SchemaOrg

A strictly-typed builder for generating SEO Schema.org JSON-LD in Elixir and Phoenix applications.

You should not have to memorise the Schema.org vocabulary. This library ships 1000+ generated struct modules (SchemaOrg.Product, SchemaOrg.Offer, …), one per Schema.org Class. Build a graph with ordinary struct literals — your editor auto-completes the valid fields and the compiler rejects the rest — then serialise it with to_json_ld/1.

%SchemaOrg.Product{
  name: "MacBook Pro",
  offers: %SchemaOrg.Offer{price: 1999.00, price_currency: "USD"}
}
|> SchemaOrg.to_json_ld()

{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "MacBook Pro",
  "offers": { "@type": "Offer", "price": 1999.0, "priceCurrency": "USD" }
}

Quick Start

mix deps.get          # fetch jason + ex_doc
mix compile           # compile the library and its generated types
mix test              # run the suite

Commands

Architecture

Two layers, one hand-written and one generated:

• Runtime API (lib/schema_org.ex, lib/schema_org/thing.ex) — hand-written. SchemaOrg.to_json_ld/1 serialises any generated struct (recursively) into a @context/@type-annotated JSON-LD map and encodes it with Jason.
• Generated types (lib/schema_org/types/) — 1000+ files, one per Schema.org Class. Each is a plain struct (every valid property, direct and inherited, is a field) plus a new/0 constructor. Build values with struct literals; a field is untyped, so it accepts Schema.org's loose value model directly (a scalar or a nested struct; a single value or a list). Never edit these by hand — they are produced by the code-generation task and overwritten on every run. (See ADR-002 for why building is struct-literal rather than pipe-setter based.)
• Code generation (lib/mix/tasks/schema_org.build_types.ex) — a maintainer-only Mix task that ingests the official Schema.org JSON-LD graph (priv/schemaorg-current-https.jsonld), maps Properties onto Classes via domainIncludes and subClassOf inheritance, and renders each module through an EEx template (priv/templates/type.ex.eex).

priv/schemaorg-current-https.jsonld ──▶ mix schema_org.build_types ──▶ lib/schema_org/types/*.ex
                                              (EEx template)

Documentation

Current docs:

• docs/specs/01-type-generation.md — Code-generation pipeline: JSON-LD parsing, EEx template, type module layout — Implemented
• docs/plans/01-type-generation.md — Task-by-task breakdown for the above
• docs/decisions/ADR-001-build-time-codegen-committed-artifacts.md — Build-time generation, committed as artifacts
• docs/decisions/ADR-002-struct-literal-api-over-pipe-setters.md — Struct-literal building API (performance)

Schema.org in Plain Terms

Schema.org is a shared vocabulary that search engines (Google, Bing, Yandex) understand. When you embed it as JSON-LD in a page, you are telling the crawler "this page is about a Product called MacBook Pro that has an Offer of $1999". That structured data is what powers rich results — star ratings, price snippets, breadcrumbs, FAQ accordions.

The vocabulary is a graph of Types (Classes) and Properties:

• A Type is a thing you can describe — Product, Person, Recipe, Event.
• A Property is an attribute — name, price, author, startDate.
• Each property declares which types it is valid on (domainIncludes) and what kind of value it expects (rangeIncludes).
• Types form an inheritance tree (subClassOf): every type ultimately descends from Thing, so every type has name, description, url, and so on.

Writing this JSON by hand is error-prone — there are ~800 types and ~1500 properties, and putting a property on the wrong type produces silently-invalid markup. This library turns that graph into typed Elixir so the compiler and your editor catch mistakes before a crawler ever sees them.

Abbreviations