For twenty years, search rewarded pages: you optimised a page, it ranked, people clicked. That still happens, but a second thing now happens alongside it. Before an AI assistant can recommend you, compare you or explain what you do, it first has to decide what you are. It builds a model of your business, and it builds that model from your website plus everything else the web says about you: reviews, maps listings, business directories, even job adverts.
Why this is suddenly concrete
This is not a vibe. In September 2023 Google filed a patent application, published in March 2025, called “Data extraction using LLMs” (WO2025063948A1). It describes an AI system that reads across a domain and synthesises what Google calls a “deep, holistic characterization” of the entity behind it. The filing is still pending, so it is a window into how Google approaches the problem rather than a confirmed live system, and we have written up its full reading, including its limits, in a companion substantiation note on citate.io.
The mechanics matter. The system reads many pages, in different formats, and does not need your markup to be tidy. It treats the result as “an interpretation of the extracted content rather than a verbatim duplication”, so it is forming a view, not quoting you. It pulls in third-party data to fill the picture, and the filing names maps data, job listings and business information by name. And it organises what it finds as a graph of relationships: your services, your products, your brand intent, and how they connect.
One example in the filing lands the whole point. The system analyses a law firm and flags a mismatch: the firm’s brand and advertising push corporate mergers and acquisitions, but most of its actual work is civil contract law. The system trusts its reconstruction over the firm’s self-presentation. If your brand says one thing and your wider footprint says another, the footprint can win. This is the engine underneath entity corroboration.
Two honest caveats, because overselling this would be the wrong move. The application is pending, not granted. And read in full, its stated use is substantially about generating and targeting digital components, which the filing defines as advertising or non-advertising content, rather than a declared search-ranking method. The mechanism is real and well described; the leap to “this is the recommendation algorithm” is an interpretation, ours included. We would rather you knew that than took the stronger claim on trust.
Your pages stop being only targets and become evidence
Once you see your site as input to an entity model, each page takes on a second job. A services page is no longer only a ranking target for a keyword; it is the system’s evidence for what you actually do. A case study is no longer only traffic; it is proof of experience in a specific area. A team page tells the system who stands behind the work. Reviews, directory listings and press feed reputation. None of this is content you have to invent. It is content you already have, now doing double duty.
The practical shift is to stop auditing pages one at a time and ask a single question about the whole footprint: if an AI rebuilt your business from your site, your reviews, your listings and every third-party mention, what would it say, and is that what you want it to say? That question surfaces gaps and contradictions that page-by-page work never reveals.
What to actually do
Make your story consistent across sources. The system reconciles many sources into one characterization, so contradictions cost you. The name, the core description, the primary service, the location and the positioning should read the same way on your site, your Google Business Profile, your directory entries, your social profiles and your recruitment pages. Identical wording is not the goal; a consistent, reconcilable picture is.
Decide the attributes you want owned, then evidence them. Pick the two or three attributes you genuinely want associated with your business, then make sure the evidence exists off your own site, not just on it. A claim of expertise is supported by case studies, citations, credentials and independent coverage, not by an adjective on your homepage.
Strengthen the relationships. The model is a graph, not a list. Make the connections legible: which services map to which audiences, which locations to which service areas, which people to which work. The clearer the relationships, the more confidently a system can place you and decide when to surface you.
Audit your entity footprint. This is the move most businesses have never made. An Entity Footprint Audit answers the rebuild question directly: it reconstructs what the web currently says about your business across all public sources, shows where that diverges from how you present yourself, and prioritises the gaps. It is where the law-firm mismatch in Google’s own example would have been caught before a machine caught it.
The two layers, so you know which problem you are solving
It helps to separate two moments, because the fixes differ. There is the training layer: how a model comes to know your brand at all, which is governed by what made it into the training data and is covered in How AI Learns Your Brand. And there is the runtime layer: how a system reconstructs and characterises your entity when an answer is being generated, which is what this filing and this guide are about. Corroboration is the currency in both. Off-site agreement is how you get remembered, and off-site agreement is how you get characterised correctly in the moment.
How this differs by business
For enterprise and B2B, the enemy is inconsistency: different departments, product pages, partner sites and recruiters often describe the company differently, and an incoherent identity becomes a weak characterization. For ecommerce, entities extend to products, so clear attributes, clean category relationships and reviews that reinforce who a product is for all feed how products are understood. For local services, much of this aligns with signals you already manage, services, reputation, sentiment and service-area consistency, with the addition that your site, your Google Business Profile and your third-party presence need to tell the same story.
One caution worth keeping
“Entity SEO” is the kind of phrase that gets turned into a buzzword and sold as a brand new discipline. It is not. Much of this is continuous with E-E-A-T, with knowledge-graph thinking, and with the consistency work good practitioners have done for years. What has changed is that the mechanism is now described in a primary source, which lets us replace assertion with evidence. Treat it as a sharper lens on familiar work, not a new religion, and be sceptical of anyone selling it as the latter.
