What Is Knowledge Graph?

How the Knowledge Graph Works

A knowledge graph, in the general sense, stores information as a network: each real-world thing is a node, and each fact is an edge that connects two nodes. Google’s Knowledge Graph applies that shape at web scale. “Leonardo da Vinci” is a node; “painted” is an edge; “Mona Lisa” is another node. Store enough of those triples and you no longer have a pile of documents about a topic — you have a machine-readable model of the topic itself. Each node is an entity, and each edge is a semantic triple of subject, relationship, and object.

The step that makes this usable is identity. Every entity in the graph carries a unique machine identifier — a Knowledge Graph MID, written like /m/0cqt90 — that names the thing independently of the words people use for it. This is what lets Google treat “NYC,” “New York City,” and “the Big Apple” as the same place, and treat “Apple” the company as a different node from “apple” the fruit. Recognizing which entity a piece of text refers to is the job of named entity recognition; the graph is where the recognized entities and their facts live.

Google assembles the graph from public structured sources and from patterns extracted across the open web, then corroborates facts across many references before trusting them. Once an entity is in the graph, it surfaces in several places: it disambiguates ambiguous queries, it fills the knowledge panel beside relevant searches, and — increasingly — it supplies the factual scaffolding that AI Overviews and other answer engines use to keep generated text tied to real, identified things when they ground a response.

What the Knowledge Graph Stores

The graph is built from three kinds of information working together:

• Entities — the nodes: specific people, places, organizations, products, works, and concepts, each with a stable machine ID.
• Facts and relationships — the edges: attributes like a birth date or headquarters, and links like “founded by” or “located in,” expressed as subject–predicate–object triples.
• Types — a category assigned to each entity (person, place, organization, and so on) that constrains which facts and relationships make sense for it.

The distinction that matters is that the graph stores things and their connections, not pages. A single fact can be corroborated by thousands of documents, and a single document can contribute facts about many entities.

Example of the Knowledge Graph

The clearest documented illustration is the launch itself. On May 16, 2012, Google engineering SVP Amit Singhal published “Introducing the Knowledge Graph: things, not strings,” announcing that Search now understood real-world entities and their relationships rather than matching keyword strings. The launch numbers are specific and verifiable: the graph contained more than 500 million objects and more than 3.5 billion facts about and relationships between those objects, seeded from public sources including Freebase, Wikipedia, and the CIA World Factbook and then expanded at far larger scale.

The announcement framed the payoff as three concrete behaviors. First, find the right thing: a search for “Taj Mahal” could be disambiguated between the monument, the Grammy-winning musician, and the casino, because each is a distinct entity in the graph rather than a shared string. Second, get the best summary: Google could assemble the key facts about an entity in one panel — the post noted that the information shown for Tom Cruise already answered 37 percent of the next questions people asked about him. Third, go deeper and broader: because entities are linked, the graph could surface related things a searcher had not thought to ask for.

Those three behaviors — disambiguation, summarization, and connection — are exactly what an entity model buys you over a keyword index, and they are why the same structure now underpins AI-generated answers. When an assistant needs to be sure that “Mercury” means the planet and not the element or the Roman god, it is leaning on the same idea Google shipped in 2012: identify the thing first, then talk about it.

Frequently Asked Questions

The Bottom Line

The Knowledge Graph is Google’s map of the real world: a web of identified things joined by checkable facts, used to read a query as an idea rather than a spelling. It moved search from matching words to recognizing entities, and it now supplies the entity backbone that both classic results and AI-generated answers reach for when they need to know what a name actually refers to.