Introduction and scope

As large language models and AI agents become intermediaries between content and the people who consume it, the publisher's question shifts from "how are my products found by search engines?" to "how are my products discovered, interpreted, and acted upon by AI agents?" At the data layer this is a knowledge-representation question: it depends on which vocabulary a catalogue is expressed in and whether the agent's protocol consumes it.

This Note (a) inventories the existing open vocabulary stack for e-commerce, with provenance and cross-namespace mappings; and (b) analyses what that stack lacks for autonomous, transacting agents, proposing that the gaps be filled by extension and crosswalk rather than by protocol-private forks. It is intended as reference material and as the evidence base for an interoperability-layer proposal aligned with the group's Agent Interoperability work.

How the layers compose

A single JSON-LD @context typically mixes namespaces: schema: for the core commerce model, gs1: for richer product and supply-chain detail, and unece: (UN/CEFACT) for units and measures. GoodRelations supplies the conceptual e-commerce core that was folded into schema.org in November 2012 [[GR-SCHEMA-2012]]; its original namespace remains authoritative in full RDF/OWL/SPARQL environments.

{ "@context": {
    "@vocab": "https://schema.org/",
    "gs1":    "https://ref.gs1.org/voc/",
    "unece":  "https://vocabulary.uncefact.org/",
    "gr":     "http://purl.org/goodrelations/v1#"
} }

Layer 0 — JSON-LD (serialization / linking syntax)

Not a vocabulary but the W3C syntax that carries the vocabularies and links them [[JSON-LD11]].

Layer 1 — schema.org commerce core

Since the November 2012 integration, GoodRelations is the e-commerce core of schema.org [[GR-SCHEMA-2012]] [[SCHEMA-PRODUCT]]. The GR origin column shows the originating GoodRelations class or property where applicable.

1a. Types / classes

1b. Key properties

Layer 2 — GoodRelations (original namespace)

Authored by Martin Hepp [[GOODRELATIONS]]; an independent OWL vocabulary that remains primary in RDF/RDFa/OWL/SPARQL contexts, with mapping axioms to schema.org. At integration the overlap was small (only 5 of schema.org's 315 classes coincided with 27 GoodRelations classes), so most GR classes were new additions rather than duplicates. Use this mapping when consuming Linked Data published in the original http://purl.org/goodrelations/v1# namespace.

Layer 3 — GS1 Web Vocabulary

The first large-scale external extension to schema.org [[GS1-WEBVOC]]: it extends schema.org under Linked Data principles with explicit mappings, adding supply-chain and consumer-facing master data (initial focus: apparel, footwear, food/beverage/tobacco). Properties derive from existing GS1 standards (GDSN, GS1 Source, GPC); Offer is a GS1-specific addition. The complete normative list is the JSON-LD ontology at https://ref.gs1.org/voc/data/gs1Voc.jsonld; terms are ratified via the GS1 Global Standards Management Process and maintained at github.com/gs1/WebVoc.

3a. Verified terms (illustrative)

3b. GS1 Digital Link and resolver link types — the hypermedia/affordance layer

GS1 Digital Link expresses GS1 keys (GTIN, batch/lot, serial…) as a resolvable Web URI [[GS1-DIGITAL-LINK]]; the resolver returns typed links to related resources. The link-type idea is rooted in Web Linking / HATEOAS / REST: it is a resource-pointer layer, not (yet) a machine-actionable API-contract layer — a distinction central to the gap analysis. Several link types are already agent-relevant (marked ●).

Related layer — UN/CEFACT vocabulary

Supplies units of measure and trade/measurement types that complement schema.org and GS1 in product markup (e.g. unece:MeasureType, unece:MeasureTypeCode) [[UNCEFACT-VOC]]. Frequently co-used in the same JSON-LD @context for precise units.

Provenance summary

Agentic extensions — gap analysis

Every vocabulary above was built to describe products for two readers: a human browsing a page, and a search engine building an index. Autonomous agents that discover, verify, negotiate, transact and act under delegated authority are a third consumer the stack was not designed for. They need three things the descriptive stack underprovides:

1. Affordances, not just descriptions — machine-actionable operations with typed parameters, not links to human-readable pages. GS1's resolver link types (paymentLink, registerProduct, scheduleTime, …) point to where actions happen but do not describe the contract of those actions.
2. Verifiability — signed, fresh, attestable facts so an agent does not act on a hallucinated or stale price. GS1 already ships two primitives (jws, verificationService); they are not yet bound into the Offer model as a signed-offer pattern.
3. Governable authority + attribution — who may act, within what limits, and how content creators are credited when an agent uses or routes value through their content. This is the open-Web question expressed at the vocabulary layer.

Principle: extend and crosswalk, don't fork. Most needs below can be met by binding existing W3C primitives to schema.org/GS1 rather than inventing protocol-private schemas — which is what keeps the Web open across MCP, UCP and ACP [[AGENT-PROTO-SURVEY]] [[ACP]] [[UCP]].

Conclusion

Rows 1–3 and 7 of the gap table are where proprietary protocols are currently filling the vacuum each in their own way — ACP's feed and checkout API, UCP's capability profiles, AP2's consent proofs. If affordance, authority and attribution semantics are standardised as open extensions to the shared vocabulary, content stays publishable once and remains legible to every agent ecosystem; if they are not, each protocol forks its own and the publisher is back to an N×N integration burden. A vocabulary-neutral interoperability layer — the direction this group proposes — could provide the crosswalk that helps prevent such a fork and keep the Web open.