Knowledge graph

A knowledge graph stores facts as things and connections. The things, called entities or nodes, might be a company, a product, a feature, a person or a document. The connections, called edges, say how they relate: this product has this feature, this feature is priced on this plan, this claim comes from this document. Because each fact is a small, typed statement rather than a sentence buried in a page, a machine can look it up, combine it with others and follow it back to where it came from.

There are two meanings in marketing conversations. Google’s Knowledge Graph is the public one that powers knowledge panels in search. A company knowledge graph is the same structure built from one organisation’s own material, used to keep facts consistent and to ground AI-written content.

Search moved from matching words to understanding entities when Google launched its Knowledge Graph, and AI answer engines depend on that understanding even more. If an engine cannot tell your product apart from a similarly named one, it cannot describe or cite you accurately.

Inside a company, a graph solves a problem every content team has: the same fact written five ways in five documents. With one node for the price and one for each feature, every draft reads the same value, and a correction made once is corrected everywhere.

• Google introduced its Knowledge Graph in May 2012 under the phrase “things, not strings”.Google, Introducing the Knowledge Graph: things, not strings (May 2012), 15 Sept 2026
• >500 millionAt launch Google said it “currently contains more than 500 million objects, as well as more than 3.5 billion facts about and relationships between these different objects.”Google, Introducing the Knowledge Graph: things, not strings (May 2012), 15 Sept 2026
• Google’s Knowledge Graph Search API “uses standard schema.org types and is compliant with the JSON-LD specification”, and Google warns it “is not suitable for use as a production-critical service.”Google for Developers, Knowledge Graph Search API, 15 Sept 2026

1. 1

   Help Google’s graph recognise you: publish Organization structured data on your homepage with your official name, logo and the profiles you control, and keep that name identical everywhere it appears.

2. 2

   List the entities your content keeps mentioning: products, plans, features, integrations, people and proof points. Those are your first nodes.

3. 3

   Attach a source to every fact, so a node records not only that a plan includes a feature but which document says so and when it was last checked.

4. 4

   Extract automatically, then review. Extraction gets relationships wrong in predictable ways, such as confusing a planned feature with a shipped one.

5. 5

   Assign an owner to update the graph when sources change, and retire facts that are no longer true instead of leaving them beside the new ones.

• Treating the graph as a one-off project. A graph that is not maintained becomes a confident source of outdated facts.
• Building nodes without provenance. A fact with no source cannot be checked, so it cannot ground anything.
• Confusing a knowledge panel with ownership of the entity. Google decides what appears in its panel; your structured data and consistent naming influence it but do not control it.

Google’s Knowledge Graph

Google’s is a public graph about the world that Google builds and controls. A company knowledge graph is private, built from your own sources, and yours to correct.

A document library

A library stores documents; you retrieve pages. A graph stores facts and their relationships; you can ask a question that spans several documents and still trace each fact.

1. 1Google, Introducing the Knowledge Graph: things, not strings (May 2012) read 15 Sept 2026
2. 2Google for Developers, Knowledge Graph Search API read 15 Sept 2026