How ChatGPT and Perplexity Choose Which Attorneys to Recommend

When a potential client opens ChatGPT and types "recommend a good estate planning attorney in Houston," something deliberate happens on the other side of the screen. The model does not guess. It does not randomly select a firm from a database. It executes a retrieval process — drawing on training data, real-time web retrieval, and structured data signals — to select the entity it considers most authoritative and trustworthy for that query.

Understanding that selection process is not optional for law firms with serious client acquisition ambitions in 2025. It is the entire game.

How AI Recommendation Systems Work: The RAG Pipeline

Modern AI assistants like ChatGPT (with Browse), Perplexity, and Google Gemini generate attorney recommendations through a process called Retrieval-Augmented Generation (RAG). The model does not rely solely on its training data. It actively retrieves current web content, processes it, and synthesizes a response. The selection of which content gets retrieved determines which attorneys get recommended.

The RAG pipeline for a legal query like "recommend a personal injury attorney in Chicago" runs roughly as follows:

1. Query interpretation: The model identifies the user intent (find an attorney), the entity type (personal injury lawyer), and the geographic constraint (Chicago).
2. Retrieval: The model queries an index of crawled web content, looking for documents that match the query context. Documents with high entity relevance scores are prioritized.
3. Ranking: Retrieved documents are ranked by a combination of factors including entity authority, content structure, and source credibility.
4. Synthesis: The model synthesizes the top-ranked information into a natural language response, naming the entities (attorneys or firms) that appeared most authoritatively in the retrieved documents.

The critical insight is step 3: ranking by entity authority and content structure. This is not keyword matching. It is a trust score applied to legal entities and the content they produce.

What Is Entity Authority for a Law Firm?

Entity authority is the degree to which an AI engine or knowledge graph recognizes a law firm as a verified, trusted, clearly-defined legal entity. It is distinct from domain authority (a link-based metric) and from keyword rankings. A firm with high entity authority is unambiguously identified by AI systems as a specific, real organization providing specific legal services in a specific location.

Entity authority is built from four types of signals:

The Role of Schema Markup in AI Selection

Schema markup is the most controllable entity authority signal — and the most frequently missing from law firm websites. When an AI engine crawls a law firm's website without schema markup, it must infer what the organization does, who works there, and what legal services are offered. When schema markup is present, this information is explicitly declared in machine-readable format.

The difference in AI behavior is significant. An AI model processing a page with @type: "Attorney" schema and explicit knowsAbout: "Personal Injury Law" properties can confidently associate that entity with personal injury legal services. A page without schema requires the model to guess — and in competitive markets with multiple attorneys, the model will prefer the entities it can verify.

The four schema types a law firm must deploy, in order of priority:

1. Organization — establishes the firm as a named legal entity with URL, description, and sameAs links to authoritative directories
2. Attorney / Person — on every attorney biography page, declaring practice areas, bar admissions, and professional credentials
3. LegalService — on every practice area page, declaring the type of legal service offered, geographic area served, and pricing information where appropriate
4. FAQPage — on content pages, structured questions and answers that AI models directly extract for RAG responses

Topical Authority: Why AI Models Trust Some Firms and Not Others

Entity authority tells AI models who you are. Topical authority tells them what you know. For law firms, topical authority is built by consistently producing clear, well-structured, factually accurate content on specific legal subjects — and maintaining that content as the authoritative reference on those subjects.

AI retrieval models evaluate topical authority through what is sometimes called a content cluster architecture: a pillar page covering a broad topic (e.g., "Personal Injury Law in Texas") supported by cluster pages covering specific subtopics (e.g., "Car Accident Claims," "Medical Malpractice Timelines," "Wrongful Death Compensation"). This architecture signals to AI engines that the firm's website is a comprehensive, coherent authority on the subject — not a collection of disconnected keyword pages.

The content itself must be structured for AI retrieval. The RAG pipeline extracts content in atomic chunks — short, self-contained passages that directly answer a specific question. A 3,000-word practice area page written as continuous prose is largely useless for RAG retrieval. The same information organized under 10–12 H2 headers, each followed by a 40–60 word direct answer, is highly retrievable.

How Perplexity's Real-Time Retrieval Differs from ChatGPT

While ChatGPT (without Browse) relies primarily on training data with periodic updates, Perplexity performs real-time web retrieval for every query. This distinction affects optimization strategy.

For Perplexity optimization, the site must be crawlable by PerplexityBot in real time (check robots.txt), the content must be in static HTML (not JavaScript-rendered), and the pages most relevant to legal queries must be indexed and fast-loading. Perplexity's retrieval algorithm weights source credibility — sites with consistent publishing history, HTTPS, and structured data are retrieved more frequently than new or unstructured sites.

For ChatGPT (with Browse enabled), the principles are similar but entity recognition from training data plays a larger role. Firms that have been mentioned in high-authority sources that ChatGPT's training data includes — legal publications, bar association resources, legal news outlets — carry a significant advantage. This is why off-site entity building in credible legal publications is not optional for LLMO.

The Entity Relationship Map: Connecting Your Firm to What Clients Search

AI engines understand the world through entity relationships. For a law firm, the goal is to associate the firm's entity with the specific legal concepts, practice areas, and geographic markets that potential clients search for.

This is achieved through a combination of on-site and off-site signals:

• On-site: sameAs properties in JSON-LD linking the firm's entity to its Avvo profile, Martindale listing, LinkedIn company page, and state bar directory page
• On-site: knowsAbout and hasCredential properties on attorney schema declaring specific practice areas and bar admissions
• On-site: areaServed properties on LegalService schema specifying geographic jurisdiction
• Off-site: Guest contributions or citations in legal publications that use the firm's full name, practice area, and location consistently
• Off-site: Verified profiles in legal directories with complete, consistent information matching the on-site schema

The more consistently these entity relationships are declared across authoritative sources, the more confidently AI models will associate the firm with specific legal queries — and recommend it accordingly.

What the Firms Being Recommended Right Now Have in Common

Analysis of the law firms that consistently appear in ChatGPT and Perplexity recommendations for competitive legal practice areas reveals a consistent profile. These are not necessarily the largest firms or the firms with the highest domain authority. They share:

• Complete, validated schema markup on all attorney and practice area pages
• Explicit bot permissions in robots.txt for ChatGPT-User, PerplexityBot, and Google-Extended
• Content structured in atomic H2 sections with direct answer blocks
• Verified, consistent presence in Avvo, Martindale-Hubbell, Justia, and state bar directory
• A topical content cluster covering their primary practice area comprehensively
• Fast page load times (LCP under 2.5 seconds) ensuring crawlers retrieve and index pages reliably

None of these are exotic requirements. All of them are absent from the majority of law firm websites we have audited. The gap between where most firms are and where they need to be is large — but it is entirely bridgeable with a systematic implementation approach.

The Compounding Advantage of Moving First

AI recommendation authority does not reset monthly like an ad auction. It compounds. A firm that establishes entity authority, topical authority, and schema infrastructure in 2025 is building a citation history that future model training cycles will incorporate. The more an AI model has encountered a firm as the cited answer to a legal query, the more confidently it will recommend that firm going forward.

The competitive dynamic is stark: the first firm in a given practice area and market to establish strong AI search authority will progressively dominate AI recommendations in that space. Competitors entering 6, 12, or 18 months later will face a model already associating that practice area with a competitor's entity — and will need to invest significantly more to achieve the same result.