Guide · AI data modeling

Taxonomy vs ontology: a tree of terms, or a model of how things relate.

A taxonomy arranges terms from broad to narrow. An ontology defines classes, the relations between them and the rules they obey — formally enough for software to draw conclusions.

This guide places both on the ladder of knowledge organisation systems, shows the same insurance domain written as a SKOS taxonomy and as an OWL ontology, traces the standards behind each, and gives a decision guide — including where each one helps an AI system read documents and answer questions.

Updated October 2026 · 7 sources

Read the guide How Botminds does it

The ladder of semantic structure, richest first

Knowledge graphan ontology plus the real instances: this claim, this policy, this vehicle

instances

Ontologyclasses, properties, constraints and axioms with formal meaning

OWL

Thesaurushierarchy plus synonyms and “related term” links

SKOS

Taxonomyterms arranged broader to narrower

SKOShierarchy

Synonym ringterms treated as equivalent for search

equivalence

Controlled listone approved term per concept, no relations

pick list

What is the difference between a taxonomy and an ontology?

A taxonomy is a controlled vocabulary arranged in a hierarchy — each term has broader and narrower terms, such as Claims › Motor claim › Windscreen damage. An ontology is an explicit, formal specification of a domain: classes, the typed relationships between them, their properties and the rules they must satisfy. In taxonomy vs ontology, the taxonomy answers “what kind of thing is this?”; the ontology also answers “how does it relate to everything else, and what follows from that?”

Side by side

Taxonomy vs ontology, compared dimension by dimension.

Both organise knowledge. They differ in how many kinds of relationship they allow and how formally those relationships are defined.

hierarchy of terms

Taxonomy

A classification: terms placed under broader terms. Built for people to browse, tag, filter and find.

One main relation: broader / narrower
Labels and synonyms per concept
Easy to build and govern
Standard: W3C SKOS

formal domain model

Ontology

A specification of classes, properties and constraints with defined meaning. Built for software to check, integrate and infer.

Any number of named relations
Domains, ranges, cardinalities, axioms
Reasoners derive new facts
Standard: W3C OWL 2

What kind of thing is this?Core questionHow do things relate, and what follows?

Broader, narrower, relatedRelationshipsAny typed property: covers, filedUnder, ownedBy

Deliberately lightFormal semanticsFormal logic (description logic for OWL 2 DL)

Navigation, tagging, search facetsTypical useData integration, validation, inference

Lower: agreed terms and a hierarchyEffort to buildHigher: domain experts and modellers

Often the backbone of an ontologyRelation to the otherUsually contains a class hierarchy — a taxonomy

Worked example

The same insurance domain as a taxonomy and as an ontology.

Noy and McGuinness describe an ontology as classes, the properties of each class and the restrictions on those properties; add individual instances and you have a knowledge base. The taxonomy is the first of those parts.

An ontologyan explicit specification of a conceptualization — Gruber, 1993

ClassesClaim, Policy, Vehicle, Person

Class hierarchyMotorClaim is a kind of Claim — the taxonomy inside

Object propertiesrelations between things: filedUnder, coversVehicle

Data propertiesvalues on things: lossDate, sumInsured

Restrictionsevery MotorClaim is filed under some MotorPolicy

Individualsclaim 4471, policy P-2209 — the instance data

As a SKOS taxonomy (Turtle)

ex:Claims  a skos:Concept ;
  skos:prefLabel "Claims"@en .

ex:MotorClaim  a skos:Concept ;
  skos:prefLabel "Motor claim"@en ;
  skos:altLabel  "Auto claim"@en ;
  skos:broader   ex:Claims .

ex:Windscreen  a skos:Concept ;
  skos:prefLabel "Windscreen damage"@en ;
  skos:broader   ex:MotorClaim .

As an OWL ontology (Turtle, abridged)

ex:MotorClaim  a owl:Class ;
  rdfs:subClassOf ex:Claim .

ex:coversVehicle  a owl:ObjectProperty ;
  rdfs:domain ex:MotorPolicy ;
  rdfs:range  ex:Vehicle .

ex:MotorClaim  rdfs:subClassOf [
  a owl:Restriction ;
  owl:onProperty ex:filedUnder ;
  owl:someValuesFrom ex:MotorPolicy ] .

What the extra formality buys. State that home and motor policies are disjoint and that a claim is filed under exactly one policy, and a reasoner flags a motor claim filed under a home policy as inconsistent; any new subclass of MotorClaim inherits the restriction. OWL assumes an open world, so a missing fact is unknown rather than false. The SKOS version only says where “Windscreen damage” sits in the tree — exactly what a search facet needs.

The standards

The standards behind taxonomies and ontologies.

Both sides have mature, open standards. Knowing them saves inventing a format — and lets a vocabulary move between tools.

1993Ontology definedGruber: “an explicit specification of a conceptualization”.
Feb 2004OWLThe Web Ontology Language — now called OWL 1 — becomes a W3C Recommendation on 10 February.
2005ANSI/NISO Z39.19Guidelines for controlled vocabularies: lists, synonym rings, taxonomies, thesauri.
Aug 2009SKOSW3C Recommendation for sharing thesauri, classification schemes and taxonomies.
Oct 2009OWL 2W3C Recommendation, with the EL, QL and RL profiles for tractable reasoning; second edition December 2012.
2017SHACLW3C shapes language for validating graph data against constraints.

SKOS, OWL 2 and SHACL in one line each

SKOSConcepts with preferred, alternative and hidden labels; broader, narrower and related links; mappings between schemes. The specification says it is not a formal knowledge representation language.
OWL 2Classes, properties, individuals and data values with formally defined meaning, so reasoners can check consistency, subsumption and instance membership.
SHACLValidation: does this data satisfy these shapes? Often the practical choice when the goal is data quality rather than inference.

Decision guide

Taxonomy vs ontology: which one your system needs.

Start from what the system must do with the structure. Most organisations need a good taxonomy long before they need an ontology, and many never need inference at all.

What must the structure do?

Tag, browse and filter contentA taxonomy. Add synonyms (alternative labels) and it becomes a thesaurus that also improves search.

Classify documents with AIA taxonomy with a written definition per term. The definitions become the model’s instructions; the terms become its allowed answers.

Join data from many systemsAn ontology, or a lighter shared schema. Typed relations let records from different sources line up on the same entities.

Check rules or infer factsAn ontology with constraints — OWL for inference, SHACL for validation — and the instance data in a knowledge graph.

Scope with questions
Write the competency questions the model must answer — Noy and McGuinness’s first step.
Reuse first
Check existing vocabularies — schema.org, industry standards — before inventing terms.
Terms and hierarchy
Enumerate the terms, then arrange the classes. This alone is your taxonomy.
Properties and limits
Add relations and attributes, then their types and cardinalities.
Instances
Populate real individuals and test the competency questions against them.

For AI and LLMs

How taxonomies and ontologies help AI read documents.

Large language models already know a great deal of general vocabulary. What they do not know is your organisation’s exact categories and relations — that is what a taxonomy or ontology supplies.

Answering questions

Search facetsRetrieval is narrowed to the right category — only motor claims — before the model reads a word.facets

Grounded answers across entitiesRetrieval follows typed relations — this policy, its claims, their vehicles — before the model writes an answer.graph RAG

Reading documents

Classification targetsTerms become the allowed values in an extraction schema; definitions become its instructions.enums

Entities and rulesExtracted names link to the right class and instance, and constraints flag a value that breaks the domain’s rules.validation

TaxonomyOntology

Five habits that keep either one working

One kind of hierarchy per linkZ39.19 separates generic (is a kind of), instance and whole-part relations. A tree that mixes them breaks inheritance.
Stable identifiers, changeable labelsGive each concept an identifier that never changes; labels can then be renamed and translated.
A definition for every termA term without a scope note is read differently by every person — and every model.
Formality only where a question needs itFormal axioms nobody queries are cost without benefit. Grow the model from competency questions.
A named ownerVocabularies drift. Someone must approve new terms and retire old ones.

On Botminds

How Botminds puts a taxonomy to work in extraction.

On Botminds the taxonomy is the list of labels a collection reads from its documents — and each label’s description is the text the extraction agent is given, word for word.

structure

Labels as a taxonomy

Labels nest and group into categories, so a model of hundreds of fields stays readable.

Definitions as instructions

Edit a label’s description and you change what the AI reads — the scope note does real work.

rules

Types and choices

Five label types; a list label holds allowed values, and values outside a new list are reported for review.

tables

Record labels

Rows — invoice lines, payment schedules — with child labels as columns.

links

Relations between collections

Parent and child collections, and link fields that join one register to another.

integrity

One direction

Records flow parent to child, so the model stays free of loops.

See AI data modeling for the full design, and the Academy lesson on taxonomy, lifecycle, tags and events.

Questions people ask

Taxonomy vs ontology: common questions.

The technology behind itAI data modelingFrom documents to typed, searchable records

What is the difference between taxonomy and ontology?

In taxonomy vs ontology, a taxonomy arranges terms in a broader-to-narrower hierarchy, while an ontology formally defines classes, many kinds of typed relationship between them, their properties and the constraints they obey — so software can validate data and infer new facts.

Is a taxonomy a type of ontology?

It is often described as the simplest one. Most ontologies contain a class hierarchy, which is a taxonomy, but a taxonomy on its own has no typed relations or formal rules.

What is the difference between an ontology and a knowledge graph?

The ontology is the schema — classes, relations and rules. A knowledge graph is instance data organised by that schema: the actual customers, policies and claims, connected by the ontology’s relations.

What is an example of a taxonomy and an ontology?

Taxonomy: Claims › Motor claim › Windscreen damage. Ontology: a MotorClaim is a kind of Claim, is filed under exactly one MotorPolicy, and that policy covers one or more Vehicles — with rules a reasoner can check.

What is the difference between a taxonomy and a thesaurus?

A thesaurus adds equivalence (synonyms) and associative (“related term”) links to a taxonomy’s hierarchy. ANSI/NISO Z39.19 describes the progression from lists to synonym rings, taxonomies and thesauri.

What is the difference between SKOS and OWL?

SKOS is a W3C standard for sharing taxonomies and thesauri with light semantics. OWL 2 is a W3C ontology language with formally defined meaning, built for reasoning. SKOS itself is defined in OWL, and the two are often used together.

Do I need an ontology for AI?

Not always. Most document AI needs a well-defined taxonomy: the categories and fields to read, each with a precise definition. Ontologies earn their cost when an AI system must join many sources or obey rules about how entities relate.

Sources

See it on
your own work.

Bring one process and its documents. We’ll show you the agents, the controls and the record, running on your data.

Book a demo AI data modeling