Google Patents Scraper

Search public Google Patents records or fetch one known publication. The Actor uses Google Patents' structured search response for speed and optionally enriches each result from its public patent page.

Modes

• search: one user-friendly keyword search.
• batch: multiple searches with fair maxItems allocation and concurrent requests.
• patent: one publication number or public Google Patents URL.

detailLevel controls cost and richness:

• search: structured result data only—fastest for large result sets.
• metadata: adds current status, application and expiry details, inventors/assignees, claims count, classifications, citation counts, legal-event count, images, and other public metadata.
• full: also adds full claims, description, citation records, and legal-event records. Patent mode defaults to full.

Example inputs

Rich search:

{
  "mode": "search",
  "query": "machine learning",
  "country": "US",
  "language": "en",
  "detailLevel": "metadata",
  "maxItems": 5,
  "proxyConfiguration": { "useApifyProxy": false }
}

Full single-patent lookup:

{
  "mode": "patent",
  "patentNumber": "US10977654B2",
  "detailLevel": "full",
  "proxyConfiguration": { "useApifyProxy": false }
}

Batch search:

{
  "mode": "batch",
  "queries": ["machine learning", "quantum computing"],
  "detailLevel": "search",
  "maxItems": 20,
  "maxConcurrency": 2
}

Search and filter inputs

Search mode supports query; batch mode supports queries. Filters include patent country, result language, assignee, inventor, priority-date range, granted/application status, and relevance/newest/oldest sorting. Limits, search/detail concurrency, request retries, request timeout, and maximum pages are bounded and configurable.

Dates accept YYYY-MM-DD; compact legacy YYYYMMDD values remain supported. A global maximum of 500 records matches the five structured pages available to the Actor.

Dataset

The source-backed 85-field contract includes stable SHA-256 IDs, canonical patent and PDF URLs, publication/application identifiers, kind and status, dates, inventors, assignees, abstracts, claims, classifications, citations, legal events, images/figures, family-country status, search context, source, and ISO timestamps. Optional values are omitted when Google Patents does not expose them—no nulls or fabricated placeholders are written.

Search-card records remain useful if an optional detail page is temporarily unavailable. In that case, detailFetched stays false and the run summary reports the enrichment failure count. A single-patent lookup fails if its requested detail page cannot be validated.

Proxy and access behavior

• Direct HTTP is the default and fastest route.
• Apify Residential and custom proxies are supported through proxyConfiguration.
• GOOGLE_SERP is deliberately rejected. It is designed for Google Search results, not patents.google.com structured and detail endpoints.
• Status and content type are validated before parsing. Malformed JSON, unexpected pages, explicit automated-query/CAPTCHA pages, and HTTP 403/429 responses fail closed and are never stored.
• The Actor does not solve CAPTCHAs or access authenticated/private material.

Performance

Search pages contain up to 100 structured results, so search-only mode normally needs one request for common workloads. Metadata and full detail pages run with bounded concurrency. Representative runs are designed to complete in seconds; very large full-detail runs depend on patent size, target rate limits, and proxy performance but remain subject to Actor timeouts.

Local verification

npm test
npm run lint
npm run validate:dataset
apify validate-schema
apify run --purge --input-file INPUT.json
node scripts/audit-dataset.mjs

The qa-inputs/ folder contains minimal, batch, filtered, full-patent, empty-result, invalid, Residential, and incompatible-GOOGLE_SERP cases.

Limitations

Google Patents data and markup change over time. Full descriptions and claims can be large. Family status, citations, legal events, classifications, and images vary by publication and jurisdiction. Search results can identify family members outside the requested country; the country input follows Google Patents' own search filter rather than rewriting source results.