simply thin, the assistant repeats it.
Each assistant leans on different directories
The directory mix is not uniform. Perplexity accounted for 9 of the 11 Justia citations, 11 of the 19 citations of attorneys.superlawyers.com and 8 of the 12 Best Lawyers citations; Gemini and Claude cited attorneys.superlawyers.com zero times. Avvo was the most evenly spread, cited by all five assistants. Martindale, long the reference directory for the profession, was cited once. If you only maintain one directory profile, the data argues for Avvo; if you want coverage in Perplexity specifically, Justia, Super Lawyers and Best Lawyers matter more.
Perplexity also stands apart on community sources: 13 of the 17 Reddit citations came from it, on questions such as red flags when hiring a lawyer and large firm versus solo practitioner. ChatGPT cited Reddit zero times in this run.
| Assistant (model) | Answers naming ≥1 provider | Providers per answer | Citations per answer | Answers pointing to directories | Notable source habit |
|---|---|---|---|---|---|
| ChatGPT (gpt-5-mini) | 9 of 20 | 3.45 | 10.25 | 5 | Cited checkmyattorney.com 6 times; no Reddit |
| Perplexity (sonar) | 7 of 20 | 2.9 | 19.05 | 1 | Most citations by far; most Reddit, Justia and Super Lawyers |
| Gemini (gemini-3.5-flash) | 11 of 18 | 3.06 | 5.56 | 2 | Named firms even on generic comparison questions |
| Claude (claude-haiku-4-5) | 11 of 20 | 2.35 | 6.05 | 6 | Fewest names per answer; often pointed to directories in text |
| Grok (grok-4.20-non-reasoning) | 10 of 20 | 2.7 | 9.9 | 7 | Most likely to tell the reader to check directories |
What to do about directories
The directory lever is less about being listed than about the listing being right. An assistant reading an Avvo or Super Lawyers page will repeat the practice areas, the office location and the description it finds. So the work is an audit. For each directory that appears in answers for your practice areas, check that the firm name matches the name on your site and your Google Business Profile. Check the address, the phone number and the practice areas, and check that the profile links to the matching practice page on your site. For best-of list sites, be careful. Some are editorial, some are pay-to-appear, and you should know which is which before paying for any of them. Our guide to third-party corroboration goes into how assistants weigh several independent sources that agree.
Finding 5: other states, legal aid and online services fill the gaps
Generic legal questions get answered from other states
The trust questions produced no firm names, but their citations reveal something a Chicago firm can use. Asked “How do I verify a lawyer’s credentials and check for disciplinary history?”, the five assistants cited, between them, the bar or discipline sites of California, Florida, Pennsylvania, Alabama, Arizona, Nevada and North Carolina, the U.S. Department of Justice and a site about lawyers in the Philippines (cited 5 times on that question). The Illinois Attorney Registration and Disciplinary Commission’s lawyer search was cited exactly once, by Perplexity. The licensing question looked similar: California, Florida, Kansas, Texas and Ontario sources, plus Wikipedia.
The same happened on the other generic questions. Answers about what to expect at a first meeting cited sources from New South Wales and the UK; answers about whether case-specific experience matters cited Australian, Canadian and British pages. None of those questions mentioned a city, so the assistants had no reason to localize. But a client in Chicago who asks them gets an answer built from other jurisdictions’ rules.
The opening for a local firm
This is the one place in the data where a firm can win citations without competing for a list position. Answering “how do I verify a lawyer in Illinois” with the actual Illinois procedure is useful, specific and currently thin in the sources the assistants find. The procedure covers the ARDC lawyer search, what “authorized to practice” means there, where public discipline appears and how to read it. Similar pages work for “what to bring to a first meeting with a Chicago family lawyer” or “retainer versus hourly fees in Illinois”. These do not produce a mention for your firm, because the answer will not name anyone, but they make your site a cited source, and they establish the firm as the one that explains the local rules.
How to write the local explainer
- Put the jurisdiction in the title and the first sentence, so the page is retrieved for local variants of the question.
- Link to the primary authority (for Illinois, the ARDC lawyer search) and describe what the reader will see there, instead of paraphrasing rules you have not checked.
- Answer the national version of the question first in two or three sentences, then the Illinois specifics. Assistants often lift the general part; you want that part to be accurate too.
- Date the page and say when it was last reviewed. Procedures and fee rules change, and an assistant cannot tell a 2019 page from a current one unless you say so.
- Do not add a sales section. A page that turns into a pitch halfway through is less likely to be cited for a neutral procedural question.
Legal aid and government pages on urgent matters
Not every name in a legal answer is a law firm, and two questions show who else competes for the slots. The eviction question (“My landlord is threatening eviction. I need legal help fast in Chicago. Who’s the best option?”) was the only recommendation question where the leading names were not firms at all. Rentervention was named by all five assistants and recommended by three. Cook County Legal Aid for Housing and Debt was named by four and recommended by all four, followed by Eviction Help Illinois, Legal Aid Chicago and Uptown People’s Law Office. The sources were dominated by government and nonprofit pages: chicago.gov was cited 12 times, all on this question, tenants-rights.org, the Metropolitan Tenants Organization’s site, 10 times, plus legalaidchicago.org, hud.gov and Cook County pages. For urgent consumer matters where free help exists, the assistants route people to it first. A private firm that handles tenant cases is not going to displace legal aid in that answer, and should not try. It can be the name given as the paid alternative, but only if its page states plainly who it represents and what it costs compared with free services.
Online services on simple documents
The online wills question (“Are online legal services better than traditional lawyers for simple documents like wills?”) produced product names: Trust & Will from four assistants, FreeWill from three and Quicken WillMaker. An estate planning firm that wants to be part of that answer needs an honest page about when an online will is enough and when it is not. The assistants are already framing the question that way, and a law firm’s own page on the topic (yolofskylaw.com) was among the sources cited on that question, 3 times.
The GEO for law firms playbook, in the order to do it
The findings point to a sequence. It is deliberately ordered by what moves first: measurement, then the pages you control, then the profiles you can edit, then the content that wins citations, then the loop.
Step 1: build a question set per practice area
For each practice area you want clients from, write five to eight questions, phrased the way a client would type them into an assistant. Use plain language, include the city and add the circumstance that makes it urgent. At least one per area should be a direct recommendation (“who would you recommend for…”), because only those produce names. Add one cost question, one comparison and one verification question for the citation side. Do not put your firm’s name in any of them: a question that names you measures nothing, because the assistant will talk about you regardless. Our guide to building a prompt set covers phrasing, intent balance and how many questions you need.
Step 2: measure a baseline across all five assistants
Run every question in every assistant on the same day, with the same location. For each answer, record four things: whether you were named, whether you were recommended, which competitors were named and which sources were cited. Record failures as failures. Two of our 100 cells failed; if we had scored them as “not named”, Gemini would look worse than it is on two of the most valuable questions. Repeat each question at least twice if you can, because a single run understates how variable these answers are.
Step 3: rebuild the practice-area pages you are losing on
Take the recommendation questions where competitors are named and you are not. For each, check whether you have a page that answers exactly that question for that city. If you have only a general practice page, write the specific one, using the structure in the page anatomy figure above. Keep it factual and checkable. If you already have the page, compare it with the competitor pages the answer cited. Look at the title, the opening, the specificity of the process and fees, and the credentials. In our data, the pages that won were not long and not clever; they were specific.
Step 4: make every directory profile tell the same story
List the directories cited in answers for your practice areas. In this sample that means Avvo, Super Lawyers, Justia, FindLaw and Best Lawyers, plus whichever best-of list sites showed up in your city. For each, verify name, address, phone, practice areas, attorney bios and the link back to the matching practice page. Inconsistent names are a particular problem for firms that have changed partners or merged: if one directory still lists the old name, the assistant may name the old firm, or split your mentions between two entities.
Step 5: publish the local explainers the national pages cannot
Pick the two or three trust and process questions closest to your practice areas, and write the Illinois version: how to verify a lawyer here, what a first meeting with a family lawyer in Cook County involves, how retainers work under the rules here. These pages are for citations, not mentions, and they compound: each one is a page an assistant can cite when the question is asked with a location.
Step 6: technical basics that stop you from being invisible
- Check that the AI search crawlers can fetch your practice pages. A firewall or CDN rule that blocks unknown bots will block them too, and your robots.txt should not exclude the search agents (OAI-SearchBot, PerplexityBot, Claude-SearchBot) even if you decide to exclude training crawlers.
- Make sure the practice pages render their text in the HTML, not only after JavaScript runs. Several AI fetchers do not execute scripts.
- Add LegalService or Attorney structured data with the name, address and telephone exactly as in your directory profiles, and sameAs links to those profiles. It will not get you named on its own; it removes ambiguity about which entity a page belongs to.
Step 7: re-measure monthly and read the deltas per assistant
Re-run the same question set monthly. Compare each assistant with itself, not the average with the average. When a page goes live, look for it in the citations before looking for your name in the answer: citation usually comes first. When an assistant drops you, look at which sources it cited instead; that tells you whether a competitor page or a directory change is responsible.
Ethics and compliance guardrails in GEO for law firms
Law firms work under rules that most GEO advice ignores, and some common tactics are a professional-conduct problem, not just a quality problem.
Communications about your services
The ABA Model Rules of Professional Conduct, which states have adopted in their own versions, prohibit false or misleading communications about a lawyer’s services and restrict claims to be a certified specialist. Those rules do not stop applying because the reader is a language model. A practice page that says “Chicago’s top DUI lawyer” or implies results the firm cannot substantiate is a problem when a person reads it, and it is the passage most likely to be quoted back to a client by an assistant. Illinois has its own rules; check them, or ask your ethics counsel, before publishing comparative or specialist language.
Reviews and testimonials
Directory reviews matter in this data because directories are cited. That creates a temptation to manufacture them. The FTC’s final rule on fake reviews and testimonials, announced on August 14, 2024, bans fake reviews and reviews that misstate the writer’s experience. It also bans compensation conditioned on a positive or negative review, undisclosed reviews from insiders, and suppressing negative reviews through baseless legal threats or intimidation, and it lets the FTC seek civil penalties against knowing violators. Ask clients for reviews in a neutral way, never pay for them, and never ask staff or relatives to write them.
Confidentiality in replies and case examples
Replying to a negative review with details of the client’s matter can breach client confidentiality, even if the client discussed the matter publicly first. The same applies to case studies on practice pages: use only matters where the client has given informed consent, or that are already public record, and describe outcomes without implying they predict future results.
Paying for placement
Some best-of list sites charge for inclusion. Paying for a listing can be permissible advertising, but presenting a paid placement as an independent ranking is not something a firm should repeat on its own site (“As featured in the Top 10…”) without knowing how the list was compiled. If an assistant cites a pay-to-play list as evidence, that is the list’s problem; if your site repeats it as an award, it becomes yours.
How to measure GEO for law firms without fooling yourself
The sample above also shows the traps in measuring this. Each of these would have produced a wrong conclusion from the same data.
Trap 1: putting your firm’s name in the question
“Is Firm X good for DUI cases in Chicago?” will mention Firm X every time. Questions that name you inflate your mention rate to near 100% and say nothing about whether a client who does not know you will hear of you. All 20 questions in this benchmark were generated without any provider name for exactly this reason.
Trap 2: counting your own mentions more generously than competitors’
If your tracking matches your firm under five spellings and competitors under one, you will appear to beat them. Use the same normalization for every name. The rfd-law.com example cuts the other way: a matcher that relies on the domain will miss a firm whose domain is an abbreviation, so check by hand for your own firm and your closest competitors.
Trap 3: averaging the assistants
Perplexity cited more than three times as many sources per answer as Gemini, and the two leaned on different directories. An average across assistants describes no assistant. Report each separately, and weigh them by how your clients actually use them if you know that. BrightLocal’s 2026 consumer survey found 45% of US consumers using AI tools for local business recommendations, up from 6% the year before, and 31% had used ChatGPT for business recommendations in the past 12 months. That makes ChatGPT a reasonable first assistant to optimize for, but not the only one.
Trap 4: treating failures as zeros
Gemini failed on the custody and estate planning questions in this run. Scoring those as “not named” would put Gemini at zero for two practice areas when it simply did not answer. Keep failed cells out of the denominator and re-run them.
Trap 5: reading one run as a trend
With most names appearing once, a single run will show firms appearing and disappearing between months for no reason other than sampling. Before you conclude that a new practice page worked, look for it in the citations, check it held across at least two runs, and check that it did not coincide with a change in the question set.
What a prospective client does after the answer
Measuring the answer is not the end of the path. In Yext’s 2026 survey of people who search for local businesses, only 5% said they move directly from an AI answer to a purchase. After an AI recommendation, 53% search Google or Bing to verify, 49% visit the business’s website, 42% click through to the sources the AI cited and 28% look for reviews. For a law firm, that means the practice page that got you cited is often also the page the client lands on next, and the directory profile that carried your name is the one they check. Both need to hold up under a second look. Our guide to local GEO covers the Google Business Profile side of that path, where local results depend mainly on relevance, distance and prominence, and where prominence includes links and reviews.
Sources
- InsightWonder industry benchmark, law firms, Chicago, Illinois sample, collected 7 September 2026: /industries/law-firms/
- Attorney Registration and Disciplinary Commission of the Supreme Court of Illinois, Lawyer Search: https://www.iardc.org/Lawyer/Search
- American Bar Association, Model Rules of Professional Conduct: https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/
- U.S. Federal Trade Commission, final rule banning fake reviews and testimonials (14 August 2024): https://www.ftc.gov/news-events/news/press-releases/2024/08/federal-trade-commission-announces-final-rule-banning-fake-reviews-testimonials
- BrightLocal, Local Consumer Review Survey 2026, AI and trust: https://www.brightlocal.com/research/lcrs-ai-trust/
- Yext, Consumer Search Behaviors Report: https://www.yext.com/resources/consumer-search-behaviors-full-report
- Google Business Profile Help, how local ranking works: https://support.google.com/business/answer/7091
- Schema.org, LegalService type: https://schema.org/LegalService