The AI Overview Takeover of Legal Search — And Why Firms Are Missing From It

Ask ChatGPT to recommend a top M&A lawyer in New York, and you'll get three names — none of them chosen by the firms themselves. Legal search has quietly changed hands, and most law firms haven't noticed.

Legal is now the most AI-Overview-saturated vertical in search. According to Semrush analysis, 78% of legal queries trigger a Google AI Overview — the highest rate of any industry. A separate Ahrefs study of 146 million SERPs found that question-style legal queries trigger AI Overviews 57.9% of the time.

Yet the firms themselves are largely missing from the answers. When users ask ChatGPT, Claude, Perplexity, or Google AI Mode for a lawyer recommendation, the citations overwhelmingly point to seven directories: Chambers, Legal 500, Super Lawyers, Best Lawyers, Martindale, Avvo, and Justia. In one test, a search for "Cravath, Swaine & Moore M&A expertise insights" returned four directory profiles in the top six results — directories beat one of America's most prestigious firms for its own practice area.

The stakes are hard to overstate:

• When an AI Overview appears, only 8% of users click a traditional result — a 47% CTR reduction versus SERPs without one.
• AI engines name only 1–3 firms per practice-area query, versus 10–15 firms in traditional rankings — a dynamic researchers call "recommendation compression."
• AI-referred prospects convert at 4.4x the rate of standard organic visitors, per recent industry research.

Compression creates winner-take-most dynamics. Once an engine names a firm for a category, it's more likely to name it again — recursive citation builds a moat around early movers while everyone else becomes invisible. Sophisticated buyers have adapted fastest: general counsel and family offices now pre-screen outside counsel with a prompt before they ever call a referral.

The irony is that reputation doesn't transfer. Prestige built through peer networks and paywalled rankings is invisible to language models, which rely on the public web. Firms that assume their standing will carry over are discovering it doesn't.

For firms (and the agencies advising them), the first step is knowing where you actually stand. Tools like AI search visibility trackers and Briefsy's AI-search visibility checks can reveal whether you appear in ChatGPT, Perplexity, or Gemini answers — or whether a directory is answering for you. You can't fix what you haven't measured, and in this window, the firms that measure first are the ones the engines learn to name.

What Actually Triggers a Citation: The Four-Stage Pipeline

Getting cited by an AI Overview isn't a single event — it's the end of a four-stage pipeline, and your firm can fail at any one of them without ever knowing why.

The process runs Discovery → Retrieval → Selection → Citation, and each stage is a separate gate. As Juris Digital explains, a page can be crawled but never retrieved, retrieved but never selected, and selected but never displayed. Critically, a citation is a displayed source event — not a "case" — so counting citations without diagnosing which earlier stage failed leads firms to fix the wrong problem.

Google's own documentation confirms the mechanics. AI Overviews use retrieval-augmented generation (RAG) and are, in Google's words, "rooted in our core Search ranking and quality systems" — meaning traditional SEO fundamentals remain the entry ticket. The system also uses "query fan-out," generating related queries to gather answer information, so the user's exact phrasing isn't the only query your content must satisfy.

Before any of that matters, your pages must clear three eligibility requirements:

• Indexed in Google with snippet eligibility
• Fully crawlable by Googlebot
• Included in generative AI features in Search Console

Google also explicitly rules out popular shortcuts: llms.txt files, special AI markup, and content chunking are not used by Search. Creating separate pages for every fan-out query variation actually violates Google's scaled content abuse policy.

The engines also behave very differently. Only 11% of domains are cited by both ChatGPT and Perplexity, per Semrush-cited analysis — visibility on one engine doesn't predict the other. Google AI Overviews correlate 93.67% with top-10 organic results, ChatGPT shows roughly 87% correlation with Bing's top 10, and Perplexity cites an average of 21.87 sources while rewarding recency.

For law firms, the practical implication is diagnostic: before investing in content, confirm which stage is actually failing. Tools like Briefsy can check whether your firm appears in AI answers across ChatGPT, Perplexity, and Gemini, so you know whether you're losing at discovery or at selection — and fix the right gate first.

What Doesn't Work — And What Google Says to Skip

Some of the loudest advice about "AI optimization" is advice Google itself has publicly debunked. Law firms are spending money on tactics that the search engine explicitly says it ignores — and in one case, tactics that can trigger a penalty.

The clearest example: llms.txt files and special AI markup do nothing for Google Search. Google's official AI optimization guide states plainly that you don't need to create new machine-readable files, AI text files, or markup to appear in generative AI search, because Google Search doesn't use them. The same documentation dismisses "content chunking" as unnecessary — there is no ideal page length — and says rewriting content specifically for AI is a wasted effort.

Worse, some vendors recommend creating a separate page for every query variation the AI might generate. Google's own guide warns this violates its scaled content abuse policy. Since AI Overviews use "query fan-out" — generating related queries to gather answer information — building a page per variation can look like spam, not strategy. Juris Digital's guidance points the other direction: build a coherent body of content around the real client decision, not one page per imagined query.

Then there's the statistics problem. Popular reports claim schema markup plus clear headings drives 2.8x higher citation rates, and that articles over 2,900 words are 59% more likely to be cited. But Juris Digital cautions that these correlations "do not prove causation" and should be treated as hypotheses, not proven causes. Google, for its part, states that no special schema is required for generative AI search. The vendors selling those numbers and the platform publishing the answers are telling you different things.

A practical checklist of what to skip:

• llms.txt files or any "AI-specific" markup — Google doesn't use them.
• Chunking or restructuring content for AI — no ideal page length exists.
• Mass-producing pages for every fan-out query variation — scaled content abuse risk.
• Inauthentic mentions across the web — Google's spam systems actively counteract them.

The honest takeaway: AI citation is built on the same fundamentals as good SEO — indexation, crawlability, unique content — plus verifiable third-party validation. What it doesn't require is a new stack of proprietary tricks.

That distinction matters when you're evaluating where to spend. Tools like Briefsy monitor your AI search visibility across ChatGPT, Perplexity, and Gemini alongside competitors and reviews, so you can see whether citations are actually happening — rather than guessing whether a vendor's tactic moved the needle.

The Real Citation Levers: Trust Signals, Directories, and Differentiated Content

The uncomfortable truth about AI citation is that the engines reward what other people say about you far more than what you say about yourself. Best Lawyers' analysis finds that AI engines surface lawyers based on peer-reviewed recognition, third-party validation, and whether claims are verifiable rather than self-promotional — and they penalize generic, duplicated attorney bios.

In practice, that means the levers look like this:

• Corroborated recognition: peer-reviewed rankings and mentions confirmed across multiple independent outlets, not one firm-authored page.
• Specific bios: practice-area detail, matter types, and verifiable credentials instead of interchangeable boilerplate.
• Direct answers: content that answers the real client question plainly, distinguishing verified experience from promotional adjectives.
• Editorial authority: bylined commentary in trade press that gives engines a citable, third-party-published voice.

This is also why prestige doesn't translate automatically. The "white-shoe visibility paradox" describes the most elite firms being under-represented in AI results precisely because their reputations traveled through closed networks — peer referrals and paywalled rankings that language models simply cannot see. A track record only produces AI signal when it's publicly reported. Meanwhile, seven directories (Chambers, Legal 500, Super Lawyers, Best Lawyers, Martindale, Avvo, Justia) dominate legal AI citations, with zero law-focused editorial sources appearing in top results in one report's testing.

Timing compounds the problem. AI engines surface only 1–3 named firms per practice-area query, and once an engine names a firm, it's more likely to name it again — a recursive-citation moat favoring early movers. The same research projects a roughly 24-month window before the cost of building AI citation authority escalates 50–80% compounded annually, though that urgency framing comes from a vendor with commercial interest and deserves a measured read.

The practical starting point is diagnosis: check which firms the major engines name for your practice areas today, and whether your own bios and claims are specific enough to survive the selection stage. Briefsy's AI-search visibility checks run that query across ChatGPT, Perplexity, and Gemini, then deliver a prioritized brief of what changed and what to fix next.

Your competitors' citation share is already forming. The question is whether yours is in the shortlist — or being written out of it.

Your Action Plan: Diagnose the Earliest Failed Stage and Monitor Continuously

Knowing which stage of the citation pipeline you're failing at changes everything about how you fix it. A firm that's crawled but never retrieved has a content problem; a firm that's retrieved but never cited has a trust problem — and treating one as the other wastes months.

Start with the diagnostic rule from Juris Digital: fix access before content expansion. That means confirming your pages are indexed with snippet eligibility, crawlable, and included in generative AI features in Search Console — the baseline Google says AI Overviews require. It also means unblocking the AI crawlers (GPTBot, ClaudeBot, PerplexityBot, and Google-Extended) in your robots.txt, a step the 5WPR/Haute Lawyer report flags as a common blind spot. Skip llms.txt files and AI-specific rewrites — Google's own documentation confirms it doesn't use them.

Once access is clean, work the selection stage:

• Claim and complete your profiles on the seven directories that dominate legal AI citations — Chambers, Legal 500, Super Lawyers, Best Lawyers, Martindale, Avvo, and Justia.
• Differentiate attorney bios: Best Lawyers' research finds generic, duplicated bios hurt surfacing, while specificity and verifiable claims help.
• Answer real client questions directly rather than building one thin page per imagined query variation — Google's query fan-out means the user's exact words aren't the only query involved.

Then measure. Track branded AI discovery through Search Console's Generative AI performance report and Bing's AI Performance reports, including Citation Share — with the caveat that grounding queries aren't a complete log, and a citation is a displayed source event, not a case. Remember that visibility doesn't transfer across engines: only 11% of domains are cited by both ChatGPT and Perplexity, so check each engine independently.

This is where continuous monitoring earns its keep. Because engines surface only 1–3 firms per practice-area query and recursive citation hardens early movers into recommended entities, a one-time audit goes stale fast. Tools like Briefsy run visibility checks across ChatGPT, Perplexity, and Gemini alongside competitor tracking, showing which stage your firm is failing and which rivals are winning the shortlist in your market.

The window matters. One report estimates roughly 24 months before the cost of building AI citation authority rises 50–80% compounded annually — a vendor figure, but the underlying logic holds: the firms inside the shortlist compound, and the firms outside it don't.