How AI finds lawyers.
When someone tells ChatGPT “I got rear-ended and the insurance offer feels low,” a referral comes back in seconds. This guide explains how that referral gets made: the sources AI reads, the signals it weighs, and the parts your firm controls.
AI refers clients to lawyers in three stages: it retrieves from sources it trusts (bar records, legal directories, case results, your site’s practice pages, review patterns), weighs them for authority, consistency, and specificity, with extra caution because legal outcomes are high-stakes, then composes an answer naming two or three firms matched to the story the client just told. The referral is earned through verifiable signals, not ad spend.
¶It starts with a story, not a search
Nobody types “best lawyer” into ChatGPT. They tell the story: “I got rear-ended and the offer feels low.” “My landlord kept the deposit.” “We need a prenup and don’t know where to start.”
The model classifies the story into an area of law before the client knows what kind of lawyer they need, then matches firms to that specific situation: practice area, jurisdiction, case size, urgency. Your visibility isn’t one number. It’s a different number for every story your future clients are telling.
1Stage one: what AI actually reads
Before naming a firm, models pull from a recognizable stack of legal sources:
- Bar records and court data. License status, admissions, disciplinary history. Foundational legitimacy signals the model checks quietly.
- Legal directories and editorial rankings. The Avvos and Super Lawyers of the world, plus local legal journals. Heavily weighted because they’re structured and vetted.
- Your practice-area pages. If your site explains exactly who you help and how, in crawlable HTML, the model can match you to stories. Generic “full-service firm” pages match nothing.
- Review specificity. “Settled my rear-end case for 3x the first offer” is a signal. “Great lawyer!” is barely one.
- Press, case results, and the model’s memory. Notable outcomes and coverage published over years. Slow to change, which is why early presence compounds.
2Stage two: how AI weighs what it reads
Retrieval gets you considered. Weighting decides whether you're named. Four properties consistently move recommendations in this category:
| Signal | What the model is checking | Influence |
|---|---|---|
| Authority | Is this claim corroborated by sources with a track record: editorial reviews, institutional records, established publishers? | High |
| Consistency | Do your practice areas, attorneys, and locations match across your site, directories, and bar records? Conflicts read as risk. | High |
| Specificity | Can the model cite something concrete, like “handles rideshare accident cases statewide, fee on recovery only,” or just slogans? | High |
| Recency & sentiment | Are mentions fresh, and do recent reviews describe the matter type being asked about? | Medium |
Missing from the table: TV budget, billboard count, firm size. A three-attorney shop with sharp practice pages, specific reviews, and two editorial citations regularly beats household-name firms in story-level referrals. The model rewards whoever it can verify.
3Stage three: how the answer gets written
The model compresses everything into two or three names, each with a one-sentence rationale fitted to the asker's situation. Three properties of that compression matter:
- There is no page two. Options four through forty simply don't exist in the conversation. Presence is closer to binary than rank.
- The rationale is the brand. The model's one sentence about you is your positioning now, and it traces to sources you can change.
- The default compounds. Models that repeatedly find you reliable in a moment keep reaching for you in that moment. Early presence becomes the default recommendation.
You can’t buy the referral. You can make your firm the easiest one for a cautious model to verify, in the exact stories your future clients are telling it.
§Why legal is held to a higher bar
Legal questions are classic YMYL territory. Models are more conservative, more citation-dependent, and more disclaimer-prone in legal moments. They recommend fewer names and lean harder on verifiable records.
That bar is a moat. Thin content gets filtered out entirely, and the firms that do the verification work face less competition inside the answer. It also means compliance-safe presence building, accurate data, earned citations, educational content, is the only kind that works.
✓The signals you control
Everything above reduces to a working checklist. This is the same gap analysis Aethon's presence map runs automatically. Here's the manual version:
- Publish story-level practice pages in crawlable HTML: “rear-ended with a low insurance offer” content, not “personal injury services.”
- Reconcile every directory and bar listing. Names, practice areas, and locations must match everywhere.
- Structure attorney and firm data with schema markup: attorneys, practice areas, jurisdictions, fees model.
- Earn citations in legal press and local journals the models already trust for your region.
- Cultivate matter-specific reviews. Ask satisfied clients to mention the case type and the outcome experience.
- Measure by story, monthly. Your presence in “DUI first offense” and “low settlement offer” are different numbers with different fixes.
Items one through five are exactly what the Action Engine generates and ships. Item six is the presence map itself.
See your firm’s presence map
30 minutes. Your market live, every moment, every gap.
More for legal teams
Be the firm in the referral.
Find out whether you're in the answer. 30 minutes. your market live across the platforms that matter.