How to Explain AI Search to Your Partners (Before Budget Season)

<!-- INBOUND LINKS STILL OWED (files outside this draft's scope — do not edit them here): 1. /case-recommendation-share — a line pointing here as the internal-sell packaging for the metric ("how to present this to partners"), per CONTENT_PLAN B4. 2. /methodology — link here from the section a skeptical partner is sent to inspect. 3. /ai-search-statistics-law-firms — link here from the adoption-statistics section, since this piece is where those numbers get their stated-preference labels for a partner deck. 4. /legal-ai-check result screen — CONTENT_PLAN §5 requires the "score / metric explained" result state to link to /case-recommendation-share and to offer a copyable, dated result line suitable for a deck, with a link back here. The result screen currently carries no internal links at all. PUBLICATION DEPENDENCY: this draft links to /blog/what-ai-visibility-vendors-can-promise-law-firms, which is still held in content/drafts/. That link is dead until B2 ships; ship B2 first or drop the sentence that carries it. -->

You are not the one who decides. You are the one who has to walk into a room where seven partners are looking at a spreadsheet and explain why a line item nobody has heard of belongs on it.

This is the briefing for that meeting: three sentences that explain AI recommendation without an acronym, one number that survives questioning, the objections to raise before a partner does, and a one-page memo you can forward. One warning first. The fastest way to kill this line item permanently is to oversell it once.

What your partners will actually ask

"Is this a scam?" Fair question, and parts of the category are recycled SEO with a new invoice attached. The part that is not: potential clients are asking assistants who to hire, the assistants answer with firm names, and until you measure it you do not know whether yours is one of them. Say the first half out loud. It buys you the second half.

"How is this different from the SEO we already pay for?" They overlap, and you should concede that: directory profiles, reviews, and answerable practice-area pages help both. The difference is what comes back. SEO reporting tells you where you rank. An AI answer either names your firm or it does not, and there is no page two to be on.

"How many clients has this brought us?" Nobody can tell you, including any vendor who says otherwise. Most AI recommendations end with no click at all. No credible signed-case data from AI referrals has been published for law firms, so what you propose to measure is the upstream half — whether the firm gets named at all. If a partner pushes, what intake and analytics can and cannot tell you is the long answer.

"What would we even do about it?" This question decides the vote, because a problem with no action item is just anxiety. The answer is unglamorous: directory profiles, review recency, name and address consistency, plain-language pages that answer client questions, earned coverage. Two diagnostics walk the causes and fixes in order: why a firm is not showing up, and why a competitor is named instead.

"Can't we just check it ourselves?" They will, in the meeting, on a phone. One check is close to a coin flip: an audit of retrieval-augmented commercial recommendation found that re-running an identical prompt returned recommendation sets with only about 50–61% overlap, and cosmetic rewording dropped that to about 29% (arXiv, 2026 — an unreviewed preprint from an AI-visibility vendor, measured on consumer buying questions rather than legal ones; say that limitation out loud if pressed). The phone test will disagree with yours, and that is the argument for repeated measurement rather than against it.

The three sentences

To explain AI search to law firm partners, describe what the assistant does rather than how it works: a person with a legal problem asks ChatGPT, Gemini, or Google's AI answers who to hire, the assistant replies with the names of two or three specific firms, and being one of those names functions as a referral that the firm never sees in any report.

When someone asks ChatGPT, Gemini, or Google's AI answers who to hire for a legal problem, the answer names two or three specific firms. Being one of those names is worth a referral, and it happens with no record on our side at all. We do not know how often we are named, how often our competitors are named instead, or which sources the assistants read to decide.

That is the whole pitch. No acronym, no slide about how large language models work. If someone asks how they work, say the mechanism matters less than the sources: assistants assemble answers largely from third-party pages, and in legal that means directories.

The one number for the slide

The metric worth defending is Case Recommendation Share: the percentage of AI answers to client-style hiring questions that recommend your firm, meaning put it forward as one to contact rather than merely mentioning or citing it. It holds up under questioning for three reasons. The formula is published, so anyone can check the arithmetic. Failed scans are excluded from the denominator, so an outage cannot manufacture a drop. Mentions and citations do not count, which produces a smaller, harsher number that is hard to accuse you of inflating.

A worked version, with fictional firms. Delacroix & Wren Injury Law (delacroixwren.example) runs 40 scans in one market over a month — eight client-style hiring questions put to each of five assistants. Thirty-six answers come back; four scans fail and are excluded. The firm is recommended in 7 of the 36, so its Case Recommendation Share is 19.4%. In the same answers, Thackeray Trial Group (thackeraytrial.example) is recommended in 15, for 41.7%. The slide reads: we are named in roughly one in five answers our potential clients would see; the firm across town is named in two of five.

The numbers that will get you caught

Two labels save you in the room. Consumer-adoption figures are stated preference: the 41.9% of US adults who say they would use ChatGPT to research a lawyer — up from 9% in 2023, n=1,110 — were asked which sources they would use, not observed using them. And the ChatGPT conversion advantage everyone quotes (15.9% versus 1.76% for Google organic) is a single unnamed client in one 2025 case study, where AI was about 0.07% of that client's organic traffic. Not a legal figure, not an industry rate. Every number here with its method label lives on the statistics hub; send partners the link, not a screenshot.

What to promise internally, and what not to

Write the promise down beforehand. Whatever you say in the room is what you are measured against in six months. And if an outside vendor is making promises of its own, the firm — not the vendor — answers to the bar for marketing claims made on its behalf: what an AI-visibility vendor can promise your law firm covers that side of it.

Safe to promise: that you will measure how often assistants recommend the firm, report it monthly with the underlying answers attached, fix the source-layer problems the measurement exposes, and say so when the number does not move.

Not safe to promise, at any budget: that the firm will be recommended, that a competitor will stop being recommended, or that a given number of cases will result. Recommendations are probabilistic, and the question's wording moves them as much as anything the firm does — the paraphrase figures above are the everyday version of that. A vendor who guarantees placement is either misunderstanding it or counting on you not to check.

What a monthly report should contain

1. Case Recommendation Share, with the denominator. Recommendations over successful answers, both numbers shown.
2. The same figure for two or three named competitors, so the number has a scale.
3. A per-assistant breakdown. One overall figure can hide an assistant where the firm sits at zero, and assistants genuinely disagree: only 11% of cited domains appeared across more than one of the four platforms tested in an analysis of 118,000 answers (Whitehat SEO, March 2026).
4. The cited sources, ranked. This is the action list: it names the directory profile or third-party page doing the work.
5. Method and changes. Question set, runs, date window, anything that changed since last month. A report you cannot audit is a report you cannot defend; ours is public.

If you are building the client deliverable now, the free law-firm AI visibility report template turns those five requirements into a copy-ready monthly structure, including the evidence appendix and the language that keeps an unmeasured value from appearing as zero.

The one-page partner briefing memo {#partner-briefing-memo}

Fill in the blanks, delete this sentence, and send it the day before the meeting. Reuse it freely, inside your firm or with your clients.

AI recommendation: where we stand Prepared by ____________ · For the ____________ meeting · Data as of ____ / ____ / ______

What this is. When a potential client asks an AI assistant who to hire for a legal matter, the assistant names specific firms. This memo reports how often it names us.

What we measured. ____ client-style hiring questions, run across ____ assistants, ____ times each, between ____ and ____. Failed scans excluded: ____.

Our number. We were recommended in ____ of ____ answers (____%). Recommended means the answer put us forward as a firm to contact, not merely mentioned or cited us.

Named instead of us. 1. ____________ (%) · 2. ____________ (%) · 3. ____________ (____%)

What the assistants read. Top cited sources: 1. ____________ 2. ____________ 3. ____________

What we propose to do. ________________________________________________

What we are asking for. $________ over ____ months, reviewed at ____ months against the number above.

What this does not tell us. It does not tell us how many signed cases came from AI, and no available method does. It measures AI outputs, not the quality of any firm's lawyering. Answers vary between runs, so single checks are unreliable and only the trend is meaningful.

Memo template last verified 17 August 2026.

Bring the objections yourself

Say these before a partner does. The credibility is worth more than the ground you give up.

The honest limits, in the room

Three limitations belong in the meeting, not in a footnote. Answers are probabilistic: identical prompts return different firm lists, so one check is a snapshot, not a measurement. A model update can reshuffle everything: assistants change without notice, which makes trends trustworthy and absolute values fragile. Attribution stays incomplete: if the number improves, your work is a plausible cause, not a proven one.

FAQ

What is the most persuasive thing to show a partner?

Your own firm's answers, unedited. Not a chart. Open two or three real AI responses to a question a potential client in your market would ask, and read the firm names out loud. The argument makes itself when the room hears a competitor's name.

How much should a firm budget for this in year one?

Measurement first, work second. The measurement is small, and the work is mostly what the firm should already be doing: directory profiles, review recency, consistent name and address, pages that answer client questions. Propose the smallest budget that produces a monthly number, an action list, and a review date: a modest ask that gets renewed beats a large one that gets cut.