Entity-Based Ranking: Knowledge Graphs and Search

From Words to Things

For most of search's early history, a search engine treated a query as a collection of characters. It looked for pages containing those characters and ranked them by how often and how authoritatively those characters appeared. The engine had no idea whether "jaguar" referred to an animal, a car, or a football team. Context was a guess, not a certainty.

That changed when search engines began thinking in terms of entities and knowledge graphs. An entity is a distinct, identifiable thing in the world: a person, a place, an organization, a concept, a product. A knowledge graph is the structured map of how those entities relate to one another. Together, they allow a search engine to move from matching strings of text to understanding meaning.

What an Entity Actually Is

The word "entity" sounds abstract, but the idea is concrete. An entity is anything that can be uniquely identified and described with attributes. Marie Curie is an entity. Paris is an entity. The Nobel Prize is an entity. What makes them entities rather than just words is that each one has a stable identity independent of how it is spelled or phrased.

This stability matters enormously. "Marie Curie," "Madame Curie," and "the first woman to win a Nobel Prize" all refer to the same entity. A system that understands entities can recognize that these phrases point to the same thing, even though the words are completely different. A system that only matches keywords cannot.

Entities also carry attributes. Marie Curie has a birthdate, a nationality, a field of research, and a set of relationships to other entities: the University of Paris, the element polonium, her husband Pierre Curie. These attributes and relationships are the raw material of a knowledge graph.

How a Knowledge Graph Works

A knowledge graph is a structured database of entities and the relationships between them. Think of it as a vast web of connected facts. Each node in the web is an entity. Each edge connecting two nodes is a relationship: "was born in," "is a type of," "won," "is located in," "is the spouse of."

Google's Knowledge Graph, which the company introduced in 2012, is one of the most prominent examples. When someone searches for a well-known person, place, or thing, the information panel that often appears on the right side of results draws directly from this graph. The engine is not scraping a webpage in that moment. It is retrieving structured knowledge it already holds about that entity.

The power of a knowledge graph lies in inference. If the graph knows that Marie Curie won the Nobel Prize in Physics and that the Nobel Prize in Physics is awarded by the Royal Swedish Academy of Sciences, it can infer a relationship between Marie Curie and that institution even if no single document explicitly states it. The graph connects dots that text alone cannot.

Why Entities Change How Ranking Works

When a search engine ranks pages using keyword signals alone, it is essentially asking: does this document contain the right words? When it ranks using entity understanding, it asks a richer question: does this document demonstrate genuine knowledge about the entity the searcher is interested in?

These are fundamentally different questions. A page can contain every keyword associated with a topic while saying very little of substance. Conversely, a page can discuss an entity in depth using varied language, synonyms, and related concepts without ever repeating the exact keyword phrase. Entity-based ranking rewards the second kind of content and penalises the first.

This shift also affects how search engines evaluate authority. An entity-aware system does not just ask whether a page mentions a topic. It asks whether the author or the publishing site is itself a recognized entity with established credibility in a relevant domain. A cardiologist writing about heart disease carries a different entity signal than an anonymous blog post using the same keywords.

The Relationship Between Entities and Search Intent

Search intent and entity understanding are deeply connected. When someone types a query, they are usually asking about an entity or a relationship between entities, even if they do not use precise language. "Best coffee in Rome" is a query about the relationship between a category of product, a quality judgment, and a geographic entity.

A search engine that understands entities can disambiguate intent far more reliably. "Mercury" as a query could mean the planet, the element, the Roman god, the car brand, or the musician Freddie Mercury. Entity understanding, combined with signals about the searcher's context and history, allows the engine to resolve this ambiguity without requiring the user to be more specific.

This disambiguation is one of the reasons semantic search represents such a departure from earlier approaches. The engine is not just pattern-matching. It is reasoning about what the searcher most plausibly means.

How Entities Enter the Knowledge Graph

Not every person, place, or thing automatically becomes an entity in a search engine's knowledge graph. Entities earn their place through a combination of factors: how widely they are referenced across the web, whether structured data has been published about them, whether they appear in trusted reference sources, and whether multiple independent sources describe them consistently.

Wikipedia plays an outsized role in this process. Search engines treat it as a high-confidence source of entity definitions and relationships because it is widely cited, editorially maintained, and structured in ways that make machine reading straightforward. Wikidata, Wikipedia's sister project, provides machine-readable structured facts that feed directly into knowledge graphs.

Structured data markup on webpages also contributes. When a publisher marks up a webpage to indicate that it is about a specific person, organization, or event using standardized vocabulary, that signal helps a search engine confirm and enrich its understanding of an entity. The markup does not create the entity in the graph, but it reinforces and clarifies the connection between a page and an entity the graph already recognizes.

Entities, E-E-A-T, and Demonstrated Expertise

Google's quality evaluation framework places significant weight on Experience, Expertise, Authoritativeness, and Trustworthiness. Entity understanding is the mechanism through which the engine operationalises these concepts at scale.

When an author is a recognized entity with verifiable credentials, publications, and institutional affiliations, the engine can connect their content to a network of trust signals. When an organization is a recognized entity with a history, a physical presence, and consistent representation across authoritative sources, its content carries a different weight than content from an unidentifiable source.

This is why the question of author and brand entity signals matters so much in modern search. It is not about gaming a system. It is about whether the real-world credibility of a source is legible to a machine that thinks in terms of entities and relationships.

The Limits of Entity-Based Systems

Entity-based ranking is powerful, but it is not without constraints. Knowledge graphs are built from information that has already been published, verified, and structured. New entities, emerging concepts, and niche topics may not yet have sufficient representation in the graph for the engine to reason about them confidently.

There is also an inherent conservatism to knowledge graphs. Because they rely on consensus across multiple authoritative sources, genuinely novel ideas or contested claims may be handled poorly. The graph reflects what is known and agreed upon, not what is new or disputed.

Understanding these limits helps explain why search engines do not always behave as expected on emerging topics, specialized domains, or rapidly changing information landscapes. The entity layer is a strength for stable, well-documented knowledge and a constraint where that documentation does not yet exist.

A Different Way of Thinking About Relevance

Entity-based ranking invites a fundamentally different understanding of what relevance means in search. Relevance is no longer about the surface similarity between a query and a document. It is about whether a document genuinely contributes to a searcher's understanding of an entity or the relationships between entities they care about.

This shift has implications for how content earns its place in results. Depth, accuracy, and genuine subject-matter knowledge matter more than keyword density. Consistency of identity across the web matters more than isolated optimization signals. The question a search engine is implicitly asking has changed: not "does this page use the right words?" but "does this page, and the entity behind it, genuinely know what it is talking about?"

That is a harder question to fake and a more meaningful one to answer.