Knowledge Graph
A store of entities and the typed relationships between them — Google's since 2012, and the reason a machine can tell your company from a word that happens to spell it.
Karl-Gustav Kallasmaa, Founder & CEOLast updated A knowledge graph is a store of entities — people, places, organisations, works, events — held as nodes with typed relationships between them, so that a system can reason over the relationship rather than match the text. Google launched its Knowledge Graph on 16 May 2012 with the framing "things, not strings," describing "an intelligent model — in geek-speak, a 'graph' — that understands real-world entities and their relationships to one another." The announcement stated it then held more than 500 million objects and more than 3.5 billion facts about and relationships between them.
The distinction from an index is the whole point. An index answers "which documents contain this phrase." A graph answers "what is this thing, and what is it connected to" — which is a question no single document has to have answered for the system to produce it.
How entities get in, and how they are queried
Population is a corroboration process, not a submission process. Structured markup, authoritative reference sources, business listings, and consistent descriptions across independent sites all act as evidence that a given entity exists, has a given type, and has given attributes. No single one of them is decisive, which is why "we added schema markup and nothing happened" is the expected outcome rather than a bug: markup is a machine-readable assertion, and assertions are weighed against other sources.
Reading is more constrained than people assume. Google's Knowledge Graph Search API returns ranked entity matches typed with schema.org types — Person, Place, Organization, LocalBusiness, Book, Movie, Event and others. Google states in that same documentation that the API "is not suitable for use as a production-critical service," and directs customers with high QPS requirements to Cloud Enterprise Knowledge Graph. Anyone building a monitoring product on top of that endpoint is building on something its own publisher describes as non-production.
The vocabulary those types come from is maintained publicly: schema.org published version 30.0 on 2026-03-19, and its type list is the shared ground on which markup, graphs and validators all agree about what an "Organization" is.
Why this matters when the reader is a machine
The problem a knowledge graph solves for a brand is disambiguation, and disambiguation is where naive systems fail loudest.
A brand whose name is also an ordinary word is ambiguous to anything doing substring or lexical matching — the string appears constantly in text that has nothing to do with the company. An entity record is what allows a system to fix a referent before it does anything else: retrieve documents about this organisation, attribute a claim to this company, decide that two mentions in different sources refer to the same thing.
That matters upstream of citation. AI search systems select passages and then attribute them; if the entity is unresolved, both steps can go wrong in ways that look like a visibility problem and are actually an identity problem. Semantic search covers the retrieval half; source citation covers what happens to the attribution.
Failure modes
Assuming markup equals inclusion. Markup is a claim. Acceptance requires corroboration from sources you do not own, and there is no queue, no ticket, and no timeline.
Inconsistent naming across properties. A legal name in one place, a trading name in another, and a stylised name in a third give a graph three weak candidates instead of one strong entity.
Treating a knowledge panel as the graph. A panel is one rendering of graph data in one surface. Its absence is not proof of absence from the graph, and its presence is not proof that every fact in the graph about you is right.
Reading entity presence as a ranking signal. Being a resolved entity affects whether a system knows what you are. It does not, on its own, decide whether you get cited for a given question; a retrievable passage that answers the question still has to exist.
Building on the public API as infrastructure. Google itself says it is not suitable for production-critical use. Treat it as a lookup, not a dependency.
Frequently asked questions
What did Google say the Knowledge Graph was when it launched?
An intelligent model that understands real-world entities and their relationships — things, not strings — announced on 16 May 2012.
How large was it at launch?
More than 500 million objects and more than 3.5 billion facts and relationships, per the announcement.
Which types does the public API return?
schema.org types, including Person, Place, Organization, LocalBusiness, Book, Movie, TVSeries and Event.
Is being in a knowledge graph the same as being cited by an AI assistant?
No. It helps a system identify you correctly; it does not supply the passage an assistant needs to quote.
Terms related to Knowledge Graph
Retrieval by meaning rather than by matching strings, what it is genuinely better at, and the class of query where it reliably fails.
The Schema.org vocabulary and the JSON-LD you embed with it, what it is genuinely good for, and the limits of what it can make happen.
Search where a model composes the answer and fetches pages through named crawlers, rather than returning a ranked list of links for you to read.
How an AI answer attributes what it says to the pages it read, and why a citation is a distinct outcome from a click or a mention.