Google’s LLM Patent Shifts SEO Toward Entity Identity

TL;DR: A 2023 Google patent describes how LLMs build a structured understanding of business entities by pulling from websites, reviews, job listings, maps data, and more — not just indexed pages. If this system influences search and AI recommendations, operators need to actively shape what Google understands about their brand, not just what their pages rank for. This is a structural shift in how SEO creates revenue outcomes.

From Ranking Pages to Modeling Entities

Google has spent over two decades helping users find documents. A query goes in, a ranked list of URLs comes out. That model works when users are browsing. It breaks down when users are asking AI to compare, recommend, or evaluate options on their behalf.

AI Overviews, AI Mode, Ask Maps, and Gemini do not return a list of links. They synthesize an answer. To do that, they need to understand the entities involved — the businesses, products, people, and services being evaluated. A patent filed by Google in 2023, titled “Data extraction using LLMs,” describes one possible mechanism for building that understanding.

The system does not simply extract text from a webpage. It collects information across multiple sources, processes it through a large language model, and generates what Google calls a “deep, holistic characterization” of an entity. The output is an interpretation — not a verbatim copy of what the site says.

That distinction matters for anyone running paid and organic acquisition at scale. If Google is forming conclusions about your business rather than just indexing your copy, then what your brand says and does across every public touchpoint directly affects how AI systems describe and recommend you.

How the Patent System Actually Works

The patent describes a four-stage process. First, the system identifies a domain and the entity associated with it. Second, it gathers content from webpages connected to that domain and processes it through an LLM. Third, the AI extracts attributes — services, reputation, social sentiment, relationships, principles, age, and geographic presence. Fourth, it supplements that picture with third-party data: Google Maps, job listings, business directories, and public reviews.

The output is not a page summary. Google’s filing includes example “entity summaries” that read like brand identity profiles — describing a hypothetical company’s positioning, tone, values, and differentiators as a synthesized narrative. The system also organizes these attributes into hierarchical graph structures, connecting entities to their services, audiences, use cases, and related concepts.

This is entity modeling, not content extraction. The question it answers is not “what does this page say?” It is “what do we understand about this business?” Those are operationally different questions, and they require operationally different responses from marketing teams.

Operators who want to understand where their brand’s entity footprint stands today should start with a structured brand and entity audit before making content or structural changes.

Webpages Become Evidence, Not Just Rankings Targets

Under the traditional SEO model, a service page targets a keyword. A location page targets a geographic market. A category page targets a product cluster. Those objectives remain important — but this patent suggests webpages now serve a second function.

Each page is evidence used to build or reinforce an entity model. A case study does not just attract organic traffic. It demonstrates experience in a specific domain. A team page helps identify the people behind the organization. Reviews contribute reputation data. Press mentions and industry citations either reinforce or challenge the system’s developing picture of who you are.

The patent explicitly states the system can extract information regardless of format — it is not limited to structured data or schema markup. That means unstructured content, social profiles, and third-party mentions are all inputs. Google is synthesizing a picture from everything it can find, not just what you’ve marked up for machine consumption.

For operators running multi-channel performance campaigns, this creates an alignment problem. If your ad copy, landing pages, Google Business Profile, and industry directory listings describe your business in subtly different ways, the entity model the AI constructs may not match the brand you intended to project.

What This Means for High-CAC Vertical Operators

In verticals where cost per acquisition runs high and purchase decisions involve research and comparison — forex trading platforms, online casinos, crypto exchanges, personal injury law firms — AI-driven recommendations are not a future concern. They are a current distribution channel.

When a user asks an AI system “which forex broker should I use for US clients” or “what’s the best online sportsbook in New Jersey,” the answer depends on how well each entity’s identity has been communicated to Google’s systems. A broker or operator who has invested only in keyword rankings may be well-understood as a webpage but poorly understood as an entity. That distinction determines who gets cited in an AI answer and who gets omitted.

For iGaming operators building brand authority in competitive state-by-state markets, consistency across licensing pages, affiliate mentions, app store profiles, and review platforms directly shapes the entity model. For forex acquisition programs, the entity signals that matter include regulatory certifications, trading platform reputation, and how consistently those attributes appear across broker review sites, social channels, and PR coverage.

The same logic applies to law firm marketing operations running mass tort or personal injury campaigns: attorney profiles, bar association listings, case outcome mentions, and client reviews are not separate SEO tasks. They are all inputs into a single entity model that determines whether an AI system recommends your firm when a user asks for help.

Four Operational Changes Operators Should Make Now

1. Audit your entity footprint across all public sources. Ask: if an AI had to describe your business using only publicly available information, what would it say? Run that exercise against your own website, Google Business Profile, industry directories, job listings, press coverage, and social accounts. Gaps and inconsistencies in that picture are gaps in your AI visibility.

2. Define the three to five attributes you want associated with your brand. The patent’s example entity summaries consistently organize businesses around a small set of core characteristics. If your brand has no defined attribute set — or if different teams are communicating different ones — the entity model AI builds will be diffuse and harder to surface in recommendations.

3. Back every claim with evidence. The patent describes building entity understanding from accumulated signals, not from assertions. “We are the most trusted broker” carries no weight without reviews, certifications, regulatory filings, and third-party citations that substantiate it. Supporting evidence is what allows the AI system to draw the conclusion you want it to draw.

4. Structure entity relationships explicitly. The hierarchical graph structures in the patent connect businesses to their services, service areas, audiences, and differentiators. Make those relationships easy to identify: which services you offer, which audiences you serve, which locations you operate in, which problems you solve. Audience-level targeting infrastructure and content architecture should reflect this same relational logic.

The Limits of What This Patent Tells Us

Patents are not product announcements. Google files thousands of patents annually, and many describe systems that never ship in recognizable form. Even when a patented system does make it into production, the implementation rarely mirrors the filing exactly.

What patents reliably reveal is how Google is thinking about a problem. This one shows Google is thinking about entity understanding as a prerequisite for AI-powered recommendation systems. That aligns directly with observable behavior in AI Overviews and AI Mode, where businesses are described, compared, and recommended — not simply linked to.

The practical implication is not to optimize for this patent specifically. The implication is to recognize that the signal environment for AI visibility is broader than a site crawl. It includes everything Google can find and interpret about your business from across the public web. Operators who build AI-native lead qualification systems and pair them with a coherent entity strategy will have an advantage as AI search matures.

Operators in crypto, igaming, legal, and forex are already navigating aggressive compliance constraints that limit what they can say in ads and on landing pages. Entity-level reputation signals — reviews, citations, PR, and consistent positioning across third-party platforms — may become the primary differentiator in AI-driven search surfaces where paid placement does not exist. Crypto acquisition teams building brand authority in a post-FTX trust environment understand this dynamic better than most.

The shift from page-level SEO to entity-level SEO is not a rebrand of existing work. It is a structural change in where the leverage lives. Your pages are now evidence. Your entity is the product.

Originally reported by Search Engine Land, June 2026.