What is a knowledge graph?
A knowledge graph is a map of things and the named links between them, such as “Nike is a brand” or “Nike makes running shoes.” Each link is one fact. Software can follow the links to answer questions, check facts and show where an answer came from.
AI Knowledge · Guide
Updated October 25, 2026
Key takeaways
A knowledge graph stores facts as things, called nodes, joined by named links, called edges.
Its rules, called an ontology or schema, say what kinds of things exist and how they may link.
Knowledge graphs let AI find facts, check them and cite its sources.
In neuro-symbolic AI, the knowledge graph is the symbolic memory the reasoning stage works from.
How a knowledge graph stores facts
Every fact is a triple: a subject, a named link and an object. Put millions of triples together and they form a graph.
Nike – is a – brand.
Nike – makes – running shoes.
running shoes – are a kind of – footwear.
footwear – is sold in – shoe stores.
Nodes and edges
The things are called nodes and the named links are called edges. Because every edge has a name, software knows what each connection means, not only that one exists.
The two layers of a knowledge graph
Every knowledge graph has a set of rules and a set of facts that follow them.
The rules: an ontology
The ontology, or schema, defines the kinds of things that exist, such as brand, product or place, and which links are allowed between them.
The facts
The facts are the actual nodes and edges that follow those rules, such as Nike, running shoes and the link “makes” between them.
Knowledge graph vs database vs taxonomy
All three organize information. They differ in what they connect and how easily they change.
Shape: Relational database: Tables of rows and columns. Taxonomy: A tree of parent and child categories. Knowledge graph: A network of things and named links.
Kinds of links: Relational database: Fixed joins between tables. Taxonomy: One: “is a kind of”. Knowledge graph: Many named links, such as makes, sold in or located in.
Adding a new kind of fact: Relational database: Change the table design. Taxonomy: Add a branch to the tree. Knowledge graph: Add a new link type.
Best for: Relational database: Transactions and reports. Taxonomy: Browsing and classifying. Knowledge graph: Connected questions and reasoning.
How a knowledge graph is built and used
Most knowledge graphs go through the same four steps.
Define the ontology
Decide which kinds of things and which links the graph will hold.
Add facts
Load facts from databases and documents, extract them with AI, or have experts add them by hand.
Connect and check
Merge duplicates, such as “Nike” and “Nike, Inc.”, and check new facts against the rules.
Query and reason
Follow links to answer questions, infer new facts and show where each answer came from.
Knowledge graph examples
Knowledge graphs already sit behind products people use every day.
Search engines
Google’s Knowledge Graph powers the information panels beside search results.
Open knowledge
Wikidata is a free, public knowledge graph that anyone can read and edit.
Shopping
Retailers link products, brands, attributes and categories so search and recommendations understand what a shopper wants.
Healthcare and science
Medical knowledge graphs link diseases, drugs, genes and symptoms to support research and diagnosis.
How AI uses knowledge graphs
Knowledge graphs give AI models facts they can check.
Grounding answers
A large language model can look facts up in a knowledge graph before it answers, which cuts down on made-up answers.
GraphRAG
Retrieval-augmented generation that follows links in a graph, instead of only matching similar text, can answer questions that need several connected facts.
Neuro-symbolic reasoning
In neuro-symbolic AI, a knowledge graph is the symbolic memory that the reasoning stage checks its conclusions against.
Explaining results
Because every fact is a named link, a system can show the exact path of facts that led to its answer.
Where ADG Precision fits
Most knowledge graphs describe the world: things and the facts about them. ADG Precision’s Intent Graph describes the request. It reads what someone asks for, from a search term to a spoken command, into offerings, brands, places, questions, intents and interests, and checks the action before anything runs.
The two work together. A knowledge graph finds the facts, and ADG works out what was asked and whether the action should run.
Knowledge graph FAQ
Short answers to the questions people ask most.
What is the difference between a knowledge graph and an ontology?
The ontology is the set of rules: which kinds of things exist and how they may link. The knowledge graph is the ontology plus the actual facts that follow those rules.
Is a knowledge graph a database?
A knowledge graph is usually stored in a graph database, but the term describes the content: facts joined by named links. A relational database can hold the same facts in tables, but questions that hop across many connections are harder to ask.
What is an example of a knowledge graph?
Google’s Knowledge Graph, which fills the information panels beside search results, and Wikidata, a free public knowledge graph, are the best-known examples.
How do knowledge graphs help large language models?
They give the model facts it can check. A system can look facts up in a knowledge graph before the model answers, or check the model’s answer against it, which reduces made-up answers and lets the system cite its sources.
What is a triple in a knowledge graph?
A triple is one fact written as subject, link and object, such as “Nike – makes – running shoes.” A knowledge graph is a large set of triples.
Further reading: Wikidata, the free knowledge graph; Google Knowledge Graph Search API.
Related guides
Keep reading in AI Knowledge.