Google patent shifts SEO agency strategy toward entity understanding

A 2023 Google patent titled Data extraction using LLMs, published as WO2025063948A1, describes systems that identify a domain and an entity, then build a characterization of that entity from extracted information and cross-source interpretation. For agencies, the win is no longer just a better page position; it is a brand footprint that reads cleanly enough for Google to classify, connect, and trust it.

The patent puts entity understanding at the center

The patent describes AI that can build an understanding of businesses, brands, products, and other entities from websites and public data by extracting information, identifying relationships, and synthesizing a deep, holistic characterization. In WIPO PATENTSCOPE, the output is "third content that represents a characterization of the entity based on extracted content." In plain English, Google is modeling a brand from fragments across the web, not just from a single page.

That is why classic SEO silos are breaking down. A title tag, a few links, and a clean homepage do not settle the question of who a business is if the rest of the digital footprint is messy. Agencies now have to treat reviews, directory profiles, public mentions, category language, and entity relationships as part of search strategy, because that is the material AI systems will use to form the answer.

Brand consistency is now a technical requirement

If Google is trying to characterize an entity, inconsistency becomes a ranking problem and an understanding problem at the same time. A brand that uses one name on the site, another in directories, a third in review profiles, and vague category language in its bios is making the machine guess. Agencies should be tightening the same details across the client footprint: legal and trading names, service labels, location data, and the wording used to describe what the company does.

Organization structured data on the homepage can help Google better understand an organization’s administrative details and disambiguate it in search results. LocalBusiness structured data goes further, letting Google understand business hours, different departments within a business, reviews, and more. That structured layer helps AI map a business to the right entity profile instead of a lookalike.

Google’s own surfaces already reward this work

Google has also made the same point through Business Profile and Knowledge Graph products, not just patents. Once a business verifies ownership of a listing, it can provide or edit address, contact info, business type, and photos, and those details can surface in Search, Maps, and the knowledge panel. Knowledge panels are drawn from the Knowledge Graph, which Google launched in 2012 as a database of facts about people, places, and things. In 2020, Google put that graph at more than 500 billion facts about five billion entities.

AI Overviews use a customized Gemini model working with existing Search systems, including ranking systems and the Knowledge Graph.

What agencies need to audit now

The agency playbook should start with a hard audit of whether the client’s footprint answers three questions clearly: who are you, why are you credible, and how are you distinct. If the answer is fuzzy, the brand is giving AI too much room to improvise.

Why this matters most for B2B and local brands

The message lands hardest for B2B and local companies because their buying cycles start before a sales call. If a prospect searches, sees a knowledge panel, checks reviews, and scans the site, the brand has already been interpreted by the machine. If Google misclassifies the company or fails to connect the right capabilities, that brand may never make the shortlist.