AI Visibility for Law Firms: How to Get Mentioned by ChatGPT & Google AI in 2026

What AI visibility is — and why it's replacing the first page of Google

AI visibility has two components, and most firms conflate them. Mentions are when an AI engine names your firm in its answer — "firms frequently recommended for family law in Tampa include…" Citations are when the engine links your site as a source for a claim it makes. Mentions drive direct client acquisition; citations drive the referral traffic and authority that eventually produce mentions. You need both, and they're earned through different mechanisms — mentions through your entity footprint across the web, citations through the structure and depth of your own content. Treating "AI SEO" as one undifferentiated thing is why most early advice on this topic is useless.

The behavioral shift underneath this is well documented. Pew Research Center found that roughly a third of US adults have used ChatGPT, about double the share from two years earlier, with the steepest adoption among adults under 50 — exactly the demographic hiring divorce, immigration, and injury lawyers. Meanwhile, the majority of Google searches now end without a click on any traditional organic result, because AI Overviews answer the question on the results page itself. When I run baseline audits on US law firms, the typical firm appears in zero of the forty client-intent prompts we test. The first page of Google took twenty years to get this crowded. The AI answer layer is, right now, almost empty.

How each AI engine picks its sources: Overviews, ChatGPT, Perplexity

Google AI Overviews are the best documented of the three. Google has publicly described a "query fan-out" approach: when a query triggers an Overview, the system issues a cluster of related searches, retrieves candidate pages from Google's existing index and ranking systems, synthesizes an answer, and attaches links to the pages that support each claim. The exact mechanics shift with updates, but two practical consequences have held. First, you generally need to already rank well — in practice, somewhere in the top ten to twenty results — for at least one related query to be a realistic source candidate — AI Overviews are downstream of SEO, not a separate channel. Second, Google's own documentation on AI features confirms there is no special markup or opt-in: the same crawlability, ranking, and structured-data fundamentals govern eligibility.

ChatGPT works through two distinct layers, and they reward different work. The parametric layer is the model's training data: when ChatGPT answers without browsing, it names firms it "remembers" from the corpora it was trained on — legal directories, news coverage, Reddit threads, bar association pages. This layer updates only when a new model ships, so it lags reality by months and rewards a long, consistent entity footprint. The retrieval layer is ChatGPT search: the model runs live web queries through its search infrastructure and its own crawler (OAI-SearchBot), and tends to cite pages that answer the question directly and render cleanly without JavaScript gymnastics. Perplexity is retrieval-first by design — every answer is grounded in a live search against its own index built by PerplexityBot, every claim carries a citation, and it is consistently the fastest engine to pick up new content, sometimes citing pages within days of indexing.

• AI Overviews: rank first, get cited second — eligibility flows from Google's index
• ChatGPT (no browsing): mentions come from training data; build the entity layer
• ChatGPT (search mode): citations favor direct answers on fast, crawlable pages
• Perplexity: always cites, indexes fast — the best early scoreboard for new content
• All three: corroboration matters; claims confirmed across multiple sources win

AEO vs SEO: what changes and what carries over

Answer engine optimization is not a replacement for SEO — it is a layer on top of it, and roughly 70% of the work is identical. Everything in my complete law firm SEO guide still applies: technical health, crawlability, topical depth, E-E-A-T signals, local presence. AI Overviews almost never cite a page Google hasn't ranked; ChatGPT's search mode leans on conventional search results; Perplexity's index favors the same authority signals Google does. A firm with broken SEO fundamentals chasing "AI optimization" is installing solar panels on a house with no wiring.

What genuinely changes is the unit of optimization. SEO optimizes pages to win a click; AEO optimizes passages to be quoted. That means question-phrased headings, a complete 60–150 word answer at the top of every section before the supporting detail, explicit factual anchors (statutes, dates, dollar figures with sources), and one idea per section so the extraction is clean. Measurement changes too: rank position and click-through rate become trailing indicators, while mention rate and citation rate across a panel of client-intent prompts become the leading ones. And the competitive geometry changes most of all — a results page has ten organic slots plus ads and maps, but an AI answer typically cites three to eight sources and names even fewer firms. AEO is winner-take-most in a way classic SEO never was.

Structured data: making your firm machine-readable

Structured data is how you remove ambiguity for machines, and law firms are unusually well served by the vocabulary. The minimum stack: LegalService or Attorney markup on your homepage and practice-area pages with areaServed, serviceType, and address filled in; Person schema on every attorney bio with hasCredential, memberOf (bar admissions), and alumniOf; Article schema with a named author and dateModified on every substantive post; and FAQPage on pages with genuine Q&A content. The sameAs property deserves special attention — it explicitly ties your site to your firm's profiles on Avvo, Justia, your state bar directory, and Google, which is precisely the cross-source corroboration AI retrieval systems use to resolve "is this the same firm?"

One honest correction to most advice you'll read: Google stopped showing FAQ rich results for most websites back in 2023, so FAQPage markup no longer buys you a visual rich snippet. It still matters — marked-up Q&A pairs are unambiguous to any parser, and unambiguous content is cheaper for a retrieval pipeline to extract and trust. Validate everything in Google's Rich Results Test, because one missing required field fails silently. When CaseGap audits law firm sites, broken or absent schema is the single most common finding — north of 80% of firms either have none or have markup that doesn't validate. It is a one-day fix with compounding returns.

Topical authority: why ten deep pages beat a hundred thin ones

Retrieval systems don't evaluate your site the way a human skimming your blog archive does — they evaluate whether, for a given question, your page contains the most complete, specific, well-supported answer available. A 3,000-word pillar on "how child custody is decided in Texas" that cites the Family Code, explains the best-interest factors, covers modification and enforcement, and answers the eight follow-up questions a parent actually has will outperform thirty 400-word posts that each gesture at the topic. Thin content was already losing in classic SEO; in AI retrieval it's not even a candidate, because there's no passage worth extracting.

The practical structure is the topic cluster I describe in my law firm content marketing guide: one pillar per practice area, eight to twelve supporting pieces that each fully answer one question, dense internal linking between them. Depth in a narrow cluster also concentrates your authority signal — engines that see your domain repeatedly supplying the best answer on Texas custody questions start treating you as the entity to consult on that topic. This is why I tell firms to win one cluster before starting a second. The practice-specific dynamics differ enough that I've written dedicated playbooks for personal injury firms and family law practices — injury queries skew toward urgent factual questions, family law toward long multi-session research — but the cluster architecture underneath is the same.

• One pillar page per practice area, 2,500+ words, answer-first sections
• 8–12 supporting posts, each fully answering exactly one client question
• Source questions from intake calls and consult transcripts, not keyword tools
• Update pillars every six months with visible dateModified — freshness is retrievable
• Win one cluster completely before opening a second

Entity building: how AI models learn your firm exists

Your website teaches AI engines what you know; the rest of the web teaches them that you exist. Language models and retrieval systems build their picture of "Smith & Rivera Family Law" from every place the firm appears: your state bar's public directory, Avvo, Justia, FindLaw, your Google Business Profile, local news quotes, university alumni pages, court records, podcast show notes, Reddit threads where someone recommends you. When those sources agree — same firm name, same attorneys, same address, same practice focus — the model forms a confident entity it can safely name in an answer. When they conflict, the safest move for the model is to omit you, and it does.

The work is unglamorous and decisive. Claim and complete every major directory profile with identical name, address, and phone — exact, down to the suite number. Keep your Google Business Profile compliant and complete, because it's a high-trust structured source every engine consumes. Make each attorney's bio page substantive: bar number, admissions, education, notable matters within bar advertising rules, and consistent with their bar profile. Then pursue the sources that feed training data disproportionately — quotes in local news coverage, your bar association's publications, American Bar Association section involvement, and genuine community presence that produces organic third-party mentions. Reviews matter here too, not just for the map pack: volume and specificity of Google reviews are corroborating evidence that the entity is real, active, and well regarded.

How a firm earns its first AI mention: an 8–12 week walkthrough

Here's how this plays out in practice, drawn as a composite from firms we've audited through CaseGap — a three-attorney family law firm in a mid-size Southern metro, decent traditional rankings, zero AI presence. Weeks 1–2: baseline and entity triage. We test forty client-intent prompts across ChatGPT, Perplexity, and Google AI Overviews: zero mentions, zero citations, and one engine confuses the firm with a similarly named practice two states away. The firm fixes NAP inconsistencies across eleven directories, claims two unclaimed profiles, rewrites three thin attorney bios, and adds LegalService, Person, and sameAs markup sitewide. None of this is creative work; all of it is load-bearing.

Weeks 3–6: content restructuring. The firm's custody pillar ranks #6 but is never cited — the answers are buried under preamble. Each section is rebuilt to open with a complete, liftable 60–150 word answer citing the relevant Family Code section, and a real FAQ block with markup goes on every pillar. Weeks 7–12: the first signals. Perplexity cites the rebuilt custody page in week eight — it indexes fastest and is usually the first scoreboard to move. An AI Overview citation on a long-tail custody-modification query follows around week ten, because the page already ranked and now extracts cleanly. By week twelve, ChatGPT's search mode names the firm in response to "family law attorneys in [city] who handle relocation cases." Unprompted parametric mentions come later, on model-release timelines. The pattern is consistent: Perplexity first, Overviews second, ChatGPT search third — and the firms that fail are the ones that skip the entity triage and start with content.

How to measure AI visibility

Manual measurement is tedious but legitimate, and every firm should do a baseline before spending anything. Build a panel of 30–50 prompts from real client intent — pull them from intake calls, not keyword tools. Mix three types: informational ("how is property divided in an Ohio divorce"), evaluative ("best employment lawyers in Columbus for severance review"), and direct-hire ("I was just fired while pregnant in Columbus, which firm should I call"). Run each prompt in ChatGPT, Perplexity, and Google (watching for AI Overviews), in fresh sessions with no logged-in history contaminating results. Record three numbers: mention rate (is your firm named), citation rate (is your site linked), and accuracy (does the engine describe your firm correctly — wrong practice areas and stale addresses are common and damaging).

Two methodological warnings from building this tooling. First, these systems are stochastic: the same prompt can produce different answers minutes apart, so run each prompt at least three times and score the aggregate, or you'll chase noise. Second, re-test monthly, because engine behavior shifts with model updates and index refreshes — a citation you earned in March can silently vanish in May. This is exactly the drudgery worth automating: CaseGap's AI-visibility engine queries the live engines with real client-intent prompts for your practice areas and metro, scores mention and citation rates against firms in your market, and tracks the trend over time. If you want the baseline without building a spreadsheet, run a free audit — the AI-visibility checks across ChatGPT, Perplexity, and AI Overviews are included, and most firms are genuinely startled by their zero.

The first-mover window: why 2026 is the year to do this

I'll make the contrarian case plainly: AI visibility in 2026 is what local SEO was in 2009 — a channel where the clients have already arrived and the competitors haven't. When we run market-level audits, fewer than one in twenty small and mid-size firms shows any deliberate AI-visibility work: no answer-first structure, no validated schema, no entity hygiene, no measurement. The big legal marketing agencies are still selling the same SEO retainers; the directories are optimizing for themselves, not for you. Meanwhile the answer slots are being allocated right now, query by query, to whoever happens to be least disorganized.

The window matters because this channel compounds in a way that punishes late arrivals. Cited content attracts links and references, which strengthen the very authority signals that earn the next citation. Mentions in AI answers generate branded searches and reviews, which feed the entity layer. And content that circulates today is the training data of next year's models — the firms named in 2026's web corpus are disproportionately the firms 2027's models will "remember" unprompted. An AI answer names three firms where a results page listed ten; when those three slots in your metro settle, displacing an incumbent will cost far more than claiming an open slot costs today. The work in this guide is eight to twelve weeks of mostly unglamorous effort. The firms that do it this year will spend years being the answer.