Technical guide · reviewed September 2026

How to extract JSON-LD structured data from a website

Parse schema.org blocks from permitted public pages

JSON-LD is JSON embedded in a page, usually in a <script type="application/ld+json"> element. It can describe products, articles, breadcrumbs, organizations, and other entities. Treat it as publisher-provided structured data: useful for discovery, but not automatically complete, current, or authoritative for every field.

Inspect a block in DevTools

For a one-off check on a page you are allowed to inspect, parse every block and keep malformed entries separate instead of letting one bad block abort the whole result.

const blocks = [...document.querySelectorAll('script[type="application/ld+json"]')];
const parsed = blocks.flatMap((element, index) => {
  try { return [{ index, value: JSON.parse(element.textContent) }]; }
  catch { return [{ index, error: 'invalid_json' }]; }
});
console.log(JSON.stringify(parsed, null, 2));

Use the page’s visible content and the schema vocabulary documented at Schema.org when deciding how to normalize a field. Do not assume that a schema block proves the value is visible, available, or eligible for a search feature.

Why raw HTTP and browser output can differ

Some sites include JSON-LD in the initial response; others insert it after JavaScript runs. Compare a raw response with the DOM after a normal browser load before selecting your method. If the required public block is client-rendered, a browser-enabled request can help.

// Raw HTTP: this synthetic response may contain no JSON-LD.
const raw = await fetch('https://demo.example/products/widget');
const html = await raw.text();
console.log(html.includes('application/ld+json'));

// Browser DOM: the page may add a block after scripts run.
console.log(document.querySelectorAll('script[type="application/ld+json"]').length);

Extract with ExtractAPI

ExtractAPI parses valid JSON-LD blocks from the page and returns them in data.structuredData. The field name is structuredData, not jsonld; malformed blocks are skipped and response-size limits still apply.

curl -X POST https://extractapi.app/v1/extract \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url":"https://demo.example/products/widget","javascript":true}'

Here is a synthetic, shortened response:

{
  "status": "ok",
  "url": "https://demo.example/products/widget",
  "data": {
    "title": "Widget Pro | Demo Store",
    "structuredData": [
      {
        "@context": "https://schema.org",
        "@type": "Product",
        "name": "Widget Pro",
        "offers": {"@type":"Offer","price":"49.00","priceCurrency":"USD"}
      },
      {
        "@context": "https://schema.org",
        "@type": "BreadcrumbList",
        "itemListElement": []
      }
    ],
    "mainContent": "Widget Pro In stock $49.00"
  }
}

Normalize without losing provenance

Keep the original block, source URL, extraction time, and schema type alongside normalized fields. A product can have multiple offers or variants, and a page can publish more than one entity.

function productOffers(structuredData) {
  return structuredData
    .filter(block => block && block['@type'] === 'Product')
    .flatMap(product => {
      const offers = Array.isArray(product.offers) ? product.offers : [product.offers];
      return offers.filter(Boolean).map(offer => ({
        name: product.name || null,
        price: offer.price ?? null,
        currency: offer.priceCurrency ?? null
      }));
    });
}

Common failure modes

• Multiple blocks: iterate all blocks; do not assume the first block is the Product or Article you need.
• Malformed JSON: record an invalid_json category and continue. Never execute the text as JavaScript.
• Client rendering: raw HTTP may miss a block that appears only after scripts run; browser mode may still be unable to load a protected or unavailable dependency.
• Stale or conflicting values: compare structured data with visible text where correctness matters; preserve unknown values rather than inventing defaults.
• Rate limits and permissions: use a small allowlist, cache observations, respect target rules, and stop on an access control.

FAQ

Get a free API key