How lawyers can stay ahead of AI: what 12 AI assistants said in 6,227 answers about lawyers

Summary

Key findings

1. AI rarely agrees on who the good lawyers are. Across 200 legal markets, the 12 assistants named 9,795 different law firms. 73% of them were named by only one assistant. Just 8% were named by four or more, but those firms took 31% of all picks.
2. A quarter of answers name no lawyer. 24% of answers about lawyers named no one. Grok named no lawyer in 87% of its answers, DeepSeek in 63% and Alexa in 44%. ChatGPT, Claude, Gemini and Mistral named one almost every time.
3. When AI names no one, it sends people to the bar and Avvo. Of the 1,492 answers that named no lawyer, 83% pointed to the state bar and 80% to Avvo. Across all answers, Avvo came up in 48% and the state bar in 47%.
4. AI describes lawyers by what they handle. 28% of picks came with a specialty, 22% with reviews, 17% with trial experience, 13% with years in practice and 12% with a peer rating such as Super Lawyers, Best Lawyers or Martindale-Hubbell.
5. The firms AI agrees on have third-party proof. For firms named by four or more assistants, 15% of picks cited a peer rating and 12% a verdict, settlement or dismissal, against 9% and 8% for firms only one assistant named. Reviews were mentioned about equally (22% and 20%).
6. Each practice area has its own pitch. Personal injury: results (18% of picks) and "no fee unless we win" (17%). Criminal defense: a former prosecutor (19%) and trial experience (24%). Business litigation: peer ratings (16%). Speeding tickets: flat fees (18%) and star ratings (12%).
7. "Affordable" turns the answer into a fee conversation. For "affordable" questions, 26% of picks came with a free consultation and 21% with a flat fee or payment plan. Every "affordable" personal injury answer explained contingency fees, and 81% of "affordable" criminal defense answers pointed to a public defender, legal aid or a referral service.
8. Web-search AI reads directories as much as firm websites. Of the sources the web-search and voice assistants cited, 47% were law-firm websites and 38% were directories and rankings, led by Justia, Avvo, Super Lawyers, FindLaw and LawInfo. Perplexity's citations were 83% directories.

What it means. Being named by one AI assistant is not the same as being named by AI. The firms that several assistants agree on are the ones with proof that lives outside their own website: peer ratings, published results, reviews, and complete directory and bar profiles. Those are things a firm can build. The last section turns the findings into a plan.

Scope. The October 2026 edition of AI Syndicate Local: 200 legal markets (50 metro areas in 9 states, times 4 practice areas: personal injury, criminal defense, business litigation and speeding tickets), three fixed questions to each of 12 assistants, asked between 1 and 7 October 2026. This paper reads all 6,227 answers about lawyers that came back and the 20,276 lawyer recommendations in them. Method and limits are at the end.

What did we read, and how?

Each month AI Syndicate Local asks 12 AI assistants three questions about a type of business in a metro area: who are the best, who do you recommend, and a good affordable one nearby. For lawyers, a question reads like "I need a criminal defense lawyer in Fort Smith, AR. Who do you recommend?" In October 2026 we asked about four practice areas in 50 metro areas across Alabama, Alaska, Arizona, Arkansas, California, Florida, Georgia, New York and Texas.

6,227 answers came back. This paper reads what the assistants said: whether they named a lawyer, which firms they named, the reasons they gave for each one, the advice they added and the sites they pointed to. We call each firm an assistant names a pick and take the words that describe it, from its name to the next firm named. We sorted those words into kinds of reason:

The twelve fall into three groups by how they answer:

• Web search: ChatGPT, Claude, Gemini and Perplexity answer with web search on. Google AI Overviews is the AI Overview Google shows for the question. Microsoft Copilot is the Copilot answer Bing shows; when Bing shows none, we reconstruct one from the same Bing results.
• Model memory: Grok, Meta AI, DeepSeek and Mistral answer from what the model learned, because no consumer live-search access is available for them.
• Voice: Siri and AAlexa cannot be asked programmatically, so we reconstruct their answers in the short, spoken style each uses. A typical Siri answer about a lawyer is under 50 words and names one firm.

A reason in this paper is one the assistant wrote, not proof of why it picked the firm. The limits are set out at the end.

Will AI name a lawyer at all?

Not always. 24% of the answers about lawyers named no lawyer. Some assistants explained that they could not recommend a specific attorney; others gave a guide to finding one instead. Whether a firm can be named at all depends mostly on which assistant the client uses.

Assistants that answer from memory named no lawyer in 43% of answers, the voice assistants in 22% and the web-search assistants in 8%. Grok declined most in speeding-ticket answers (98%) and least in business litigation (75%). Perplexity and Microsoft Copilot held back most on criminal defense (30% and 27%).

The question matters too. Asked for "the best" lawyers, the assistants named no one in 16% of answers; asked "who do you recommend?" or for an affordable lawyer, in 28–29%. Recommending a person seems to be the line some assistants will not cross.

When an answer names no lawyer, it still sends the client somewhere. Of the 1,492 answers that named no one, 83% pointed to the state bar (usually its lawyer referral service or its licence lookup), 80% to Avvo, 54% to Martindale-Hubbell, 42% to Google and 38% to Super Lawyers. For a firm that an assistant will not name, those profiles are the answer.

Do the AI assistants agree on which lawyers to name?

Very little. Across the 200 legal markets, the 12 assistants named 9,795 different firms, a median of 47 per market. 73% were named by only one assistant. 15% were named by two or three, and 8% by four or more.

Those few firms carry most of the weight. The 8% named by four or more assistants received 31% of all picks. In 195 of the 200 markets at least one firm was named by four or more assistants, so a short consensus list exists almost everywhere. Below it, each assistant has its own long tail of names.

Firms named per practice area, 50 metro areas each, October 2026.

Criminal defense and speeding tickets are the most scattered: many small practices, and little agreement on which of them to name. Personal injury and business litigation have more firms that several assistants know.

Heavy advertising does not buy the list. Morgan & Morgan, a national personal injury firm, was named in 20 of the 50 personal injury metros, but received 1.9% of personal injury picks. In speeding tickets, firms with many offices did better: The Ticket Clinic was named in markets across seven states, most of all in Florida and California, and Off the Record, an online ticket service, was named in 30 of the 50 speeding-ticket metros, mostly by Gemini.

What does AI say about the lawyers it recommends?

Usually. 72% of picks came with at least one reason. The rest were a name, sometimes with a phone number or address, and nothing about why.

Web-search assistants lean on evidence they can read: reviews (26% of picks against 14% for memory assistants), years in practice (18% against 7%) and peer ratings (16% against 6%). Memory assistants describe what a firm handles: specialty comes up for 47% of their picks, against 18% for web search. Mistral gave a specialty for 81% of its picks.

Each assistant has its habits. ChatGPT cites peer ratings (27% of its picks), board certification (13%) and gives a phone number for 55%. Claude leans on reviews (44%) and quotes a star rating for 20%. Gemini talks about trial experience (40%) and local knowledge (32%). Perplexity writes short answers, about 120 words, and cites peer ratings for 23% of its picks.

Each practice area is described differently.

Share of picks in each practice area mentioning each reason, 12 assistants, October 2026. Specialty is 23–32% in all four.

The pattern follows what clients worry about. A personal injury client is told about results and how the fee works. A criminal defendant is told whether the lawyer used to be a prosecutor. A business is told about peer recognition. A driver with a ticket is told about price and reviews.

What do the lawyers most assistants agree on have in common?

We compared what the assistants wrote about the 754 firms named by four or more assistants with what they wrote about the 7,190 firms only one assistant named. To keep the comparison fair, we compare single picks, not firms, so a firm named more often does not get more chances to collect a reason.

The firms several assistants agree on are described with third-party proof: a peer rating (1.7 times as often), a verdict, settlement or dismissal (1.6 times), trial experience (1.5 times) and years in practice (1.4 times). Reviews come up about equally for both groups. So does a free consultation, and the firms only one assistant names are slightly more often given a star rating.

Two readings fit. Firms with peer ratings, published results and long track records leave more evidence across the web, so more assistants find them. Or, once several sources describe a firm the same way, every assistant repeats that description. Either way, the reasons that separate the consensus firms are ones that live outside the firm's own marketing: rankings, case results reported elsewhere, years of public record.

Consensus firms are also known across both kinds of AI. 72% of the firms named by four or more assistants were named by at least one web-search assistant and at least one memory assistant. Only a firm that is both easy to find today and already written about widely reaches both.

What changes when someone asks for an affordable lawyer?

The "affordable" question changes the answer more than anything else we measured. The assistants stop talking about rankings and start talking about how a lawyer is paid.

Asked for "the best", AI cites peer ratings for 18% of picks. Asked for "affordable", it cites them for 3%, and free consultations (26%) and flat fees (21%) take over. Trial experience drops from 21% to 11%.

The fee advice is specific to the practice area:

Share of answers to the "affordable" question, 12 assistants, October 2026.

For a firm, this means a client who asks AI for an affordable lawyer arrives already knowing how fees in that practice usually work, and often with a price in mind. A firm that publishes how it charges gives the assistant something to say about it. A firm that does not leaves the assistant to describe the market in general, or the competitor that did publish.

Which sites does AI send people to, and which does it cite?

Two different lists matter: the sites an answer tells the client to check, and the sources a web-search answer cites.

Sites named in the answer. Lawyers have a list of their own, different from every other trade we measure.

Memory assistants name these sites far more often: Avvo in 85% of their answers and the state bar in 76%, against 26% and 34% for web-search assistants, which name firms instead.

Sources cited. Eight of the assistants cite sources: ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews, Microsoft Copilot and our reconstructions of Siri and Alexa. The four memory assistants cite nothing. Across 16,201 citations in 3,369 legal answers:

The directories cited most were Justia, Avvo, Super Lawyers, FindLaw, LawInfo, Best Lawyers, Expertise.com and Lawyers.com. 71% of answers that cited anything cited at least one directory. The assistants differ sharply:

Share of each assistant's citations in answers about lawyers, October 2026. ChatGPT also cited bar and court sites for 25% of its citations. Microsoft Copilot cited too few sources in legal answers (76) to compare.

So a firm's own website is the main source for Gemini and ChatGPT, and directory profiles are the main source for Perplexity and Claude. A firm needs both to be found by all of them.

What advice does AI give people hiring a lawyer?

Most answers end with advice on choosing a lawyer. Across the 12 assistants, 34% told the client to ask about fees, 32% to talk to more than one lawyer, 25% to check the lawyer's licence or standing with the state bar and 21% to read reviews. 34% mentioned legal aid, a public defender or a bar referral service, mostly in criminal defense answers (62%).

Some assistants hand over more of the homework than others. Grok told clients to check bar standing in 69% of its answers, Meta AI in 47% and Microsoft Copilot in 46%; Google AI Overviews, Siri and Alexa almost never did. Meta AI said its answer was not legal advice in 71% of answers and Grok in 57%. Personal injury answers told clients to talk to several lawyers most often (50%).

For a firm, the advice is a preview of the first call. A client who found you through AI has often been told to check your bar record, compare you with another firm and ask exactly how you charge. Firms that answer those three things up front, on the page and on the phone, match what the client has been primed to ask.

How can a law firm stay ahead of AI?

These are patterns in what 12 assistants wrote across 200 legal markets, not a guaranteed way to be recommended. With that limit, the findings point to where a law firm can start. Each step is tied to the finding behind it.

1. Measure every assistant, not one. 73% of firms are named by only one assistant, and Grok, DeepSeek and Alexa often name no lawyer at all. Checking ChatGPT alone tells you little about where you stand. Ask each assistant the questions your clients ask, for your city and your practice area, and track the answers over time.
2. Earn and display third-party proof. The firms most assistants agree on are described with peer ratings, results and trial experience far more often than firms only one assistant names. Super Lawyers, Best Lawyers, Martindale-Hubbell AV ratings, Chambers and board certification are the recognitions AI repeats. Publish them in plain text on your site, not only as badge images.
3. Publish results where the rules allow. Verdicts, settlements and dismissals are named for 18% of personal injury picks. Follow your state's advertising rules on past results and disclaimers, and write each result in plain words: the type of case, the outcome and the year.
4. Complete the profiles AI sends clients to. Avvo and the state bar are named in nearly half of all answers, and in more than 80% of the answers that name no lawyer. Claim and complete Avvo, Justia, FindLaw, LawInfo, Martindale-Hubbell and Super Lawyers, and make sure your state bar record is current. For Perplexity and Claude, these directories are most of what gets cited.
5. Make your own website readable. For Gemini, ChatGPT, Siri and Google AI Overviews, a law firm's own site is the main source. One page per practice area and per office, written in plain text, stating what you handle, where, for how long and how you charge.
6. Say how you charge. For "affordable" questions, free consultations, contingency fees and flat fees replace rankings as the reason AI gives. If you offer a free consultation, work on contingency or charge a flat fee for a ticket or a first offence, say so clearly and consistently everywhere.
7. Lead with the reason clients in your practice area hear. Personal injury: results and "no fee unless we win". Criminal defense: prior prosecution or court experience and trial record. Business litigation: peer recognition and trial work. Traffic: price, speed and reviews.
8. Keep reviews coming and keep them consistent. Reviews are named for 22% of picks and Claude names them for 44%. They do not separate the consensus firms from the rest, but they are the price of entry: a firm without them gives the assistants one reason fewer.
9. Fix what memory assistants remember. Meta AI, Mistral and DeepSeek answer from what they learned, and quote phone numbers and specialties they cannot look up. A firm that has renamed, moved or changed its practice areas should make the current facts consistent across its site, its directory profiles and the bar, so the next model learns the right version.
10. Prepare for a client who has been coached. AI tells clients to check your bar standing (25% of answers), compare lawyers (32%) and ask about fees (34%). Answer those three questions before they are asked.

How this data is used. This research feeds the AI Syndicate platform, so our clients have the leading edge. Law firms on the platform see where they stand in their market's AI Local ranking and how they have moved, inside their dashboard, and their markets are measured more often than the rest of the index. What we learn across the whole index, such as which reasons the assistants give, which directories they cite and how each assistant answers questions about lawyers, is built into the twelve-assistant measurement the platform runs for each firm's own practice areas and questions.

How was this measured, and what can it not tell you?

Method. The October 2026 edition of AI Syndicate Local asked 12 assistants three fixed questions in each of 200 legal markets (50 metro areas, 4 practice areas), with the city named in the question and no user location or account history. A model extracts the firm and lawyer names from each answer. Each pick's description runs from its name to the next firm named, ending at a new heading. We then matched the description against fixed word lists for each kind of reason, and the whole answer against word lists for advice and sites. We checked a random sample of matches for each kind by hand and tightened lists that caught unrelated text.

How the twelve are reached. ChatGPT, Claude, Gemini and Perplexity answer with web search on. Google AI Overviews is the AI Overview Google shows. Microsoft Copilot is the Copilot answer Bing shows, reconstructed from the same Bing results when Bing shows none; in this edition every Copilot answer was a reconstruction. Siri and Alexa are reconstructions in the short spoken style each uses. Grok, Meta AI, DeepSeek and Mistral answer from model memory, because we cannot reach their live search; their consumer apps may answer differently.

What this cannot tell you.

• A reason in this paper is one the assistant wrote. Models do not reliably report why they chose something, so these are stated reasons, not causes. The comparison of consensus firms shows what is said about them, not what made them consensus firms.
• Siri, Alexa and Microsoft Copilot are reconstructions. Read them as close approximations, not recordings.
• Word lists miss reasons phrased in unusual ways and occasionally catch text that is not a reason. "Local" and "results" are the broadest lists and the least precise.
• Firm names are matched within a market by normalised text. Two spellings of the same firm can count as two firms, which makes agreement look a little lower than it is; a lawyer named personally and the same lawyer's firm count separately.
• We did not check whether the ratings, results, fees or phone numbers the assistants quoted are correct, or whether the firms named are in good standing.
• The extraction model sometimes misses a firm named only in passing, which can make an answer look as if it named no one. Google AI Overviews is most affected, because its firm names often appear in link cards that our record does not keep.
• Some assistants returned fewer usable answers: Google AI Overviews (285 of 600) and Microsoft Copilot (266) had none in Alabama, Alaska or Arizona, and Claude (431) and Alexa (483) had gaps in several states. Figures for those assistants rest on fewer markets.
• Each question is asked once a month, so a single answer can vary. The figures are averages over thousands of answers, not predictions for any one.
• Nothing here is legal advice or advice on legal advertising. Lawyers should follow their state's rules of professional conduct when they publish results, ratings or fees.

What stays private. The public index publishes each market's rankings and the assistants that named each firm. The individual answers are not published.

Statistical notes

• Unit of measurement. Shares of picks pool every pick by the 12 assistants across all legal markets. Shares of answers pool every answer. A market is one metro area and one practice area.
• Ranges. 95% ranges come from resampling the 200 markets 1,000 times. For example: firms named by only one assistant, 73.4% (72.3–74.4%); firms named by four or more, 7.7% (7.1–8.3%); their share of all picks, 31.1% (29.2–32.8%); answers naming no lawyer, 24.0% (23.0–24.9%); picks with a stated reason, 72.1% (70.9–73.3%); picks with a peer rating, 11.7% (10.8–12.6%).
• Missing answers. A question that returned no answer is left out, not counted as naming no one.
• Agreement. A firm's count is the number of different assistants that named it in that market, across the three questions. Names were matched after ignoring case, punctuation, "&" and "and", and the words law, firm, group, office, attorneys, lawyers, legal, injury, accident, trial, defense, associates and legal suffixes such as PLLC, LLP and PC.
• Consensus comparison. Shares are of picks, so each mention counts once; this keeps firms named more often from collecting more reasons by chance.
• Citations. A citation counts as a law-firm website when its domain contains a distinctive word from the name of a firm named in the same market, or a legal word such as law, attorney or injury, and is not a known directory, review site, bar or government site. Dictionary pages are left out.
• Typical values. "Typical" and "median" mean the same.

How to cite this report

Quote any figure in this report with credit to AI Syndicate Research and a link to this page. You may quote up to 50 words of text, or share a single chart with its source line intact. Reproducing the report or large parts of it needs written permission. A suggested citation:

A PDF copy and a plain-text Markdown version are available. The rankings behind the report are published under a CC BY 4.0 licence at aisyndicate.com/ai-local/, with machine-readable copies at /ai-local/data.json and /ai-local/all.md. Journalists, bar associations and researchers who need a cut of the data that is not here can write to support@aisyndicate.com.

Key figures

Every key figure in this report, in one table. The same figures are in findings.json and findings.csv, free to reuse under a CC BY 4.0 licence. Ranges are 95% intervals from resampling markets.

Sources

1. AI Syndicate Local: the October 2026 edition — AI Syndicate. Read Oct 9, 2026.
2. AI Syndicate Local methodology: how the index is measured — AI Syndicate. Read Oct 9, 2026.
3. AI Syndicate Local, machine-readable rankings with the assistants that named each business (CC BY 4.0) — AI Syndicate. Read Oct 9, 2026.