CNN Search Scraper

Runs keyword searches against CNN.com's own search index and returns the matching articles, videos and photo galleries.

Covers	www.cnn.com only — see Limits
Content types	articles, videos, photo galleries
Returns	headline, URL, thumbnail, modified date, teaser; optionally full article body, author, section and publish date
Method	HTTP only, one unauthenticated JSON endpoint. No browser, no login, no API key

Example input

{
  "queries": ["climate change", "artificial intelligence"],
  "contentTypes": ["article", "video"],
  "sort": "relevance",
  "maxItemsPerQuery": 50,
  "includeFullBody": true
}

Output

Every row carries _input, _source, _scrapedAt, recordType. Three record types share the dataset: SEARCH_RESULT (one per hit), SEARCH_SUMMARY (one per query + content type, with CNN's own reported total and the filters actually applied) and ERROR. Every input maps to at least one row.

Limits — read these, they are not obvious

• Flagship only. CNN's search index covers www.cnn.com. It does not include CNN en Español, Arabic, Brasil, Indonesia or any licensee edition — verified by searching mexico and brasil and finding only www.cnn.com hosts in every result. Use the CNN Articles Scraper for the other editions.
• 1,000-result hard ceiling. CNN answers any offset at or beyond 1000 with an HTTP 403. A query can therefore never yield more than 1,000 rows regardless of the total it reports. Runs that reach it say so via _warning on the summary row rather than looking quietly truncated.
• searchTeaser is not the article body. The search API's own text field is a one-sentence teaser (measured 103–230 characters against real bodies of 2,000–7,000). It is emitted under its own name so it cannot be mistaken for full text; set includeFullBody for the real thing.
• The index is small. Measured totals: trump 23, 2026 233, russia 109, news 15. Common stopwords (the, people) return zero.
• Only three content types and two sort orders are real. See below.

Two silent-fallback traps this actor guards against

CNN's API does not reject an unknown types or sort value — it ignores it and returns something plausible:

• an unrecognised types returns the unfiltered result set (measured of=157 for every invalid value, versus 14 / 132 / 407 for article / video / gallery)
• an unrecognised sort silently behaves as newest

Both are therefore rejected in code, not just in the Console enum — an API or CLI run never sees a Console enum. Passing contentTypes: ["live-story"] returns a named invalid_input error instead of a full unfiltered dump that looks like a successful narrow search.