When AI-visibility measurement is a waste of money for a law firm

<!-- PUBLISHER NOTES (files outside this draft's scope — do not edit them here): 1. CTA SUPPRESSION IS A BUILD REQUIREMENT, NOT A COPY PREFERENCE. See schema_notes. If the blog layout injects a global CTA band, footer signup, or exit modal, this route must opt out. The page tells the reader that two of its three verdicts end without a CTA. A component that contradicts that is a defect, and this audience will screenshot it. 2. INBOUND LINKS OWED TO THIS PAGE: a. /blog/what-ai-visibility-vendors-can-promise-law-firms — one line pointing here as the situational counterpart to its regulatory question. b. /blog/ai-referral-intake-attribution-law-firms — one line pointing here from the intake card, since Verdict 2 routes readers into that piece. c. /about and /methodology — the non-ICP is published by name here; link it. d. B6 (/blog/vet-ai-visibility-vendor-law-firm) is being drafted in a parallel session. The two are deliberately disjoint: B6 = "is this vendor's number real", this page = "is this category worth anything to my firm". Whichever ships SECOND opens by pointing back at the first; wire the pair of links then. Neither file should link to the other before both are live, or one of them ships with a 404. 3. STATS HUB CORRECTION OWED (/ai-search-statistics-law-firms): the hub currently attributes the 23.6% legal AI Overview figure to the 5WPR report. That attribution is wrong at both ends — the 5WPR release contains no 23.6% figure, and it cites Ahrefs nowhere. The primary is Ahrefs' AI Overview triggers study — 146,122,391 desktop SERPs, September 2025, Legal category 354,573 of 1,499,855 SERPs. Reattribute to Ahrefs and drop 5WPR from that line entirely rather than demoting it to a waypoint; it is not the route by which that figure reached the industry. 5WPR is the source of the 78% claim and nothing else here, and the hub's "no locatable primary" note applies to that figure alone. Flag for whoever owns the hub: of the seven statistics in the 5WPR release, only the 78% carries an attribution of any kind, so it should be treated as a weak source wherever the hub leans on it. 4. Also owed to the hub: the Ahrefs 58% CTR figure now has a locatable primary (ahrefs.com/blog/ai-overviews-reduce-clicks-update/, published February 4, 2026, December 2025 data). The hub currently cites the Business Wire announcement, which timed out on re-verification. Point it at the study. -->

We sell AI-visibility measurement for law firms. This page is about the firms that should not buy it, from us or from anyone, and it ends in three verdicts of which two are "spend nothing."

Two of those verdicts carry no call to action. No free check, no demo, no email box, no "just reply and I'll take a look." If you land on one of them, the page is over and you owe us nothing, including your attention. That is checkable: scroll to the bottom of Verdict 2 and Verdict 3 and see whether anything is being sold. If something is, we broke our own rule and you should say so publicly.

Here is why a vendor would write this. Our published customer definition names the firms we do not serve, and it names them in a strategy document we wrote for ourselves before we wrote this page: BigLaw and B2B practices whose buying journey is not a local consumer shortlist, rural generalists, solo firms without a marketing budget. Selling into that list wastes their money and our time. Publishing the list is cheaper than discovering it on a call.

The question this page is not answering

There are two different questions a skeptical firm asks, and vendors happily blur them.

"Is this vendor's number real?" is a methodological question. It has testable answers: show me the prompts, the run count, the date window, the raw answers, the denominator. That question is answered against a specific vendor, in a demo, after you have decided the category might matter. It is not this page.

"Is this category worth anything to my firm?" is situational. It has nothing to do with whose software you are looking at, and you can answer it in ten minutes without contacting anybody. That is this page.

Answer the second one first. Most of the money wasted in this category is wasted by firms that skipped it.

The five inputs {#five-input-test}

You already have all five. None requires a tool, a trial, or a conversation.

This test may be reproduced and used freely with attribution to askbriefly.ai.

Two notes on the inputs, because the wrong reading of them flatters us.

Input 2 is a total, not a proposed budget. The arithmetic that matters is: take whatever price a vendor quotes you, multiply by twelve, and divide by that total. If measurement would consume more than about a tenth of everything you spend on demand generation, you are buying a gauge that costs a tenth of the engine. That cut line is our judgment, not a measured finding, and we are labelling it as judgment because this page is worthless if it smuggles opinions in as data.

Input 4 is a count, not a feeling. Eight is also our line. It is a proxy for whether an AI assistant asked "who should I call" in your market has enough candidates to produce a contested shortlist. In a market with three firms, the answer names all three and there is nothing to measure. Lawyer density varies by roughly a factor of four and a half across US states, from about 2.1 per 1,000 residents in the least dense (South Carolina, Arizona, Idaho) to 9.6 in New York, against a national average near 4, per the American Bar Association's Profile of the Legal Profession. One caveat we owe you: americanbar.org returns HTTP 403 to automated retrieval, so we could not open the report to confirm which edition's table those state figures come from — the same spread has been in circulation since at least the 2023 edition. It is the ranking the gate rests on, not the decimal. Our market catalog concentrates on twenty-five metros for exactly this reason. If you are outside that kind of density, the shortlist is short and you can read it yourself.

The three verdicts

Run the gates in order. Stop at the first one that catches you.

Verdict 3 — this is not your channel {#verdict-not-your-channel}

Any one of these is disqualifying on its own:

• One attorney, no real demand-generation budget. If input 2 is a rounding error, every dollar of it belongs in intake capacity or a referral relationship, not in a dashboard. Measurement tells you about a shortlist you have no capacity to act on.
• You cannot name eight competing firms for your case type in your market. The shortlist is not contested. Ask the assistants yourself once a quarter and read the answer.
• Your work does not arrive through a local consumer shortlist. Panel counsel, RFPs, institutional referrals, in-house rosters, partner networks, B2B procurement. If nobody types "best [practice] lawyer in [city]" on their way to hiring you, the thing this category measures is not the thing that produces your matters.
• Twelve months of the quoted price exceeds a tenth of input 2. The instrument costs too much relative to what it observes.

That is the end of this verdict. Nothing is being sold to you here, and nothing further on this page is addressed to you. If the category becomes relevant later, the gates above are the same gates.

Verdict 2 — instrument intake first, revisit in ninety days {#verdict-instrument-first}

You cleared every gate above, and input 5 is a no: you cannot say, from a recorded field rather than a memory, how last month's signed matters arrived.

Then measuring AI recommendations now buys you a number you cannot connect to anything. You will get a chart, the chart will move, and in six months you will have no way to tell whether the movement mattered. Worse, you will be susceptible to the vendor's version of causation, because you will have nothing of your own to check it against.

The fix is unglamorous and free: change one intake field from open text to a required dropdown with a named option for AI assistants, brief whoever answers the phone, and run a fifteen-minute monthly reconciliation. The protocol is written out, including the exact wording of the question and the dropdown values, in the intake attribution piece. Nothing in it is specific to any vendor, and no part of it requires an account.

Give it ninety days. Then you will have a baseline of your own, and the gates above will answer themselves.

That is the end of this verdict too. There is nothing to buy here.

Verdict 1 — measuring now is defensible {#verdict-measure-now}

You cleared all five. Multiple attorneys, real demand-generation spend, a contested local shortlist, eight nameable rivals, and intake instrumented well enough to notice a change.

Then measurement is a reasonable line item, with two conditions attached.

Buy a report, not a forecast. Nobody outside a model developer can commit to what an assistant will say about you. What a vendor can commit to is running a fixed question set across named assistants on a stated schedule, storing every answer, and telling you what came back. The four-rung version of that distinction, and the bar-advertising exposure that rides on it, is in what an AI-visibility vendor can legally promise your firm.

Insist on repeated runs. A single check is a snapshot of one moment. A 2026 audit of retrieval-augmented recommendation found identical reruns of the same prompt returning recommendation sets that overlapped only about 50 to 61 percent, and cosmetic rewording dropped that to about 29 percent (arXiv preprint, unreviewed, authors affiliated with an AI-visibility vendor, measured on commercial buying questions rather than legal ones — we cite it against our own interest, since it argues that per-prompt tracking is unreliable). Any number quoted to you from one run of one question is noise wearing a decimal point. Ours is defined over repeated scans for this reason, which is what Case Recommendation Share means.

If you want to see what the assistants currently say about your firm before deciding anything, the check is free and ungated.

Run the free check →

No signup, no card. You read the actual answers.

Before you pay anyone: the free version

Whatever your verdict, one thing costs nothing. Open ChatGPT, Gemini, and Perplexity, ask the questions a client would actually type, and read the answers. Write down which firms get named. Repeat it next month.

That is a worse instrument than a scheduled scan, for the reasons in the paragraph above about rerun variance, and it is still better than a vendor's screenshot. If you want the question set rather than inventing one, our local hiring-question packs are ungated, ask for no email, and contain no measurement of any kind. They are a list of questions.

The numbers you will be shown, and what each one is actually worth

A pitch for this category leans on four or five figures. Every one below is real and sourced. Not one of them answers the question of whether your firm should spend money, which is the entire problem.

Consumer intent is stated preference. In a 2026 survey of 1,110 US adults aged 18 to 65, 41.9% said they would use ChatGPT to research a lawyer, 9.5% said they would use AI sources only — no Google, Facebook, Yelp, or YouTube — and 76% of respondents aged 45 to 60 said they would use AI broadly to research law firms (iLawyerMarketing). Every one of those is what people said they would do. Stated preference runs hot. It is direction, not magnitude, and it is not a forecast of your phone.

Referral growth is real, large, and not legal-specific. Monthly LLM-referred sessions across 166 tracked GA4 properties rose from 65,249 to 644,478 from November 2024 to May 2026 (Previsible, 6.77 million sessions analyzed). That is a cross-industry panel. Legal is among the verticals it tracks, and there is no legal-only cut of that growth figure.

The conversion figure everybody quotes is one client. ChatGPT-referred visitors converting at 15.9% against 1.76% for Google organic comes from a single unnamed client, industry undisclosed, measured October 2024 to April 2025 (Seer Interactive). On that same site, AI accounted for roughly 0.07% of organic traffic. Use it as evidence about intent quality. It is not an industry conversion rate and it is not a legal one.

AI Overview incidence on legal queries is an unreconciled range, not a number. The low end has a locatable primary: Ahrefs' analysis of 146,122,391 desktop SERPs from September 2025 puts the Legal category at 23.6%, which is 354,573 AI Overview SERPs out of 1,499,855 legal SERPs, against a 20.5% all-SERP baseline and 57.9% for question-phrased queries generally (Ahrefs). The high end does not: a June 2026 press release claims "78% of legal queries now trigger a Google AI Overview" and attributes it to "Semrush analysis of more than 10 million keywords" (5W Public Relations). We could not locate that figure in Semrush's own published study, which reports 15.69% of queries triggering AI Overviews in November 2025 against a July peak of 24.61%, with no legal-specific breakout. We are not picking one. If a deck quotes you 78%, ask where the study is.

The click-loss figure is well-evidenced and is about Google, not about assistants. Ahrefs re-ran its CTR study on December 2025 data across 300,000 keywords using aggregated Search Console data, and found the presence of an AI Overview reducing the click-through rate for position 1 by about 58%, up from the 34.5% it measured on March 2025 data (Ahrefs, February 4, 2026). That is a real finding about the value of a ranking. It says nothing about whether your firm gets recommended, which is a different measurement.

All of the above, with method labels, lives on our statistics hub.

The number nobody has, including us

There is no credible published figure for signed cases produced by AI referrals. Not for the legal industry, not from the agencies, not from the platforms, and not from Briefly. We said so on the intake page and it is still true. The instruments in the market cannot produce it: Google Analytics 4's AI Assistant channel, live since May 13, 2026, counts sessions from a list of assistants it does not publish and explicitly excludes Google's AI Overviews and AI Mode; Search Console's generative-AI report, published June 3, 2026 with data from May 18, 2026, reports impressions only — no clicks, no click-through rate, no queries. Neither sees a recommendation that ended in a phone call.

Anyone who quotes you a signed-case number for this channel has estimated it. We could estimate one too. It would be fiction, and offering it would be the fastest way to deserve the skepticism this audience already has.

What a lead costs, and why no published benchmark will tell you

Input 2 tempts people toward an industry benchmark. There isn't one worth using.

Two named sources, both reachable, both openly published, measuring legal advertising cost:

Those differ by two orders of magnitude. Neither is wrong. They measure different things over different periods with different instruments, and neither is audited by anyone. Treat both as what buyers are told, not as what anything costs. This is exactly why input 2 is your profit and loss statement and not a benchmark: your own number is the only one that has been through your own bank account.

Numbers we cut from this article, and why

Publishing the cutting-room floor is cheap for us and useful to you, since these figures circulate and you will meet them.

• "Law firms spend 2–10% of gross revenue on marketing." Every version of this we could reach traces to agency blogs and aggregator posts quoting each other. No audited primary located. Cut.
• "54% of firms increased marketing budgets; 74% at firms with 51–100 attorneys." The primary is the 2025 LMA and Above the Law Legal Marketing Decision-Makers Survey. The PDF on legalmarketing.org returns HTTP 403 to automated retrieval, and Above the Law's own announcement page carries no budget percentages. We do not cite through aggregators, so both figures are cut rather than laundered.
• "Personal injury cost per click is up 568% since 2021." The comparison is stated on the agency page it is usually credited to, but with no published 2021 baseline and no stated method behind it, and everywhere else it appears it is quoted secondhand from there. A percentage with no denominator and no way to reproduce it is not a benchmark. Cut.
• "78% of legal queries trigger an AI Overview." Kept above only as the unlocatable upper endpoint of a disputed range, never on its own.

If you can send us the primary for any of these, we will add it and credit you.

What this page costs us

A page that disqualifies readers is a page that converts worse. That is the trade, and it is deliberate: we would rather have twelve conversations with firms that clear the gates than forty with firms that do not, and we would rather the reader who does not clear them remember that we said so.

There is a version of this article that would be a lead-qualification form with a disclaimer on top: two soft "not yet" verdicts, each with a helpful little box at the bottom offering a free check anyway. That version is worse than useless with this audience, because everyone recognizes it, and it would make the promise in the second paragraph a lie. So the two don't-buy verdicts end where they end. Our editorial policy holds us to that, and our methodology states what our own measurement cannot see.

Briefly measures what AI assistants say. It does not rank, rate, or endorse attorneys, and nothing on this page is legal advice or guidance on your professional obligations.

FAQ

Is AI visibility worth it for a small law firm? Often no, and the disqualifying conditions are specific rather than about size: a single attorney with no demand-generation budget, a market where you cannot name eight firms competing for the same case type, work that arrives through referrals or panels rather than a local consumer shortlist, or a quoted price whose annual total exceeds about a tenth of everything you spend on demand generation. A multi-attorney consumer firm in a contested metro that already spends materially on ads or SEO is a different case. Run the five inputs above before talking to anyone.

How much should a law firm spend on tracking AI recommendations? We cannot tell you, and neither can the published benchmarks. The commonly quoted "2–10% of gross revenue" figure has no primary source we could locate, and legal advertising cost figures from named sources span from about $9 per click to $1,000 per click depending on what was measured and when. The arithmetic that works: take the quoted price, multiply by twelve, and compare it to your own total demand-generation spend from last year's books.

Can I check ChatGPT myself once and skip the tools entirely? You can check, and you should, but one check proves very little. An audit of retrieval-augmented recommendation found identical reruns of one prompt returning sets overlapping only about 50 to 61 percent, and cosmetic rewording dropping that to about 29 percent — an unreviewed preprint from vendor-affiliated authors, measured on commercial buying questions rather than legal ones. A self-check tells you whether you appear at all, which is usually the only question a firm at Verdict 2 or 3 needs answered. It does not tell you whether anything changed.

---

John Rice builds and operates the scan engine behind Briefly, which runs client-style lawyer-hiring questions across ChatGPT, Gemini, Perplexity, Google AI Overviews, and Google AI Mode on a recurring schedule and stores every answer it collects. He is not a lawyer, gives no legal advice, and nothing here is guidance on your firm's professional obligations; he measures what AI assistants say, with receipts. The exclusions on this page are the customer definition Briefly wrote for itself, published verbatim. Methodology · About John

Sources

Consumer and traffic data

• iLawyerMarketing, "What Online Sources Do People Use to Research and Find Attorneys in 2026?" — stated-preference survey, n=1,110, ages 18–65: https://www.ilawyermarketing.com/what-online-sources-do-people-use-to-research-and-find-attorneys-in-2026/
• Previsible, 2026 AI Traffic Report — 166 GA4 properties, 6.77M LLM-driven sessions, November 2024 to May 2026: https://previsible.com/seo-strategy/ai-traffic-report-july-2026/
• Seer Interactive, "How Traffic from ChatGPT Converts" — single unnamed client, industry undisclosed, October 1 2024 to April 30 2025: https://www.seerinteractive.com/insights/case-study-6-learnings-about-how-traffic-from-chatgpt-converts

AI Overviews

• Ahrefs, "What Triggers AI Overviews? 86 Factors and 146 Million SERPs Analyzed" — 146,122,391 desktop SERPs, September 2025; Legal 354,573 of 1,499,855: https://ahrefs.com/blog/ai-overview-triggers/
• Ahrefs, "Update: AI Overviews Reduce Clicks by 58%" — Ryan Law and Xibeijia Guan, February 4 2026, December 2025 data, 300,000 keywords: https://ahrefs.com/blog/ai-overviews-reduce-clicks-update/
• 5W Public Relations, "AI Overviews Are Reshaping Legal Search" (June 11, 2026) — origin of the 78% claim: https://www.prnewswire.com/news-releases/ai-overviews-are-reshaping-legal-search-302776459.html
• Semrush, "AI Overviews Study" — 10M+ keywords, January to November 2025; no legal-specific figure and no 78% appears in it: https://www.semrush.com/blog/semrush-ai-overviews-study/

Platform reporting

• Google Analytics Help, "[GA4] What's new in Analytics" — AI Assistant channel release note, May 13, 2026: https://support.google.com/analytics/answer/9164320
• Google Analytics Help, "[GA4] Default channel group": https://support.google.com/analytics/answer/9756891
• Google Search Central Blog, "Introducing Search Generative AI performance reports in Search Console" (June 3, 2026; data from May 18, 2026): https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports
• Search Console Help, "Generative AI performance report (Search)" — impressions only: https://support.google.com/webmasters/answer/16984139

Cost figures, both unaudited, both labelled above as what buyers are told

• LocaliQ, "Legal Search Advertising Benchmarks" — 256 US campaigns, April 2022 to March 2023: https://localiq.com/blog/legal-search-advertising-benchmarks/
• iLawyerMarketing, "Most Expensive Google Ads Keywords in the Legal Industry 2025" — Google Keyword Planner estimates, 21,000+ keywords: https://www.ilawyermarketing.com/most-expensive-google-ads-keywords-legal-industry-2025/

Market structure

• American Bar Association, Profile of the Legal Profession — lawyers per 1,000 residents by state. americanbar.org returns HTTP 403 to automated retrieval, so we could not open the report to confirm which edition's state-density table these figures belong to; the same spread appears in coverage of the profile from December 2023 onward. Cited above without an edition year for that reason: https://www.americanbar.org/news/abanews/aba-news-archives/2025/12/aba-2025-profile-of-the-legal-profession-report/

Instability

• "Paraphrase Brittleness in Production Retrieval-Augmented Commercial Recommendation" (arXiv, May 22 2026) — unreviewed preprint, authors affiliated with an AI-visibility vendor, commercial buying questions rather than legal: https://arxiv.org/abs/2605.27440

Located but not cited — the primary for the LMA/Above the Law budget figures is unreachable to automated retrieval (HTTP 403), so those figures were cut rather than sourced secondhand: https://www.legalmarketing.org/Portals/0/CMO%20Survey%202025_LMA-ATL_1.pdf

Related on this site

• Did AI send you that case? The intake question most firms get wrong — the instrument Verdict 2 requires.
• What an AI-visibility vendor can legally promise your law firm — the promise ladder and who carries the bar exposure.
• AI Search Statistics for Law Firms — every third-party number, with its method label.
• Case Recommendation Share — the metric, and why it is defined over repeated scans.
• Methodology · Editorial policy