A taxonomy sorts things into a hierarchy of broader and narrower terms, so it answers one question: what kind of thing is this? An ontology defines the types of things in a domain, the named relationships between them, and the rules a machine can reason over, so it answers how things connect and what follows from that. A taxonomy can become the backbone of an ontology, but a taxonomy on its own cannot say that a supplier ships a part or that a policy applies to a dataset.
How do an ontology and a taxonomy compare side by side?
The two come from different traditions. Taxonomies grew out of library and information science; ontologies grew out of computer and data science, a split that Heather Hedden of Enterprise Knowledge describes along with how the two have converged on shared W3C standards. The table sets them next to each other.
| Taxonomy | Ontology | |
|---|---|---|
| Question it answers | What kind of thing is this? | How do these things relate, and what follows? |
| Shape | A tree, or a polyhierarchy where a term has two parents | A graph of classes, properties and instances |
| Relationship types | One: broader and narrower | Many, each named and directional (supplies, part of, owns) |
| Rules | None | Axioms and constraints a reasoner can check |
| Main W3C standard | SKOS (2009) | OWL 2 (2012), validated with SHACL (2017) |
| Other standard | ANSI/NISO Z39.19 | Domain ontologies such as the Gene Ontology |
| Main consumer | People browsing, tagging and filtering | Queries, applications, reasoners and AI agents |
| Build cost | Lower: extending an existing taxonomy is easier than building an ontology from scratch | Higher: hard to build and slow to coordinate experts |
Read the table row by row and one pattern shows. Every cell on the right adds something to the cell on its left. Many organizations already have a taxonomy, so Hedden recommends extending it into an ontology rather than starting from scratch.
What exactly is a taxonomy?
A taxonomy is a controlled list of preferred terms arranged by hierarchy. The ANSI/NISO Z39.19 standard, reaffirmed in 2010, defines it as a controlled vocabulary of preferred terms that are all connected in a hierarchy or polyhierarchy. The only relationship a taxonomy needs is parent and child.
A retail product taxonomy shows the idea:
- Tools
- Power tools
- Cordless drills
- Impact drivers
- Hand tools
- Power tools
Each step down is narrower. “Cordless drills” is a kind of power tool, and that is all the structure says. It does not say which battery a drill takes or which store stocks it.
On the web, the W3C publishes taxonomies with the Simple Knowledge Organization System (SKOS), a Recommendation from 18 August 2009 that covers thesauri, taxonomies, classification schemes and subject heading systems. SKOS gives you skos:broader and skos:narrower for hierarchy and skos:related for loose association. The specification says plainly that SKOS is not a formal knowledge representation language, and that skos:broader is not transitive.
A thesaurus is the next step up. Z39.19 describes it as a vocabulary with explicit devices for synonyms, hierarchies and associative relationships, and ISO 25964-2:2013 covers how to map a thesaurus to other vocabularies.
What a taxonomy is good at: consistent tagging, faceted search and topic browsing. Heather Hedden of Enterprise Knowledge lists these as the core taxonomy uses, along with normalizing different names and languages for the same concept.
The AI Context Stack
A brief on the layers AI needs to understand enterprise data, from metadata and lineage to semantic definitions and ontology.
Get the AI Context StackWhat exactly is an ontology?
An ontology is a formal model of a domain: its classes of things, the typed relationships between them, their attributes, and the rules that hold. Thomas Gruber’s 1993 paper in Knowledge Acquisition 5(2) is the one usually cited for defining an ontology as a formal specification of a conceptualization. For the full component breakdown, see what an ontology is.
The W3C standard is the OWL 2 Web Ontology Language, a Recommendation whose second edition is dated 11 December 2012. OWL 2 ontologies hold classes, properties, individuals and data values, and its formal semantics let reasoners test class consistency, work out which classes sit inside others, and retrieve instances. SHACL, a W3C Recommendation from 20 July 2017, validates RDF graphs against a set of conditions.
Take the retail tools taxonomy (Power tools > Cordless drills) and add typed relationships:
- Cordless drill requires Battery
- Battery is supplied by Supplier
- Supplier operates in Region
- Store stocks Product
Now a query can follow the chain: which drills depend on a supplier in a region with a port delay? A taxonomy cannot answer that, because it has no edge for “supplied by.”
Real ontologies mix relationship types on purpose. The Gene Ontology uses is_a for subtypes, part_of and has_part for part-whole links, and regulates, positively regulates and negatively regulates for biological control. The is_a links alone would form a taxonomy. The other relations are what make it an ontology. Two more public ontologies show the range: schema.org for describing things on web pages, and FIBO, the Financial Industry Business Ontology, written in OWL.
Where does a taxonomy stop and an ontology start?
The line falls at two things: named relationships beyond broader and narrower, and rules a machine can apply. Once your model says “a drill requires a battery” and “every battery has exactly one supplier,” you have left taxonomy behind.
Here is the same drill modeled both ways, in Turtle:
@prefix ex: <http://example.org/> .
@prefix skos: <http://www.w3.org/2004/02/skos/core#> .
@prefix owl: <http://www.w3.org/2002/07/owl#> .
@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .
ex:CordlessDrills skos:broader ex:PowerTools .
ex:requires a owl:ObjectProperty ;
rdfs:domain ex:CordlessDrill ;
rdfs:range ex:Battery .
The first line is all SKOS can say: one concept sits under another. The second statement declares a typed relationship and the classes of its subject and object, which a reasoner uses to infer types.
The standards keep this line visible. The SKOS Primer explains that SKOS concepts are individuals in OWL terms, and that OWL-DL users who link OWL classes to SKOS concepts have to keep the two formally distinct. In practice, that means a “Power tools” concept in your taxonomy and a PowerTool class in your ontology are two different objects, linked on purpose.
Practitioners disagree on where a taxonomy sits relative to an ontology. Heather Hedden of Enterprise Knowledge sets out both views:
- The ontology is a semantic layer over separate vocabularies, and taxonomies stay outside it.
- The ontology includes every member of its classes, so taxonomies become a component inside it.
Either way, the taxonomy brings something the ontology lacks on its own: labels. SKOS supports alternative labels, labels in other languages and several kinds of notes. An ontology without a taxonomy or thesaurus has a hard time with synonyms and homonyms in search.
The reverse also holds. An ontology is a schema for meaning, separate from the data that fills it. The graph of actual instances is a knowledge graph, a distinction covered in ontology vs knowledge graph.
AI Agent Context Readiness Checklist
Check whether your glossary, relationships, and ownership are ready for AI agents to reason over.
Get the ChecklistWhen should you use a taxonomy, an ontology, or both?
Pick by the questions your users and systems need answered.
| If your problem is | Start with |
|---|---|
| The same thing is tagged five ways | A taxonomy and a controlled vocabulary |
| Search misses synonyms and acronyms | A thesaurus layer on the taxonomy |
| Questions cross two or more domains (products and suppliers, customers and policies) | An ontology |
| AI agents pick the wrong table or join | An ontology with typed relationships and rules |
Most teams should build in that order. Heather Hedden of Enterprise Knowledge recommends a bottom-up path: treat each existing taxonomy, facet or SKOS concept scheme as a candidate ontology class, then add verb-based relationships between the classes, such as Employee knows Subject or Event belongsTo Division. The taxonomy you already tagged content with keeps working, and the ontology inherits that tagging.
Earley Information Science gives the matching warning: a taxonomy defines categories within a single domain, an ontology links several of them, and it is easy to over-model. Start with one function, such as marketing or support, prove value, and grow from there. A small ontology is easier to extend than a large one is to trim.
AI raises the stakes. A language model reading a taxonomy learns that “net revenue” sits under “Finance metrics.” It learns nothing about which table holds it, which filter defines it, or who owns it. Those facts are relationships, and only an ontology records them.
How does Atlan handle taxonomy and ontology?
Atlan keeps both layers in one place, because data teams need the taxonomy for people and the ontology for agents.
The taxonomy layer is the business glossary. In Atlan’s glossary, you nest terms under categories and subcategories to build a hierarchy. On top of that hierarchy, terms carry typed links that go past broader and narrower: synonyms, antonyms, recommended terms, “translates to,” “valid values for,” and “classifies.” That is thesaurus territory, and the first step toward an ontology.
The ontology layer sits on the Enterprise Data Graph. Atlan’s Context Agents auto-generate descriptions, metrics and business ontology across the graph, which 80+ connectors fill from the warehouse, BI tools and business applications. Atlan defines that ontology as the map of the business: its entities, metrics and relationships. Human experts certify it, and certified context reaches downstream tools through SQL, APIs and the Atlan MCP server.
At CME Group, the glossary is the foundation. Kiran Panja, Managing Director, Cloud & Data Engineering, said: “With Atlan we cataloged over 18 million assets and 1,300+ glossary terms in our first year, so teams can trust and reuse context across the exchange.”
If you are weighing an ontology against the metric definitions in your BI layer, ontology vs semantic layer covers that question.
Atlan in Action: Live Context Layer Demos
See how Atlan keeps a business glossary and a business ontology on one Enterprise Data Graph that people and AI agents can use.
Watch the Live DemoFAQs about ontology vs taxonomy
Is a taxonomy a type of ontology?
A taxonomy is the simplest structure on the same spectrum. It has one relationship, broader and narrower, and no rules. In OWL DL, the W3C SKOS Primer says SKOS concepts and OWL classes must be kept formally distinct, so you link the two on purpose rather than treat one as the other.
Is a thesaurus the same as a taxonomy?
No. A thesaurus adds two relationships a taxonomy lacks: equivalence, which ties synonyms and acronyms to one preferred term, and association, which links related terms outside the hierarchy. ANSI/NISO Z39.19 and ISO 25964-1 both govern thesaurus construction. A thesaurus still has no formal rules for a reasoner to apply.
Can SKOS represent an ontology?
SKOS represents taxonomies, thesauri and classification schemes. The W3C specification says SKOS is not a formal knowledge representation language, and that a thesaurus or classification scheme asserts no axioms. For classes, typed properties and inference, use OWL 2. Many teams keep both: SKOS for labels and search, OWL for the model, with explicit links between concepts and classes.
Do AI agents need an ontology, or is a taxonomy enough?
A taxonomy helps an agent find and tag the right assets. It cannot tell the agent how tables join, which definition of a metric applies, or which policy covers a dataset. Those are typed relationships and rules, which only an ontology records. For retrieval alone, a taxonomy can be enough. For reasoning across domains, it is not.
How do you turn a taxonomy into an ontology?
Treat each taxonomy branch, facet or concept scheme as a candidate class. Then name the relationships between classes with verbs, such as “supplied by” or “owns,” and add attributes and constraints. Keep the original taxonomy for labels and tagging. Start with one business function and expand once the first set of questions is answered.
Sources
- Extending Taxonomies to Ontologies, Heather Hedden, Enterprise Knowledge
- SKOS Simple Knowledge Organization System Reference, W3C
- OWL 2 Web Ontology Language Document Overview (Second Edition), W3C
- Shapes Constraint Language (SHACL), W3C
- ANSI/NISO Z39.19-2005 (R2010) Guidelines for the Construction, Format, and Management of Monolingual Controlled Vocabularies, NISO
- Relations in the Gene Ontology, Gene Ontology Consortium
- What Is the Difference between Taxonomy and Ontology?, Earley Information Science
- Structures of Controlled Vocabularies, ANSI/NISO Z39.19 text hosted by Marcia Lei Zeng
- ISO 25964-2:2013, Thesauri and interoperability with other vocabularies, Part 2, ISO
- A Translation Approach to Portable Ontology Specifications, Tom Gruber
- Schema.org, Schema.org
- Financial Industry Business Ontology (FIBO), EDM Council
- SKOS Simple Knowledge Organization System Primer, W3C
