Part of a series. See the hub: Buyers Ask ChatGPT Before They Hire a Lawyer.
The work of winning Citation Share inside AI engines is a stack. Five layers. Each layer compounds the next. Most law firms have built one or two layers. Almost no firm has built all five. The firms that complete the stack first will set the citation patterns the engines learn from. This is the playbook.
Layer 1 — Entity definition
The AI engine cannot cite what it cannot identify. The first layer is making the firm, its practice groups, and its attorneys legible as structured entities. Firm-level: a canonical entity definition with consistent name, founding year, practice areas, headquarters, leadership, and key identifiers across every page on the site and every off-site reference. Practice-group level: each practice group as a discrete entity with named partners, representative deals or matters, regulatory or industry context, and a clear taxonomy of subspecialties. Attorney level: each attorney as a structured entity — name, degree, bar admission, practice areas, named matters where ethically permissible, publications, speaking history, professional affiliations.
The work is schema, not copy. Each entity gets JSON-LD markup. Attorneys carry Person schema. Practice groups carry Organization or Service schema. The firm carries LegalService schema. The engines read this structure directly. Firms that ship it move Citation Share before they ship a single new piece of content.
Layer 2 — Third-party citation density
The engines weight breadth of citation across the source mix. A firm cited only in Law360 and The American Lawyer carries thinner authority than a firm cited in Law360, The Wall Street Journal, Bloomberg, Reuters, the firm's local business press, and one or two industry publications relevant to its client base. The work is earned-media placement outside the legal trades — pitching attorneys as named sources for stories adjacent to their practice, not directly inside it.
The mechanics: attorneys with bandwidth for monthly placements outside legal trades; a media-relations function calibrated to business and industry publications rather than only to Law360; a willingness to comment on stories before they are filed, not after; and a habit of being quoted in entity-consistent framings ("M&A partner at [Firm]" repeated across twelve sources moves the engines; varied descriptions weaken the signal). Press releases distributed without earned-media follow-through do not move Citation Share. Earned placement does.
Layer 3 — Primary-source research
Original analysis that an AI engine can pull as a primary source is the highest-value content a law firm publishes. The forms: dated, sourced analyses of regulatory developments with attorney bylines; structured case-data analyses (enforcement-action breakdowns, deal-volume tracking, jury-verdict summaries); white papers with named research methodology and dated entries; industry-specific reports the firm publishes annually with versioned data.
The disqualifying form: the client alert. Client alerts are restatements of public information without original analysis, framed as marketing content. The engines do not cite client alerts. They cite original primary-source research because that is what their training data and retrieval systems are built to weight. Replacing client-alert volume with one-tenth the volume of original, sourced, attributed analysis moves Citation Share substantially within ninety days.
Layer 4 — Structured site architecture
The firm website is the engine's reference document on the firm. It is no longer the buyer's destination. The architecture that supports retrieval differs from the architecture that supported marketing. Attorney pages need credentialed structure (Person schema, expertise markup, publication lists with dated entries). Practice pages need entity-graph structure (Service schema, named subspecialties, deal or matter data where ethically disclosable, FAQ schema for buyer-intent questions). Insights or research pages need Article schema with named authors, dated publication, and clear source attribution. The site speed, internal-link graph, and crawlability all become retrieval factors.
The fix is technical and one-time. Most firms can rebuild the schema layer of an existing site within ninety days. Most firms have not done it because the marketing budget still sits inside the directory-and-brochure model.
Layer 5 — Refresh velocity
The engines reward currency and penalize stale. A firm that publishes monthly original analysis ranks higher in retrieval than a firm publishing quarterly with identical content quality. The mechanism: engines weight recency as a freshness signal because users penalize answers built on outdated information. A practice page last updated in 2023, an attorney bio last refreshed in 2024, a research report dated 2022 — each becomes a downgrade signal.
The work: a content-refresh cadence written into the editorial calendar. Attorney bios updated quarterly. Practice pages updated every six months with new matters, regulatory context, and team additions. Research and analysis published at a steady rhythm — weekly, biweekly, monthly — at minimum one piece per practice area per month. Stale content is a measurable drag on Citation Share. A refresh discipline reverses it.
The Citation Share audit
The stack is built against a baseline. The audit is the baseline. Methodology: define the firm's top 20 buyer-intent prompts ("best M&A firms in 2026," "law firms for SEC parallel proceedings," "top class action defense firms," and so on). Run each prompt against ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews. Score the firm on three dimensions per prompt: appearance frequency (did the firm get named), position (how prominently), and depth of context (was the firm described accurately, with the right practice areas, partners, and matters).
The output is a 100-point Citation Share score per practice area per engine — 1,500 data points across 20 prompts and five engines. The score becomes the baseline. The same audit run quarterly tracks the firm's progress as it builds the stack. The audit is also the diagnostic — it identifies which practice areas are losing the answer, which engines are most exposed, and which named competitors are taking the citation share.
Sequencing the work
Build in order. Layer 1 (entity definition) is the prerequisite — without it, the engines have nothing to compound. Layer 4 (structured site architecture) is the second priority, because it makes Layer 1 retrievable. Layers 2, 3, and 5 (third-party citation, primary-source research, refresh velocity) run in parallel after the foundation is in place. A firm that ships Layer 1 and Layer 4 in the first 90 days and begins Layers 2, 3, and 5 in parallel sees measurable Citation Share movement by month four. A firm that starts with content and skips schema sees movement that does not compound.
This is the AI Communications playbook for legal. The companion analyses — Big Law Is Losing the Answer and Chambers Doesn't Move the Engine — diagnose who is exposed and why the directory tier no longer carries the buyer. This piece is the build order. The firms that complete it in 2026 will own legal Citation Share for the rest of the decade.
What are the five layers of the GEO stack for law firms?
Entity definition (firm, practice groups, and attorneys as schema-clean structured entities); third-party citation density (earned placement outside the legal trades); primary-source research (original dated, sourced, attributed analysis); structured site architecture (Person, Service, Article, FAQ schema); refresh velocity (monthly cadence at minimum).
Which layer should a firm build first?
Layer 1 — entity definition. Without structured entity data, the engines have nothing to compound. Layer 4 — structured site architecture — runs second because it makes Layer 1 retrievable. The remaining layers run in parallel after the foundation is in place.
How long does it take to move Citation Share?
Firms that ship Layers 1 and 4 in the first 90 days and begin Layers 2, 3, and 5 in parallel typically see measurable movement by month four. The compounding accelerates from month six onward as third-party citations and original research accumulate.
Why don't client alerts move Citation Share?
Client alerts are restatements of public information framed as marketing content. AI engines weight original primary-source analysis with named attribution and dated entries. The disqualifying signal is the absence of original reasoning, not the length or topic of the piece.
What does a Citation Share audit measure?
Appearance frequency, position, and depth of context for the firm across the top 20 buyer-intent prompts on each of ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews — approximately 1,500 data points. The output is a baseline score per practice area per engine, tracked quarterly.
Is Generative Engine Optimization the same as SEO?
No. SEO optimizes for ranking on a search results page. GEO optimizes for inclusion inside the AI-generated answer buyers see before any results page. The two share underlying infrastructure but pursue different objectives.
