There's a quiet kind of news that's easy to miss in a week full of algorithm updates and AI Overview screenshots, and this is one of those weeks. Rich Sanger broke down a Google patent for Search Engine Land called "Data extraction using LLMs," and once you actually sit with what it describes, it reframes a question most of us in SEO and content marketing have been asking wrong for two decades.

We've spent years optimising pages through client conversations focused on title tags, headers, internal linking, and keyword strategy. All of that still matters. But Google's patent suggests something has shifted underneath all of it: the company isn't just trying to understand your webpages anymore. It's trying to understand you, the entity behind them, in something close to the way a person would.

That's a genuinely different target to aim for, and it's worth understanding properly before everyone else catches up.

01 Breaking Down Google's Patent

Filed in 2023 and recently surfaced by Sanger, the patent describes a system where an AI model - built on a large language model - pulls information from a business's website and then goes further: maps, business listings, job postings, reviews, social sentiment, whatever public data it can find. It doesn't just extract facts. It interprets them and builds what Google calls, in its own words, "a deep, holistic characterisation" of an entity.

That word, entity, covers more than businesses. Google's definition stretches to people, products, places, and even concepts. But the part that should make any content marketing or SEO person sit up is this: the system is explicitly designed to work even when information "has not been structured for parsing." In plain terms, it doesn't need you to format anything specially. It reads the mess of your actual online presence and tries to make sense of who you are anyway.

The process Google describes breaks down into five clear stages:

1
Identify the entity and its associated domain
2
Pull information from webpages tied to that domain
3
Interpret that information rather than just copying it
4
Extract relationships, attributes, reputation signals, and sentiment
5
Supplement with third-party data - maps, listings, reviews, press mentions

The output isn't a page summary. It's closer to a profile, organised into something Google calls a hierarchical graph, where your services, audience, locations, and reputation all connect to each other rather than sitting as isolated facts on isolated pages.

From the patent filing
"The system isn't memorising your homepage copy. It forms an interpretation of who you are and presents it in the most appropriate format for the context."

Google's own patent filing makes this concrete with a fictional example - a company it calls "Example Search Co." One version describes the brand as a flowing paragraph: simple, accessible, trustworthy, with a friendly tone and approachable marketing. A second version takes the exact same underlying ideas and reformats them as a clean bullet list of brand attributes. Same entity, same characteristics, two different output shapes.

02 Why This Patent Matters Now

Patents get filed by the thousand, and most go nowhere. Sanger is upfront about that, and so are we. But a patent tells you how a company is thinking, even when it never ships as a literal product - and the thinking here lines up with where AI search has clearly been heading for a while now.

Think about what an AI Overview, or AI Mode, or Gemini actually has to do before it recommends your business over a competitor's. It can't just find a page that mentions you. It has to decide who you are, what you're good at, whether people trust you, and whether you're actually the right fit for whatever the person asked. That's an entity-level judgement, not a page-level one.

Google's Knowledge Graph already does some of this. E-E-A-T pushed in a similar direction. This patent suggests the LLM layer is what ties it all together - building a single coherent picture of an entity from dozens of scattered sources rather than ranking individual pages on their own merits.

03 Your Website Is No Longer Enough

Here's the part that genuinely changes the brief for content marketing teams. A service page used to exist to rank for a service keyword. Job done. Under this model, that same page becomes one piece of evidence among many, feeding into a much bigger picture Google is quietly assembling about who you are.

Case studies
No longer just traffic bait - they're proof of actual experience feeding the entity profile.
Team pages
Tell the system who's behind the business - a key authority signal the patent reads.
Reviews
Feed directly into the reputation signal. The patent explicitly mentions extracting sentiment from reviews.
Press coverage
Reinforce - or contradict - whatever the system has already concluded about you from other sources.
The uncomfortable part
If your website says you're the trusted, established option, but your Google reviews and social sentiment tell a messier story, the patent suggests the AI system isn't obligated to take your word for it. It's synthesising across sources, and the sources don't always agree.

This is precisely why brand reputation management stops being a "nice to have" running alongside SEO and starts looking like a direct input into how AI search understands you in the first place. The reviews you've been treating as a customer service task are quietly doubling as training data for how Google's AI systems describe your business to other people.

04 What Businesses Can Do About It

The patent doesn't hand out an optimisation checklist - that's not what patents are for. But reading between the lines, a few practical moves stand out.

  1. Check whether your story actually matches across sources. Pull up your website, your Google Business Profile, your LinkedIn, a recent press mention, and a handful of reviews, side by side. Would an AI system reading all of that conclude the same thing about who you are? Inconsistency here is the exact kind of noise this patent's interpretation layer has to wade through.
  2. Decide what you actually want to be known for, then back it up. Trustworthy. Fast. Specialist. Whatever it is, the patent's own examples lean heavily on attributes like trust, innovation, and accessibility - not generic service descriptions. Vague positioning gives the system nothing solid to latch onto.
  3. Treat reviews and third-party mentions as part of the SEO job. The patent explicitly lists reputation and social sentiment as signals it extracts. That puts brand reputation management squarely inside the same strategy as technical SEO and content marketing, not in a separate silo.
  4. Make the relationships between things obvious. Which services connect to which audiences? Which locations serve which markets? The patent's hierarchical graph structure rewards businesses that make these connections explicit rather than scattering them across disconnected pages.

05 The B2B and Enterprise Blind Spot

B2B marketing teams should expect this to sting a bit - because B2B businesses are often the worst offenders here. Sales decks say one thing, the careers page says another, investor materials lean into a third angle entirely, and nobody's checked whether any of it agrees with what the company actually wants to be known for in the market.

Run this exercise on most B2B accounts, and you'll find at least three competing versions of the same company living in three different documents. If an AI system is synthesising an entity profile from all of those sources at once, that inconsistency stops being an internal communications quirk and starts actively working against you.

The same logic applies at scale for enterprise SEO managing multiple brands, regions, or business units - where the entity-consistency problem just multiplies with every additional layer of the organisation. Anyone running enterprise SEO across a portfolio of sites should treat this patent as a reason to audit entity consistency before the next quarterly review, not after.

06 The Road Ahead for SEO

None of this replaces page-level SEO. Keywords, structure, and technical performance - all of that work still matters and isn't going anywhere. What's changing is what sits on top of it.

Google increasingly seems to be asking a bigger question than "does this page answer the query," and it's closer to "do I actually understand this entity well enough to recommend it?"

The opportunity
The businesses paying attention to this shift now - whether that's through tighter content marketing, more deliberate enterprise SEO, or finally treating B2B marketing as one coordinated story instead of five disconnected departments - get a real head start on whoever notices later. Entity consistency is not yet on most agencies' standard checklist. That gap is, for the moment, an advantage.

07 Frequently Asked Questions

Is this Google patent already live in search results today?
There's no public confirmation that this exact system is in production. Patents describe how a company is exploring a problem, not a guaranteed feature. The value here is in understanding Google's direction, not assuming this specific system is already running.
Does fixing entity consistency mean rewriting all my website content?
Not necessarily. It often starts with auditing existing content marketing assets - your Google Business Profile, social bios, and press mentions - for contradictions, then fixing the gaps rather than starting from scratch. Most businesses find the inconsistency sits in a handful of high-visibility spots, not everywhere.
How is this different from traditional E-E-A-T advice?
E-E-A-T has always pointed toward trust and authority signals. This patent suggests those signals are increasingly synthesised by an LLM across many sources at once into a single entity profile, rather than evaluated page by page - which raises the stakes on cross-channel consistency specifically.
Why would a small local business need to care about a patent built for entity-scale understanding?
The patent's framework applies to any entity, including a single local business with one location and one Google Business Profile. The principle - consistent information across your website, listings, and reviews - scales down just as well as it scales up.
Should B2B marketing teams treat this differently from B2C?
The underlying problem - scattered and inconsistent messaging across departments and channels - tends to be worse on the B2B side, simply because B2B organisations usually have more disconnected content sources: sales materials, investor content, recruiting pages, and product documentation all built by different teams.
Does this patent change how enterprise SEO should be structured?
It's a strong argument for enterprise SEO to include a regular cross-channel consistency audit as a standing deliverable, not a one-off project - especially for organisations managing multiple brands, regions, or business units where inconsistency compounds with scale.