// Entity Recognition
Your Brand Isn’t a Graph Node Anymore
The technical difference between how Google's Knowledge Graph recognises your brand and how a generative AI system decides what to cite, and why tactics built for the first don’t transfer to the second.
Talk to us about your brand’s footprintThe Distinction
Two Completely Different Systems Called The Same Thing
For fifteen years, "entity SEO" meant one thing: getting resolved into Google's Knowledge Graph. A deterministic pipeline, mention detection, candidate generation, disambiguation against a curated knowledge base, that either resolved your brand to a specific graph node or it didn’t.
Generative AI retrieval works nothing like that. AI Overviews, ChatGPT Search, and Perplexity compute dense contextual embeddings over passages of text, not graph-node lookups, unless a tool explicitly queries a structured index. Your visibility depends on passage-level factual density and extraction salience, not whether you’ve been formally recognised as an entity anywhere.
Traditional Entity Linking
A deterministic pipeline: mention detection, candidate generation, then disambiguation against a curated knowledge base like Google's Knowledge Graph. Either resolves your brand to a specific graph node, or it doesn’t.
Generative Retrieval
Computes dense contextual embeddings over passages of text, not graph-node lookups. Visibility depends on passage-level factual density and extraction salience, not formal recognition anywhere.
Why It Matters
Why Traditional Entity SEO Tactics Don’t Carry Over
This distinction isn’t academic. It’s the reason a lot of "entity optimisation" advice being sold right now describes a system that no longer governs the outcome it’s being sold to influence. NAP consistency and Schema.org markup were built to feed the graph-lookup model. Neither has a documented path into how a generative model actually decides what to cite, we cover exactly why on our dedicated Structured Data page.
The practical risk: a business can be fully "recognised" in the old sense, consistent NAP, a live Knowledge Panel, clean schema, and still be functionally invisible to a model that never queries any of that data in the first place.
What Actually Works
What Actually Makes an AI System "Recognise" You
In the embeddings-based paradigm, recognition isn’t a one-time event, it’s a statistical property of how consistently and specifically your brand’s facts appear across the sources a model draws from when answering a query. There’s no database entry to claim. There’s only a body of text, across your site and others, that either reliably describes you the same way or doesn’t.
This is why the entity work that still matters is about consistency and factual density, not registration in any single database. We go deeper on the consistency side on our Cross-web Consistency page, and on the content side on our Content Architecture page.
Questions
Frequently Asked
Isn’t Google’s Knowledge Panel still an entity recognition system?
Yes, and it still matters, for classic Search and for local and business-listing surfaces that pull from it. But it’s a separate system from how AI Overviews, ChatGPT Search, or Perplexity decide what to cite. Having a Knowledge Panel doesn’t guarantee AI citation, and lacking one doesn’t rule it out.
Does this mean structured data and schema markup are useless for entity recognition?
Not for classic search, they still support rich results and Knowledge Graph association there. For AI citation specifically, no platform documents them as required. We cover this in more depth on our Structured Data page.
What’s the actual first step for improving entity recognition in AI search?
Auditing whether your brand’s core facts, what you do, who you serve, what makes you different, are stated consistently and specifically across your own site and the other sources models draw from. That consistency work is covered on our Cross-web Consistency page.
When does this not matter as much?
If your brand rarely appears in adjacent, comparable searches, a genuinely novel category with no real competitors being cited yet, there’s less for a model to get inconsistent about. This matters most once you’re one of several plausible answers to the same question.
See How Entity and Brand Optimisation Brings This Together
Entity recognition is one piece of a larger picture, alongside structured data and cross-web consistency. See the full approach on the Entity and Brand Optimisation page.
Back to Entity and Brand Optimisation