GEO for Law Firms: Get Cited in AI Answers

The all-site numbers are backdrop, not legal evidence: Ahrefs found an AI Overview correlated with a 58% lower clickthrough rate for the top-ranking page across 300,000 keywords [ev-263], and Gartner forecasts search volume falling 25% by 2026 as queries move to chatbots [ev-264]. Neither is legal-specific; What Is Generative Engine Optimization? is where both are measured. What makes legal separate is the intent mix underneath.

Legal queries split into three bands, and content only reaches one. The transactional band — a named practice, a city, a phone number — survives the answer layer, because a summary cannot be a firm; proximity and review substance decide it anyway. The convergent band is where the answer is the deliverable: how long a claim usually takes, whether a settlement figure is normal for this case type, what a statute actually requires. The conditional band — can I sue if, what are the penalties for — needs a branch, not a sentence, and a summary that strips the condition is worse than silence. The opportunity is the middle band: publish the jurisdiction, the instrument, and the qualifier inside the sentence that will be quoted.

Clio's framing is the useful one: AI systems pull from many sources and cite only a subset, so a firm can be named in an AI answer without ranking first, and one that ranks first may be passed over [ev-265]. AI Search Visibility Audit is a separate objective with its own method, not a side effect of Topic Authority. Generalist mechanics sit on Modern SEO.

Martindale-Avvo states that AI search models synthesize a firm's website, legal directories, client reviews, and news articles rather than matching keywords alone [ev-266] — a low-confidence vendor source, carried here as a lead to verify rather than a mechanism to cite. Its NAP advice lives in SEO for Family Lawyers and SEO for Personal Injury Lawyers. The increment is the assembly step: a profile built from scattered sources rewards a firm whose facts agree, turning a verifiable fact into a citation input and an unreviewed claim into a liability.

Google's rater guidelines (§3.4.1) let pages sharing first-hand life experience on clear YMYL topics count as high E-E-A-T when the content is trustworthy, safe, and consistent with expert consensus, while some YMYL advice must come from experts [ev-257]. A firm's case diary is the intake call that did not go as scripted — a legitimate E-E-A-T Standards asset if it stops at the boundary: describe what happened, state the general lesson, do not hand the reader the outcome they came for.

OpenAI documents that OAI-SearchBot controls whether a site surfaces in ChatGPT search answers while GPTBot controls whether crawled content may train its models, and that the two robots.txt settings are independent, with a change taking roughly 24 hours [ev-262]. Blocking one does not block the other, which makes "we block AI crawlers" half a decision. The choice needs a named owner and a written record — AI Crawl Control.

Google's people-first guidance asks whether content shows clear sourcing and background about the author, with a link to an author page or a site's About page [ev-256]. Vendors argue stricter: anonymous YMYL legal content is low-trust by default, the attorney's name and bar admission belong on the content rather than a general About page, and Local Schema Markup missing areaServed reads as incomplete [ev-268] — a low-confidence source making a reasonable argument. Google accepts the About page, so it never asks for the signature in the body text. What survives scrutiny is the mechanism, not the mandate: an answer layer lifts a passage, not a page, so every qualifier has to sit inside the span that gets lifted. What Is Generative Engine Optimization? develops that mechanism; the legal increment is which qualifiers cannot be separated from the claim.

Our two existing legal guides — SEO for Personal Injury Lawyers and SEO for Family Lawyers — already tell a firm to attribute every article to a named attorney with bar number and to run LegalService markup site-wide; this page does not repeat those instructions. It changes one thing — where the signature sits. Before: "Most settlements in this case type resolve in eight to fourteen months," with the admitting state, the practice name, and the qualifier living in the byline, the title, and the footer. After: "Colorado car-accident claims most often resolve in eight to fourteen months, and that range assumes a filed suit rather than a demand-letter settlement." The claim now travels with its jurisdiction and its assumption intact, so an assistant can lift it without inheriting a promise the page never made.

The site-wide shift belongs to Modern SEO in 2026.

The ABA's Comment on Model Rule 7.1 prohibits misleading truthful statements and spells out how: an accurate report of a lawyer's achievements may still mislead if it leads a reasonable person to expect the same results are available to other clients, and a disclaimer or qualifying language may preclude that finding [ev-258]. The exposure is rarely the invented number; it is the true number with its conditions stripped off, which is why the qualifier must live in the same sentence as the outcome.

The FTC's Policy Statement Regarding Advertising Substantiation reaffirms that advertisers need a reasonable basis for objective claims before the claim goes out, and that a firm's failure to possess and rely on one is an unfair and deceptive practice under Section 5 of the FTC Act [ev-260]. The operational reading is about sequence: substantiation is a precondition of publication, and hunting for support afterward inverts the standard.

The same standard governs a firm's statement about its own AI visibility: no deterministic assistant position can be promised, so a guarantee is false on its face before anyone checks the data. What survives scrutiny is the shape Rule 7.1 and the FTC policy statement describe: what was measured, over what sample, on what date. AI Search Visibility Audit is the discipline that audit belongs to, and the site's own bar is the same shape: every figure published here carries a named source and an access date, or it does not run.

The model rule is a floor; adopting states write their own comments. North Carolina's adopted Rule 7.1 carries a Comment restating that unsubstantiated claims about a lawyer's or law firm's services or fees, and unsubstantiated comparisons with other lawyers' fees, may be misleading if presented with specificity that implies they can be substantiated [ev-259]. A multi-state firm needs a claim matrix keyed to jurisdiction, each state's adopted text as the governing surface, so the page says which rule it answers under.

ABA Formal Opinion 512 (July 29, 2024) applies existing Model Rules — 1.1 competence, 1.6 confidentiality, 5.3 nonlawyer assistance — to lawyers' use of generative AI rather than creating new ones. That reading comes from a law library's summary, not the ABA's own text [ev-261]. There is no AI carve-out from competence: an output no attorney has read is not work product, it is a liability with a confident tone.

Rule 1.6 travels in the same summary and does not stop at drafting [ev-261]. A prompt carrying a client name, case number, exhibit excerpt, or settlement figure has moved client information into a third party's system — a crawler-access question, not only a drafting one. The workflow that survives an audit drafts on a sanitized record: de-identified or hypothetical patterns, no live identifiers, and an attorney reading the finished draft against the actual file. A fabricated citation is a claim wearing a URL.

"The statute of limitations for a car accident is two years" is not an answer. In any given state it is a defined period with exceptions, tolling provisions, and a different rule where a government entity is the defendant. Name the jurisdiction inside the answer sentence, so the qualifier survives extraction on its own.

Statutes, rules, and ethics opinions have versions, and a legal answer without a date is a claim with an expiry nobody wrote down. Cite the instrument with its jurisdiction and effective date, and record a review date so a stale answer reads as stale — the cheapest signal that the firm knows what a rule said in 2019 and what it says now.

Answer from the primary instrument first — rule text, statute, opinion — and add commentary as a labeled second reference, so a secondary summary never becomes a translation step between the answer and the authority.

What Is Generative Engine Optimization? owns the sampling protocol. The legal increment is what goes in the prompt list: intake questions a prospective client actually types, each run once per state the firm practises in, plus the surfaces a session starts from — a query that surfaces a Google AI Overview, a chatbot thread that began on a directory listing. Avvo's marketing page shows what happens without the method: it states that 86% of AI-generated legal answers cite its legal directory networks, but publishes no query sample, date range, or methodology, so it is a vendor claim rather than a measured benchmark [ev-267] — the exact shape of a number a firm should refuse to repeat. Treat visible drift as the deliverable, not a position you can promise.

This guide is distilled from a larger research-backed knowledge system. The XKnowLab SEO Vault contains 571 interconnected notes on SEO fundamentals, local search, industry playbooks, and content systems — Markdown + Obsidian.

Get the complete SEO Knowledge Vault →

• Google — Search Quality Rater Guidelines (official, 2025) — YMYL pages under the most scrutiny for page quality; trust singled out as especially important. (accessed 2026-09-29, ev-255)
• Google Search Central — Creating Helpful, Reliable, People-First Content (official) — E-E-A-T weighting on YMYL; expertise questions ask for author background. (accessed 2026-09-29, ev-256)
• Google — Rater Guidelines §3.4.1, "YMYL Topics" (official, 2025) — first-hand experience as high E-E-A-T when consistent with expert consensus. (accessed 2026-09-29, ev-257)
• ABA — Comment on Rule 7.1 (official) — misleading truthful statements, unjustified expectations, disclaimers. (accessed 2026-09-29, ev-258)
• North Carolina State Bar — Rule 7.1 (official) — an adopting state's own text and Comment as the compliance surface. (accessed 2026-09-29, ev-259)
• FTC — Policy Statement Regarding Advertising Substantiation (official) — reasonable basis before dissemination; Section 5. (accessed 2026-09-29, ev-260)
• UNC Law Library — ABA Formal Opinion 512 (expert source, 2025) — second-hand summary of the opinion, not the ABA text; issued 2024-07-29, applying Rules 1.1, 1.6, 5.3. (accessed 2026-09-29, ev-261)
• OpenAI — Overview of OpenAI Crawlers (official) — OAI-SearchBot versus GPTBot, independent robots.txt settings, ~24 hours. (accessed 2026-09-29, ev-262)
• Ahrefs — AI Overviews Reduce Clicks by 58% (primary research, 2026) — April 2025 study re-run on aggregated Search Console data, comparing December 2023 against December 2025; 300,000 keywords. (accessed 2026-09-29, ev-263)
• Gartner — search engine volume will drop 25% by 2026 (industry research, 2024) — forecast, not measurement, and not legal-specific. (accessed 2026-09-29, ev-264)
• Clio — GEO for Law Firms (industry research, 2026) — AI systems cite a subset of sources; citation is a second goal. (accessed 2026-09-29, ev-265)
• Martindale-Avvo — How AI Search Ranks Lawyers (industry research, 2026, low confidence) — synthesis across site, directories, reviews, news; NAP consistency. (accessed 2026-09-29, ev-266)
• Avvo — Lawyer Marketing (industry research) — 86% figure published without query sample, date range, or methodology. (accessed 2026-09-29, ev-267)
• Lexicon Legal Content — GEO for Law Firms: 2026 Guide (industry research, 2026, low confidence) — byline credentials and LegalService areaServed. (accessed 2026-09-29, ev-268)
• How we research, source, and correct these guides: XKnow research standards

• What Is Generative Engine Optimization?
• SEO for Family Lawyers
• SEO for Personal Injury Lawyers
• Modern SEO in 2026
• SEO Content Strategy