Skip to main content
Enable formats.json and provide jsonParams.schema. An AI model reads the page’s Markdown and returns an object in json.data.

Define the object

Send this body to POST /web/scrape. See the Quickstart for authentication and SDK setup.
Abridged response:

Design for missing facts

Use optional or nullable fields for facts a page might omit. Required, non-nullable fields receive a best-effort value and can introduce guesses. Review extracted values before using them for consequential decisions. The schema must describe a top-level object. Wrap a list in an object property:
Zod users can pass z.toJSONSchema() output. jsonParams.instructions can specify which facts to prioritize or how to interpret fields.

Choose an extraction method

Limits

Schemas are limited to 50 KB when serialized; instructions to 2,000 characters. Extraction reads the first 100,000 characters of Markdown after content filtering. A page with no text returns {}. A failed or timed-out extraction has success: false and data: null.