Say you work on an integration that joins customer data from two vendors. You have heard that JSON-LD is how machines learn that one company's customer is the same thing as another company's customer. The APIs.io catalog has a JSON-LD section, so you go looking for contexts to lean on.
The section lists 7,381 contexts from 5,465 providers. That sounds like a deep well. A post published on November 5, 2026 says to slow down before you drink from it.
About 85% of those contexts declare schema.org. Declaring it is not the same as using it. Of 373,037 term mappings the catalog could parse, 40,413 point at a schema.org term. That is 10.8%. The rest mostly point at a namespace that belongs to the provider.
That matters to you because of what a provider-only mapping does. It gives a term a unique name. It does not tell a machine that one company's customer is another company's customer, which the post calls the reason to use JSON-LD at all. A JSON Schema gives you the shape of a field. A term mapped to schema:email carries meaning across vendors. A term mapped to vendor:email does not, and neither does a context that declares schema.org and leaves it unused.
So you look at who publishes the most. Adyen has 855 contexts, 11.6% of the collection, with a median of 4 properties. HubSpot has 61, FactSet 60, Palo Alto Networks 54 and Snowflake 38. At the other end, 5,188 providers have exactly one context.
There is a caution in the post about the word publishes. All 7,381 context files are served from artifacts.apis.io, the catalog's own artifact host. They sit on a provider's page because they describe that provider's API. They are not a claim that the provider ships JSON-LD itself. You should read them as a semantic layer built over each contract, unless the record shows otherwise.
You want a model to copy, and the post gives you one. Shopify's context maps Product to schema:Product, Customer to schema:Person, Order to schema:Order and Address to schema:PostalAddress. Things with no public equivalent, such as Fulfillment and InventoryItem, stay in the shopify: namespace. It also brings in GoodRelations and Dublin Core, and 80 of its 104 counted terms reach schema.org. Square and Meta go further, with every one of 148 and 126 counted terms mapped to it. Stripe nests a context per type, so Customer.name maps to schema:name while Customer.id stays stripe:customerId. The post calls that exactly the right distinction.
Then you look at the weak version, which the post says is common. Adyen's main context sets @vocab to https://adyen.com/schema# and maps each term to adyen:, some with an xsd type. It is tidy and specific to Adyen, and it adds little beyond the OpenAPI schema it came from. Across its 855 contexts, 167 of 8,081 parsed terms, or 2.1%, reach schema.org. Worse, Mocki has a class named GetV1D4867D8BB5D54A48A4Ab79131B5809B8Response, a URL turned into a type name. A rough pattern match finds about 58 contexts with class names like that.
Here is the turn, and it is aimed at the catalog itself. Before you blame the providers, the post lists what is the catalog's own to fix. Provider-namespace URIs like Adyen's are identifiers, and the author has not tested whether they resolve. 259 records are truncated in the store, so they sit outside these counts. And 1,426 contexts report zero classes, including Stripe's, which plainly defines Customer, and Square's. The catalog's extractor appears to miss some nested and schema-mapped definitions, so the class count on those pages undersells them. If you trusted a zero on one of those pages, you would have been misled by the catalog, not by the vendor.
So you adjust how you work. You stop asking how many contexts a provider has, or whether it declares schema.org. You ask the question the post recommends: how many of its terms point outside the provider?
What you take away is simple. The collection is broad and the semantics are thin. The post suggests that mapping the twenty most common nouns in a contract to schema.org would change that for most providers. Until then, treat a context as a starting point you have to check, and treat its class counts as a floor.
Read the original: https://apis.io/2026/11/05/json-ld-contexts-on-apis-io-map-only-10-8-percent-of-terms-to-schema-org/