We ran an audit last month on a regional law firm that has held the top three Google positions for “personal injury attorney [their city]” for the better part of a decade. Their organic traffic is healthy. Their lead form converts. Their domain rating is 47.
We asked ChatGPT, Perplexity, Claude, and Gemini the same question their target buyer types: “best personal injury attorney in [their city].” The firm was cited zero times across the four platforms.
Their three competitors, ranked four through seven on Google, were cited fifteen times between them. One of those competitors, ranked seventh, was the single most-cited firm across all four AI platforms in that market.
The firm we audited had spent twelve years dominating Google. They had spent zero on the test the AI platforms are actually grading.
This is not a niche test anymore. BrightLocal’s 2026 Local Consumer Review Survey puts the share of consumers who trust AI platforms to recommend local businesses at 45%. The AI citation is the new third-party reference. A buyer asking ChatGPT for the best personal injury attorney in their city is doing the same job a buyer used to do with a Yelp star rating or a Google review average. If the firm does not appear in the answer, they did not exist in that decision.
The two engines work differently.
Traditional Google ranks pages. The AI platforms answer questions. The distinction matters because the page that ranks for a query is rarely the page that gets cited in an AI answer.
A Google-ranked page is optimized to be clicked. It has a sharp meta title, a hook in the meta description, internal links to deep funnel pages, and just enough body content to satisfy E-E-A-T. The model on the other end is figuring out which page is the best destination for that query. The user gets ten options. The user picks one. The user clicks through. The user converts on the page.
An AI-cited page is optimized to be quoted. It has a question in the H2, a literal answer in the next sentence, and enough surrounding context that the citation is defensible if the user clicks through to verify it. The model on the other end is figuring out which page is the best source for the assertion it is about to make. The user does not pick. The model picks. The user gets one answer, with three or four citations stacked underneath, and the user reads the answer instead of clicking through.
These are different jobs. The personal-injury firm above has excellent destination pages. They have approximately no source pages. Their entire site reads like the funnel the firm spent the last decade tuning: short benefit statements, lots of internal CTAs, a contact form on every page. None of it is built to be quoted.
What a source page looks like.
We rebuilt three of the firm’s service pages as source pages. The structural changes were small. The information-architecture changes were not.
The H2s are now questions. Not “Our Process” but “How long does a personal injury case in [state] take to settle?” Not “Why Choose Us” but “What does a personal injury attorney charge in [state]?” The H2 is the prompt the AI is matching against. The model literally searches for content where the heading aligns with the question being asked. Heading text that reads as marketing slogan (“Built On Trust”) does not match any real query. Heading text that reads as a buyer question matches dozens.
Each H2 is followed by a one-sentence literal answer. Then the conditions, the edge cases, and the legal nuance. The model needs the answer in the first sentence to be confident enough to cite it. If the page starts with “It depends…” the model moves on. A real example of the difference: “How long does a personal injury case in Texas take to settle? Most cases settle in six to fourteen months, with claims involving disputed liability or significant medical treatment running eighteen months or longer.” That sentence is citation-ready. “Every case is unique” is not.
Citation-grade evidence is in the prose, not in a sidebar. Statistics, state-bar references, case-result data, and the firm’s actual settlement ranges. Numbered, sourced, and inside the paragraph body where the model is reading. The model does not parse decorative callout boxes the way it parses prose. A pull quote in a separate <aside> element with no surrounding context is invisible. The same statistic embedded in a sentence that names the source is cited.
The schema is tightened. LegalService instead of generic Organization. Attorney schema for each named attorney with memberOf, award, and alumniOf populated. FAQPage schema for the answered questions, with the acceptedAnswer.text matching the literal answer in the body. The full schema architecture is documented in our editorial content service, where the FAQ and HowTo schema ships standard on every page we publish. Google’s Search Central documentation is explicit on this point: the structured data has to agree with the visible prose, and a field that contradicts the prose is worse than a missing field. Most AI platforms inherit that constraint and apply it harder, because they cannot rely on the user clicking through to spot the disagreement.
Platform-by-platform behavior.
The four major AI platforms differ in ways that matter when you are debugging why one platform cites you and three do not.
Perplexity crawls aggressively, refreshes citations weekly, and rewards pages with clean FAQPage schema heavily. It is the platform that moves first when a source page goes live. We typically see Perplexity citations inside two to four weeks of a rebuild.
ChatGPT search weighs entity consistency strictly. If your name, address, and phone number disagree between your site and Google Business Profile, you lose. If the schema says your firm is headquartered in Austin but the footer says Round Rock, you lose. ChatGPT search citation lag is six to eight weeks behind Perplexity.
Claude rewards authorship signals harder than the others. Anonymous pages get demoted aggressively. Pages with Person schema tied to a real sameAs URL (LinkedIn, bar association, credentialing body) get prioritized. Claude takes eight to twelve weeks to start citing a new source page.
Gemini leans on the same signal stack as Google itself. Pages that rank in Google’s AI Overviews tend to be cited in Gemini, with a roughly two-to-four-week lag. If your traditional SEO foundations are solid, Gemini follows. If they are not, Gemini lags every other platform.
We score all four platforms separately inside our AI visibility audit because the differences are diagnostic. A site cited by Perplexity but not ChatGPT search usually has an entity-consistency problem. A site cited by Gemini but not Claude usually has an authorship problem. A site cited by none has a structural problem.
What changed for the law firm.
Eight weeks after the rebuild, the firm was cited by Perplexity in eleven of the top fifteen prompts we track. ChatGPT cited them in nine. Claude in seven. Gemini in five.
Their Google traffic was flat to the prior eight weeks. The two systems didn’t trade. They stacked.
The firm now operates under both signal stacks together. The Google rankings hold the bottom-of-funnel queries (people who already know the firm and search by name). The AI citations capture the top-of-funnel queries (people researching attorneys for the first time, asking a chatbot to recommend three). The pipeline now has two distinct entry points instead of one.
We do the same work across other regulated verticals like medical and dental practices where citation rate matters as much as Google rank. Same architectural moves, calibrated to the schema rules each vertical follows.
What this is not.
The work is sometimes pitched as “GEO” (Generative Engine Optimization) or “AEO” (Answer Engine Optimization). The acronyms do not matter. What matters is whether you are writing pages that can be quoted, with schema that agrees with the prose, with authorship signals that real models recognize, served from a fast site that the AI crawlers can index efficiently.
It is not “add an FAQ to the bottom of your page.” A FAQ section bolted onto an existing destination page does not make the page citation-ready. The H2s on the rest of the page still read as marketing slogans. The schema is still wrong. The page weight is still 2.3 MB.
It is not “rewrite the meta descriptions.” Meta descriptions matter for Google. The AI platforms barely parse them. The work is upstream of the meta description, inside the content architecture and the schema entities.
It is not handled by a chatbot widget on your site. The chatbot does not speak with the model the buyer is using to find you. They are unrelated systems. The chatbot is a different layer of automation that handles inbound; it does nothing for whether the buyer finds you in the first place.
Common Questions.
Can our existing site be retrofitted, or does it need a full rebuild?
Both work in different cases. A site on a static or static-first architecture can usually be retrofitted with schema and content work alone. A site on a heavy CMS with custom theming usually rebuilds faster than it retrofits. The decision lives in the week-one audit.
How long until we see citation rate move?
Six to twelve weeks on the fastest platforms (Perplexity, ChatGPT search), twelve to twenty-four weeks on the slower ones (Claude, Gemini). The pattern is consistent across verticals.
Does this work for service-area businesses (HVAC, plumbing, roofing)?
Yes, often faster than for professional-services firms because most service-area competitors have ignored the work entirely. A well-optimized service-area site can pull citation rate from zero to thirty in a single quarter when the field is open. The visibility side pairs with local SEO as the same engagement.
What happens if a competitor copies our source pages?
They will. The platforms cite the page with the cleaner schema, the better authorship signals, and the older publish date. First-mover advantage on source pages is real and durable.
Does AI optimization replace Google SEO?
No. The two stack. We run them together inside a single national SEO engagement for clients with mixed local and national footprints, because the schema work and the citation work build on each other when they ship in sequence.
The audit.
If you want a per-page audit of where your site stands inside the AI answer engines, the free version at Lighthouse Local will tell you. The done-for-you tier rebuilds the pages we flag.
