Google News Scraper — Headlines, Sources & Dates by Keyword

Search Google News by keyword and get one row per article: the clean headline, the publisher, a link and an ISO publish date. Any country and language, a recency filter from the last hour to the last year, and bulk searches in a single run.

No API key, no account, no scraping setup. $0.005 per article — a search that returns nothing is recorded and not charged.

Input

{ "query": "artificial intelligence", "maxArticles": 25, "when": "day" }

Bulk, with a country edition:

{
  "queries": ["tesla", "rivian", "lucid motors"],
  "maxArticles": 50,
  "when": "week",
  "country": "GB",
  "language": "en-GB"
}

Field	Type	Notes
query	string	Supports Google operators — "tesla" OR "rivian", site:reuters.com, intitle:.
queries	string[]	Up to 20 searches per run.
maxArticles	integer	1–100 per search, default 25. You are charged per article returned.
when	enum	any, hour, day, week, month, year. Applied by Google itself.
country	string	Edition country code — US, GB, DE, IN. Default US.
language	string	Edition locale — en-US, en-GB, de, fr. Default en-US.

Output

One row per article:

{
  "ok": true,
  "query": "artificial intelligence",
  "title": "Viral AI researcher's warning 'scary as hell,' Cruz says",
  "source": "Politico",
  "link": "https://news.google.com/rss/articles/CBMik...",
  "publishedAt": "2026-09-10T14:23:13.000Z",
  "publishedRaw": "Thu, 10 Sep 2026 14:23:13 GMT",
  "when": "any",
  "language": "en-US",
  "country": "US"
}

A row per article is what a spreadsheet, a filter and an integration all expect — no unpacking a nested array.

What it gets right

• The headline is clean. Google appends " - Publisher" to every title. That suffix is removed, and the publisher is returned in its own source field — but only when it genuinely is the publisher, never by blindly cutting at the last dash, because plenty of headlines contain dashes of their own.
• publishedAt is a real ISO-8601 timestamp, not the raw RFC-822 string. publishedRaw is kept alongside it if you want exactly what Google said.
• Recency is applied by Google, not filtered afterwards, so "last 24 hours" returns a full page of recent articles rather than a page of mixed results cut down to three.
• ⚠ month is sent as 30d on purpose. Google accepts when:1m, answers HTTP 200, and returns zero items — measured against every other bucket in one sweep (30d → 100 items, 7d → 100, 1d → 100, 1h → 25, 1y → 100, 1m → 0). A "last month" search built the obvious way silently returns nothing and looks like a quiet news cycle.
• XML entities are decoded in the right order, so " becomes " rather than ".

Pricing

$0.005 per article returned. No start fee. A search that returns no articles produces an articleCount: 0 record and is not charged, so probing a thin keyword is free.

Set maxArticles to control spend: 10 articles across 5 searches is 50 rows, $0.25.

Limits

• 20 searches per run, up to 100 articles each (Google's own ceiling per feed).
• Links point at Google News, not the publisher. Resolving them is deliberately not done: following one returns Google's /sorry interstitial with HTTP 429, so an Actor that tried would be fast on the first call and blocked on the rest.
• Returns article metadata, not article text. For full text, pass the resolved URL to a page extractor such as Webpage Text Extractor.
• Google News coverage varies by edition; a country/language pair with no edition returns few or no results.

FAQ

Do I need a Google API key? No. This reads Google News' own public RSS.

Can I get the publisher's direct URL? Not reliably — see Limits.

Why did my search return nothing? Either the keyword genuinely has no coverage in that edition, or the recency window is too tight. Empty searches are free, so widen when and retry.

Changelog

• 0.1 (2026-09-10) — first release. One row per article, clean headlines, ISO dates, Google-side recency, bulk queries, per-edition country and language.