Knowledge graph

A knowledge graph represents information as entities, such as people, organisations, places, services or documents, connected by named relationships. Each entity has a stable identifier, so the same thing can be recognised wherever it appears. Each relationship has a type, such as “is provided by”, “is governed by” or “replaces”. People and software can then follow connections rather than only search for matching words.

Google popularised the term in 2012 when it launched its Knowledge Graph, describing a shift from matching keywords to understanding “things, not strings”. The panels of facts beside many search results draw on it.

A knowledge graph might link a parking permit service to the council that provides it, the guidance that explains it and the policy that governs it. Relational databases can represent such relationships too. A graph is a modelling choice, and its value depends on which questions the connections help answer.

Why it matters for content

Content teams often meet knowledge graphs without the name. Structured data such as schema.org markup describes entities and relationships on a web page, such as an organisation and its address or an event and its date, and search engines add those facts to their own graphs. A content model with references between types, linking an article to its author and topic, is a graph in miniature.

A taxonomy is a natural starting point. It already gives concepts agreed labels and a hierarchy. Adding identifiers and other relationship types, such as “related to” or “is a step in”, moves it towards a graph.

Content estates contain relationships that pages often hide. A knowledge graph can connect terms, owners, evidence, audience segments and related guidance across channels. That supports discovery, reuse and impact analysis when something changes.

Graphs need consistent identifiers, schemas and governance. If the same organisation appears under three names or relationships are vague, the structure becomes impressively connected confusion. Not every content problem needs a graph. They help most where relationships are numerous, important and hard to represent in a simple hierarchy.

How AI changes the picture

AI systems can use knowledge graphs to retrieve context, constrain answers and follow explicit relationships. An assistant could show which recorded policy a guidance item depends on. That makes a connection traceable when the system exposes it, but doesn’t establish that either item is current.

AI can also propose entities and links from documents, speeding up graph creation. Human review remains important because inferred relationships may be ambiguous or wrong.

Graphs and language models complement one another. The graph provides recorded claims and connections, while the model handles flexible language. Those claims still need reliable sources, and neither replaces content ownership.