AI search visibility for law firms: how to get cited by ChatGPT, Perplexity and Google AI

The distinction from SEO is the output. A search engine returns a ranked list of pages; the user picks. A generative engine returns one answer that names a few sources; the engine picks. Everything a firm does for visibility now has to survive that second selection.

Demand is no longer hypothetical. In August 2026, 41.9% of surveyed consumers said they would use ChatGPT to research which lawyer to hire, up from 9% in 2023, while Google use fell to 71.9%. The heaviest AI adopters were 45 to 60 years old, the age group most likely to need estate, elder and business counsel (iLawyer Marketing survey).

Studies of citation behavior through 2026 agree on the pattern:

• Answer position. About 55% of cited passages come from the first 30% of a page. Bury the answer and it is not quoted.
• Mentions beat links. Across 75,000 brands, YouTube mentions correlated with AI visibility at 0.74, unlinked web mentions at 0.66, backlinks at 0.22 (Ahrefs).
• Earned sources dominate. Reddit, Wikipedia, YouTube, LinkedIn and major publishers supply the majority of citations; roughly 68% flow through about fourteen domains.
• Freshness. Content under three months old is cited roughly three times as often as older content.
• Not page one. 29.8% of domains cited in AI Overviews rank outside Google's top ten; only 38% of AIO citations come from the organic top ten.

The engines differ in how generous they are. Perplexity names brands in about 13% of answers; ChatGPT in under 1%. That makes every ChatGPT citation rare and valuable, and it means a firm should track all four engines rather than optimize for one.

1. 1Entity consistency. Identical name, address, phone and practice list on the State Bar, Google, Avvo, Martindale, Justia and Yelp. A mismatch reads as two different firms.
2. 2Schema. LegalService or ProfessionalService for the firm with sameAs links to every profile; Person for each attorney; FAQPage on question content; Article on guides. Google dropped the FAQ rich result, but the markup still helps extraction.
3. 3Answer-first content. Question-phrased H2s, a 40-to-60-word answer, then depth: tables, steps, costs, timelines. Named author bylines with credentials.
4. 4Reviews and mentions. Four to eight reviews a month across platforms, attorney profiles on LinkedIn with published articles, a YouTube presence even if small, bar-association and community-organization listings.
5. 5Crawler access. robots.txt allowing GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended. llms.txt is harmless but has no measurable effect; ship it and move on.

The structure we use on every page, including this one:

• A question as the H2.
• A direct answer of 40 to 60 words immediately under it, set apart so both readers and engines find it.
• Then the detail: lists, tables, examples, local specifics.
• An FAQ block of eight to ten questions with FAQPage schema.
• A visible 'last updated' date and a real author.

For Los Angeles firms, local specificity is the moat. Courthouse names, rent stabilization figures, county timelines and neighborhood references get cited over national content farms that cannot include them. See how this plays out per practice area and per sub-market.

Two cautions. AI answers vary run to run, so every query is run more than once and results are averaged. And citation rate is a leading indicator: it moves weeks before call volume does, which is exactly why it is worth tracking.

The practical rule: track all four, optimize the foundation once. The detail on each engine's source selection is in how AI Mode, AI Overviews and ChatGPT each pick a lawyer.

1. 1One `@id` per entity, reused on every page. Two ids for the firm reads as two firms.
2. 2`sameAs` only to profiles that carry the same name and phone. Fix the profile first, then link it.
3. 3Mark up only what is visible on the page. Engines compare markup to text.
4. 4Validate after every deploy; a broken JSON-LD block is silently ignored.

Copy-ready JSON for each type is in the schema markup guide, and the schema generator produces the firm and attorney blocks in the browser.

The directory consistency checklist lists the twenty listings to align first and the order to do them in.

The same page, described as an engine reads it:

• Entity. LegalService @id matches the Google, Avvo and State Bar profiles; knowsLanguage lists Armenian and English.
• Answer. The custody question and its 45-word answer sit in the first 20% of the HTML, server-rendered.
• Corroboration. Reviews mention custody outcomes at the Burbank courthouse; the attorney's LinkedIn article on the same topic links here.
• Freshness. dateModified within the last quarter; the visible 'last updated' line matches it.
• Local proof. Courthouse, neighborhoods (Adams Hill, Rossmoyne) and the Armenian community are named, not implied.

• Two versions of the firm. 'APC' on one profile, 'A Professional Corporation' on another, an old office on a third.
• Retrieval bots blocked by a CDN toggle or a security plugin. See which AI crawlers to allow.
• Client-side rendering. The practice page is an empty shell until JavaScript runs; most AI bots never run it.
• The answer on line forty. Three paragraphs of 'our firm was founded in 2009' before the question is addressed.
• No author. Content signed 'Admin' or unsigned. Engines and bar rules both want a responsible lawyer named.
• Stale dates. dateModified from 2022 on a page about 2026 rules.
• Fabricated review markup. AggregateRating from reviews the firm wrote itself; a penalty risk and a bar-rule problem.
• Template city pages. Twelve pages, one text, city name swapped. Zero citations for all twelve.

Every number above is on the pricing page with what each tier excludes.