Entity Recognition (NER)

How AI identifies the who, what, and where in your content — and why schema is the cheat code.

// The Concept

Named Entity Recognition (NER) is the task of identifying and classifying named entities in text: people, organizations, locations, dates, products, concepts. Every search engine and every large language model runs some form of NER on your content. It's how Google builds the Knowledge Graph. It's how Bing determines that a page about "Apple" is about the company, not the fruit. It's how AI systems understand that "Green" on your site refers to a person, not a color.

NER is one of the oldest tasks in natural language processing, dating back to the MUC conferences of the 1990s. But the modern incarnation — powered by transformer models with billions of parameters — operates at a level of sophistication that earlier systems couldn't approach. Today's NER doesn't just find entities. It disambiguates them. It resolves coreferences. It links mentions to knowledge base entries. It understands that "the company," "Apple Inc.," "the Cupertino giant," and "Tim Cook's employer" all refer to the same entity.

For content creators and SEO practitioners, NER is the gateway to entity-based search. If AI systems can't identify your entities, they can't build knowledge about them. If they misidentify your entities — classifying your company name as a common noun, or conflating your founder with someone else who shares the same name — the downstream effects ripple through every AI system that processes your content. Getting NER right is the foundation of AI visibility.

And getting NER right, it turns out, is something you can directly influence. That's where schema markup enters the picture.

// How It Works

Modern NER systems use transformer-based models that examine each token in context. For every token in your content, the model produces a probability distribution across entity types: is this token part of a person name (PER), an organization (ORG), a location (LOC), a date (DATE), a miscellaneous entity (MISC), or not an entity at all (O)?

The critical insight is that confidence scores. In the example above, "Green" as a person name gets 0.72 confidence — the model isn't sure. "Novel Cognition" as an organization gets 0.68 on the first token. These marginal confidences mean that different NER systems, or the same system on different days, might classify these entities differently. That inconsistency creates problems downstream: Google might not build a Knowledge Panel for your entity if its NER system isn't confident it's an entity at all.

Schema markup eliminates this uncertainty entirely. When you declare {"@type": "Person", "name": "Guerin Green"} in your JSON-LD, you're not suggesting to the NER system that "Green" might be a person. You're declaring it with machine-readable certainty. The confidence goes from 0.72 to effectively 1.0. No ambiguity. No guessing. No inconsistency across systems.

This is why the correlation between schema markup and AI citations is so strong. In analysis of ChatGPT-cited sources, 70.4% include Person schema. That's not because schema is a ranking factor in the traditional sense. It's because schema makes entity recognition trivially easy — and entities that AI systems can identify with certainty are entities they can cite with confidence.

// Why It Matters for Search

NER is the gateway to entity SEO, and entity SEO is the foundation of AI visibility. The logic chain is direct: AI systems can only cite entities they can identify. They can only identify entities their NER systems recognize. Schema markup makes recognition certain. Therefore, schema markup is the single most impactful technical intervention for AI visibility.

But NER matters beyond just schema. The way you write about entities in your content affects how NER systems classify them. Consistent naming helps enormously. If you refer to your company as "Novel Cognition" on one page, "NovCog" on another, and "the agency" on a third, NER systems may treat these as three different entities. Use consistent, full entity names — especially in first references — and establish abbreviations explicitly before using them.

Co-occurrence patterns also influence NER. When two entities consistently appear together across multiple documents, NER systems learn to associate them. If "Guerin Green" and "Novel Cognition" co-occur in author bios, about pages, and article bylines across multiple domains, the NER system builds a strong association between person and organization. This is the entity graph at work — and it's why a Distributed Authority Network strategy, where the same entities appear with consistent markup across dozens of domains, creates such powerful entity signals.

Cross-domain entity consistency is the compounding mechanism. A single domain declaring Person schema builds a weak signal. Fourteen domains declaring the same Person schema with the same @id creates an entity that NER systems recognize with near-certainty. Each additional domain reinforces the entity's existence in the knowledge graph. The @id is the linchpin — it tells every system that encounters it: this is the same entity, not a new one.

// In Practice

Implement comprehensive schema on every page. Start with the foundation: Person schema for authors, Organization schema for your company, Article schema for content. These three types cover the core entities that AI systems need to understand your content's provenance and authority.

Use @id cross-references religiously. Every Person and Organization entity should have a canonical @id (e.g., https://novcog.com/#guerin-green) that remains identical across all pages and all domains. When you reference the author on a blog post, don't create a new Person object — reference the @id. When your entity appears on a partner site, use the same @id. This tells AI systems that the "Guerin Green" on fourteen different domains is the SAME entity, not fourteen different people who happen to share a name.

Implement sameAs links to authoritative profiles. Link your Person entity to your LinkedIn, GitHub, Google Scholar, and other verifiable profiles. These sameAs declarations give NER systems external anchors for entity disambiguation. When multiple authoritative sources confirm the same entity relationships, the NER confidence score approaches certainty.

For content writing, use clear entity introductions. The first time you mention an entity on a page, use its full name with enough context for unambiguous identification: "Guerin Green, founder of Novel Cognition and AI strategy consultant" rather than just "Green." This contextual framing gives NER systems the surrounding tokens they need to classify the entity correctly, even without schema. Then use schema to confirm what the NER system already suspects. The combination of good writing and comprehensive schema creates entity signals that are essentially impossible for AI systems to misinterpret.

Audit your existing content for entity consistency. Search your site for every mention of your key entities and check: are they consistently named? Do coreferences (pronouns, abbreviations, descriptions) resolve unambiguously? Are there pages where your entity appears without schema markup? Each inconsistency is a crack in your entity architecture — a place where NER systems might get confused and break the chain of association that links your mentions into a unified entity.