What is an ontology in AI?
In AI, an ontology is a shared definition of the kinds of things in a subject area, their properties and how they may relate. It is the rulebook behind a knowledge graph, and in neuro-symbolic AI it gives the reasoning stage the concepts and constraints it reasons with.
AI Knowledge · Guide
Updated October 25, 2026
Key takeaways
An ontology defines classes (kinds of things), their properties and the relations allowed between them for one subject area.
A knowledge graph is an ontology filled in with real facts.
In neuro-symbolic AI, perception maps what it sees onto ontology concepts, and reasoning checks the result against the ontology’s rules.
A good ontology is specific enough to reason with and simple enough to keep up to date.
The parts of an ontology
Most ontologies are built from four kinds of pieces.
Classes
The kinds of things that exist, such as Product, Brand or Store. Classes nest: running shoes are a kind of footwear.
Properties
The details each thing can have, such as a color, a price or a size.
Relations
The named links allowed between things, such as a brand makes a product or a store sells it.
Rules and constraints
Statements that must always hold, such as every product belongs to at least one category.
Ontology vs taxonomy vs knowledge graph
The three terms are related and often mixed up.
What it is: Taxonomy: A tree of categories. Ontology: The definitions and rules for a subject area. Knowledge graph: An ontology filled in with real facts.
Links: Taxonomy: One: “is a kind of”. Ontology: Many named relations, plus rules. Knowledge graph: The actual links between real things.
Example: Taxonomy: Footwear › Running shoes. Ontology: A Brand makes a Product. Knowledge graph: Nike makes running shoes.
Used for: Taxonomy: Browsing and classifying. Ontology: Shared meaning and reasoning. Knowledge graph: Answering questions and checking facts.
How neuro-symbolic AI uses ontologies
The ontology is the shared vocabulary between the neural part and the symbolic part.
Perception maps to concepts
The neural network turns raw input into ontology concepts. A photo becomes Cat, Mat and the relation on. A search becomes a Product, a Brand and a Place.
Reasoning applies the rules
The symbolic stage uses the ontology’s classes, relations and constraints to infer new facts and reject answers that break the rules.
Feedback refines both
When a result breaks a rule, the system corrects its perception or, with a person’s approval, extends the ontology with a new concept.
Why a shared vocabulary matters
Without an ontology, the neural part might call something a “sneaker” while the rules expect a “running shoe”. The ontology gives both parts the same names for the same things, so their outputs line up.
Well-known ontologies
Many fields share public ontologies so their data and AI systems can work together.
Schema.org
A shared vocabulary for web pages that search engines use to understand products, articles, events and businesses.
SNOMED CT
A clinical terminology that gives health systems one set of concepts and relations for diseases, findings and procedures.
Gene Ontology
Describes the functions of genes and proteins across species, so research results can be compared and combined.
Industry ontologies
Retail, finance and manufacturing teams build ontologies for their own products, accounts and parts.
How to build a useful ontology
An ontology only helps if people keep it accurate and AI can use it.
Start from real questions
List the questions the system must answer, and model only what those questions need.
Reuse before you invent
Start from a public ontology such as Schema.org where one fits, then add your own concepts.
Keep it versioned
Treat changes like code: review them, record who made them and keep old versions so past answers can still be explained.
Let AI propose, people approve
Neural models are good at suggesting new concepts and links. A person approving them keeps the ontology trustworthy.
Where ADG Precision fits
ADG Precision keeps a curated structure for the requests it reads: offerings from sector and industry down to the most specific niche, and the roles a word can play, such as brand, place, question, intent or interest. New concepts are proposed and then approved by people before they count.
That structure is what lets ADG check an action against rules before it reaches Google Ads or any other live system.
Ontology FAQ
Short answers to the questions people ask most.
What is an ontology in simple terms?
An ontology is an agreed list of the kinds of things in a subject, the details they have and how they can relate. It works as the dictionary and rulebook for a knowledge graph.
What is the difference between an ontology and a knowledge graph?
The ontology defines the kinds of things and the rules. The knowledge graph is that ontology filled in with real facts, such as specific brands, products and the links between them.
What is the difference between an ontology and a taxonomy?
A taxonomy is a tree with one kind of link, “is a kind of”. An ontology adds properties, many kinds of relations and rules, so a system can reason with it.
Why does neuro-symbolic AI need an ontology?
The ontology gives the neural and symbolic parts the same concepts. Perception maps what it sees onto those concepts, and reasoning applies the ontology’s rules to them.
What languages are ontologies written in?
Common standards are RDF and OWL from the W3C, often stored in a graph database and queried with SPARQL. Many teams also keep simpler ontologies in spreadsheets or code.
Further reading: Schema.org; W3C Web Ontology Language (OWL).
Related guides
Keep reading in AI Knowledge.