Structured Data to JSON — JSON-LD & Schema.org extractor

Turn public product, article, event, organization, local-business or documentation pages into validated JSON with evidence for every field. Unlike a raw JSON-LD dump or a prompt-based scraper, this Actor publishes six fixed contracts, keeps unsupported values null, and explains where every returned value came from.

It is built for AI agents, RAG pipelines, monitoring jobs and data workflows that need predictable output without invented values.

At a glance: $0.009 per successfully validated page (nothing is charged for invalid, blocked, login or empty pages) · 1–20 public HTML URLs of one type per run · six fixed schemas, evidence for every field, no LLM. Try it: click Start with the prefilled example page and the product schema.

What it extracts

schemaType	Fixed fields
product	name, price, currency, availability
article	headline, datePublished, description, section
event	name, startDate, endDate, location
organization	name, description, organizationType, location
localBusiness	name, category, city, openingHours
documentation	title, description, product, language

The deterministic pipeline checks JSON-LD first, then microdata, OpenGraph/meta tags, semantic HTML and restrained page-text rules. It does not call an LLM.

Input

{
  "urls": [
    "https://books.toscrape.com/catalogue/a-light-in-the-attic_1000/index.html"
  ],
  "schemaType": "product"
}

Send 1–20 public HTML URLs of the same type per run.

Output

{
  "url": "https://books.toscrape.com/catalogue/a-light-in-the-attic_1000/index.html",
  "finalUrl": "https://books.toscrape.com/catalogue/a-light-in-the-attic_1000/index.html",
  "schemaType": "product",
  "data": {
    "name": "A Light in the Attic",
    "price": "£51.77",
    "currency": "GBP",
    "availability": "In stock"
  },
  "evidence": {
    "name": {
      "method": "html",
      "pointer": "article h1, main h1, h1",
      "snippet": "A Light in the Attic"
    }
  },
  "absentFields": [],
  "source": {
    "contentType": "text/html; charset=utf-8",
    "bytes": 6812,
    "sha256": "..."
  }
}

The real item includes one evidence record for every field. A missing field is returned as null with:

{ "method": "absent", "pointer": null, "snippet": null }

Evidence proves which publisher text or metadata produced the value; it does not independently prove the publisher's claim is true.

If a copied value or evidence snippet contains an email address, a phone-like number or an obvious social/profile URL, that contact is replaced with [redacted contact]. Its evidence record includes "redactedContact": true. Version 1 has no contact fields, so this rule applies to both personal and business contacts wherever they occur inside otherwise supported text.

Pricing

$0.009 per successfully validated page. A page is charged only when a dataset item with a grounded identity field and schema-valid output is written. Invalid URLs, private destinations, robots denials, login/challenge pages, HTTP failures and empty results produce no paid dataset item. Apify also adds its small platform Actor-start event.

That is 70% below Apify's general AI Web Scraper price of $0.03/page on the FREE tier. This Actor is narrower: it does not accept arbitrary prompts or arbitrary schemas.

Safety and limits

• Respects robots.txt; policy fetch errors fail closed.
• Resolves DNS before connecting, refuses private/reserved addresses, pins the approved addresses for the connection, and rechecks every redirect.
• Accepts only public HTML/XHTML, up to 2 MiB after decompression.
• Uses no cookies, login, form submission, CAPTCHA solving, Cloudflare bypass or browser fingerprinting.
• Parses the returned HTML and rejects password controls regardless of tag/attribute case, quoting or form action; it also rejects login-form and common challenge markers. These pages produce no paid dataset item.
• Defines no email, phone or profile fields. Contact fragments found inside any copied field or evidence snippet are replaced with [redacted contact] and marked with redactedContact: true; this includes business contacts because none of the six version-1 schemas permits contact data.
• Does not render JavaScript. If facts exist only after client-side rendering, they remain null or the page can fail its identity check.
• A single run accepts one schema type. Split mixed page types into separate runs.

For AI agents

Through Apify's MCP server, an agent can find and call this Actor, then inspect data, evidence and absentFields before using a value. The fixed contracts are intentionally easier to validate than prompt-generated JSON.

Why not a universal AI scraper?

The public 20-page gate measured 97.5% field accuracy with deterministic extraction alone. A local quantized model added no correct fields and only increased latency, so it is not included in version 1. See ./VERIFY.md for the measured gate, market comparison, economics and live checks.

FAQ

Does it extract any field I describe in a prompt?

No. It supports the six documented schemas only. Use a general AI scraper for arbitrary prompts.

Does it copy all JSON-LD from a page?

No. It selects and normalizes only the documented fields, supplements them from grounded page metadata, validates the final object and attaches evidence.

What happens when a value is missing?

The value is null, its evidence method is absent, and the field is listed in absentFields. The Actor does not guess.

Can it scrape private or logged-in pages?

No. Public pages only; no credentials, cookies or login flows are accepted.

Is it affiliated with Schema.org or the sites it reads?

No. Schema.org names describe public vocabulary types. This is an independent Apify Actor.