The semantic web standards a knowledge graph rests on come from several bodies, not one. The W3C publishes RDF, SPARQL, OWL, SHACL, and JSON-LD; the WHATWG maintains Microdata inside the HTML Standard; schema.org and the Open Graph protocol sit outside both. This page groups documents by the question each answers, with the publisher, maturity level, and when a practitioner has to open the document rather than a summary.
How to read a W3C document
W3C technical reports move along a track. A Working Draft can change in any direction; a Candidate Recommendation is stable enough that the group is asking for implementations; a Proposed Recommendation has gone to the membership for review; a Recommendation is the finished standard. Alongside that track, the W3C publishes documents that are not standards at all: Notes and, more recently, Statements. Several documents cited daily are Notes, and some widely used vocabularies are Community Group reports with no W3C review. The status line says whether the syntax in front of a reader is settled.
Inside a document, text is either normative or informative. Normative text defines conformance and uses the capitalized keywords MUST, SHOULD and MAY in the sense fixed by RFC 2119. Informative text is explanation: examples, notes, and sections marked “non-normative”. Every W3C specification says near the top which of its sections are which.
That distinction is why the Recommendation is usually the wrong place to start. It is written so implementers cannot disagree about edge cases; the Primer or Note beside it is written to be understood. Read the primer, build something, and open the Recommendation for the questions only conformance language answers: a grammar rule, a property’s domain, a function’s return type.
The RDF data model and its serializations
The RDF 1.1 document set became W3C Recommendations on 25 February 2014, splitting into one abstract model and several concrete syntaxes.
RDF 1.1 Concepts and Abstract Syntax defines what a triple, an IRI, a literal, a blank node, and a dataset are, and says nothing about how to write them down. Open it to settle a question about identity or datatypes, not to learn syntax. RDF Schema 1.1 supplies the small vocabulary most graphs actually use: rdfs:Class, rdfs:subClassOf, rdfs:domain, rdfs:range, rdfs:label. RDF 1.1 Semantics gives the entailment regimes that make those terms mean something to a reasoner.
The syntaxes are separate Recommendations from the same set: Turtle (the readable one, and the one to learn first), TriG (Turtle with named graphs), N-Triples and N-Quads (one statement per line, for streaming and diffing), and RDF 1.1 XML Syntax. The RDF 1.1 Primer is a W3C Working Group Note and the shortest useful introduction to it all. See RDF and triples: subject, predicate, object.
RDF 1.2 is in progress at the W3C RDF and SPARQL Working Group. In 2026, the W3C invited implementations of RDF 1.2 Concepts and Abstract Data Model and RDF 1.2 Semantics, putting both at Candidate Recommendation, while RDF 1.2 Turtle and RDF 1.2 Schema were still Working Drafts. RDF 1.1 remains the published standard; check the W3C status line before treating the newer work as fixed.
Querying and rules
Eleven SPARQL 1.1 specifications became W3C Recommendations on 21 March 2013. Four carry most of the weight. SPARQL 1.1 Query Language defines graph patterns, OPTIONAL, FILTER, property paths, aggregation, and subqueries. SPARQL 1.1 Update defines INSERT DATA, DELETE WHERE and graph management. SPARQL 1.1 Protocol defines how a query travels over HTTP, which makes public endpoints interchangeable. SPARQL 1.1 Federated Query defines SERVICE. The SPARQL 1.1 Overview maps the other ten and is the right first click. Working details are at SPARQL.
SPARQL 1.2 documents, including the Query Language, were still Working Drafts through 2026. Write against 1.1 and check the W3C page for the current state of 1.2.
Ontology and vocabulary languages
OWL 2 is the W3C Recommendation for describing classes, properties, and the constraints between them, published as a Second Edition set on 11 December 2012. Its Document Overview lists every part and says which are normative; the OWL 2 Primer is the readable introduction, and the Structural Specification and Functional-Style Syntax is the normative core. Most people write OWL in Turtle and never touch the functional syntax. OWL reasons under the open-world assumption, so a missing value is unknown rather than wrong, which is what decides most arguments about whether OWL or a validation language is the right tool. See OWL and ontologies.
Validation
Shapes Constraint Language (SHACL) is a W3C Recommendation from 2017 defining shapes, targets, constraint components and a machine-readable validation report, plus SPARQL-based constraints for conditions the core components cannot express. Rules are a different thing and live elsewhere: SHACL Advanced Features is a separate W3C Note covering rules and node expressions. A SHACL 1.2 set is in progress, with Core, SPARQL Extensions, and Rules at First Public Working Draft in 2025, and Profiling and User Interfaces following in 2026. None of it is settled; implementations follow the 2017 Recommendation. Details are in SHACL.
Shape Expressions (ShEx) solves the same problem with a compact grammar. It is a community specification maintained at shex.io, not a W3C Recommendation, and it is what Wikidata EntitySchemas are written in.
Vocabularies, and the languages that express them
RDF, RDFS, and OWL are languages. A vocabulary is a set of terms written in one of them, governed separately, and a vocabulary can be widely used while carrying no standards status at all.
schema.org is the vocabulary behind almost all structured data on the web, and it is not a W3C Recommendation. It is developed through an open community process, described on its own “How we work” page, with a Community Group hosted at the W3C, and it releases on its own schedule. Search engine support for a given type is a separate question, documented by each search engine. See schema.org and structured data.
DCMI Metadata Terms, published by the Dublin Core Metadata Initiative rather than the W3C, supplies dcterms:title, dcterms:creator, dcterms:date and the rest of the bibliographic core. FOAF describes people and their relationships; its Vocabulary Specification 0.99 carries a namespace document dated 14 January 2014 and is a community specification with no W3C status, still common in older datasets.
SKOS is the exception in this group. SKOS Simple Knowledge Organization System Reference is a W3C Recommendation from 18 August 2009, with a companion Primer, and models thesauri through skos:Concept, skos:broader, skos:narrower and skos:prefLabel: the right shape for a subject hierarchy that is not a class hierarchy.
PROV-O, the PROV Ontology, is a W3C Recommendation from 30 April 2013 and models provenance: who or what generated an entity, from which other entity, through which activity. It answers “where did this triple come from” without inventing a scheme.
Syntaxes embedded in HTML
Three ways exist to put structured data inside markup, specified in three different places.
RDFa is the fully standardized one. RDFa Core 1.1 Third Edition, HTML+RDFa 1.1 Second Edition and RDFa Lite 1.1 Second Edition are W3C Recommendations published on 17 March 2015, with an RDFa 1.1 Primer alongside. RDFa Lite is the subset worth learning: five attributes vocab, typeof, property, resource and prefix. More at RDFa.
Microdata has an unusual history. It began in the W3C HTML effort, was split out, and was published as a W3C Working Group Note on 28 January 2021, with the W3C’s own repository now pointing to the WHATWG. The normative definition of itemscope, itemtype, itemprop and itemref lives in the WHATWG HTML Standard, a Living Standard with no version numbers and no Recommendation status, because the WHATWG does not use that track. Citing the W3C Note as the specification is a common error. See Microdata.
The Open Graph protocol is not a W3C standard and never was. It originated at Facebook, and its specification is the single page at ogp.me. It borrows RDFa’s property convention on <meta> tags but defines its own vocabulary of og: terms, and each consuming platform adds requirements of its own. See Open Graph protocol.
The JSON-LD family
Three documents became W3C Recommendations on 16 July 2020: JSON-LD 1.1, defining the syntax and the @ keywords; JSON-LD 1.1 Processing Algorithms and API, defining expansion, compaction, flattening and conversion to RDF; and JSON-LD 1.1 Framing, defining the operation that reshapes a graph to a requested shape. W3C republished JSON-LD 1.0 as a Superseded Recommendation in November 2020.
Anyone hand-writing schema.org markup needs the syntax document and nothing else. The other two matter when JSON-LD is processed as RDF rather than read as JSON. Practical guidance is at JSON-LD.
Mapping existing data into RDF
R2RML and A Direct Mapping of Relational Data to RDF are W3C Recommendations from 27 September 2012 and cover mappings from SQL result sets to triples. RML extends the idea to CSV, JSON, and XML and comes from the W3C Knowledge Graph Construction Community Group, whose reports carry no W3C endorsement.
For developers
Bookmark three documents and skim the rest: SPARQL 1.1 Query Language for the grammar, RDF 1.1 Turtle for the syntax, and the OWL 2 Primer for the modeling vocabulary. When two triple stores behave differently, the answer is usually in the normative text of one of these.
For SEOs
The specification that governs a page’s markup rarely governs whether a rich result appears. JSON-LD 1.1 and schema.org define what is valid; each search engine’s documentation defines what is required. Check both, and expect disagreement. See entity SEO and structured data testing tools.
Related pages
FAQ
What are the main semantic web standards?
RDF defines the data model, RDF Schema and OWL define ontology vocabulary, SPARQL defines the query language, and SHACL defines validation. JSON-LD, Turtle, TriG, N-Triples, N-Quads, and RDF/XML are syntaxes for writing RDF down. All are W3C Recommendations. schema.org, the Open Graph protocol, and FOAF are widely used vocabularies with no W3C status.
Is schema.org a W3C standard?
No. schema.org is a vocabulary developed through an open community process with a Community Group hosted at the W3C, and it releases on its own schedule rather than through the W3C Recommendation track. The languages used to publish it, JSON-LD and RDFa, are W3C Recommendations. Search engine support is documented separately by each search engine.
Which W3C document should you read first?
The Primer or the Note, not the Recommendation, which is written so implementers cannot disagree about edge cases and is dense by design. The RDF 1.1 Primer, the OWL 2 Primer, and the SPARQL 1.1 Overview each give the shape of a technology in an hour, after which the Recommendation becomes a reference for looking things up.
Is RDF 1.1 still current?
RDF 1.1 has been the published W3C Recommendation since 25 February 2014 and remains the version implementations support. RDF 1.2 is in progress: the W3C invited implementations of RDF 1.2 Concepts and RDF 1.2 Semantics in 2026, while other documents in that set were still Working Drafts. Check the W3C status line before relying on it.
Sources and further reading
- W3C, Standards: Semantic Web: https://www.w3.org/standards/semanticweb/
- W3C, RDF 1.1 Primer: https://www.w3.org/TR/rdf11-primer/
- W3C, RDF 1.2 Concepts and Abstract Data Model: https://www.w3.org/TR/rdf12-concepts/
- W3C, SPARQL 1.1 Overview: https://www.w3.org/TR/sparql11-overview/
- W3C, OWL 2 Web Ontology Language Primer (Second Edition): https://www.w3.org/TR/owl2-primer/
- W3C, Shapes Constraint Language (SHACL): https://www.w3.org/TR/shacl/
- W3C, JSON-LD 1.1: https://www.w3.org/TR/json-ld11/
- WHATWG HTML Standard, Microdata: https://html.spec.whatwg.org/multipage/microdata.html
- The Open Graph protocol: https://ogp.me/