Brave AI Mode Scraper

Extracts completed answers from Brave Search's public AI Summarizer for one query or a bounded batch. The Actor opens the website, requests AI mode, waits for a recognized completed answer container, and stores one validated record per answer.

Modes

• Single query with query
• Multiple queries with query plus queries
• Bounded parallel query pages with maxConcurrency
• Country and language localization
• Configurable summary and navigation timeouts
• Direct, authorized Apify Residential, or custom proxy routing

Output

Each successful record includes a stable query/locale ID, query/global positions, plain summary text, sanitized structural HTML, character/word/estimated-token counts, related and follow-up questions, source/citation arrays when Brave displays them, model information when displayed, accompanying organic results when present, exact search context, provenance, and scrape timestamp.

Optional attributes are omitted when Brave does not display them. Legitimate zero counts remain numeric zero. Ordinary organic-result text or full-page body text is never accepted as an AI answer, and failed batches store no partial dataset.

Example

{
  "query": "What is machine learning?",
  "queries": ["What is quantum computing?"],
  "maxItems": 2,
  "country": "us",
  "language": "en",
  "waitForSummary": true,
  "summaryTimeoutMs": 45000,
  "navigationTimeoutSecs": 60,
  "maxConcurrency": 2,
  "maxRequestRetries": 1,
  "proxyConfiguration": { "useApifyProxy": false }
}

Reliability and limitations

The Actor uses one internally consistent Firefox profile and stable proxy session per run, validates HTTP status and content type, detects explicit block pages, closes every page/context/browser, and retries only bounded temporary failures. CAPTCHA and access-control pages are never solved or repeatedly retried.

Brave controls whether AI mode is available for a query and which optional sources, citations, model names, or organic results appear. A confirmed empty search completes with an empty dataset. A non-empty search page that never produces a genuine AI answer fails clearly instead of storing a placeholder.