Google News Article Ledger

Google News scraper for keyword searches, Google News topics and publisher sites. Each row is one article: headline, publisher, publication date, the resolved publisher URL and, when the publisher page can be read, readable article text, description and image. For scheduled monitoring, supply a successful full-snapshot dataset or an accumulated list of seen URLs. Each run returns only articles absent from that state; update the state between runs to avoid repeat delivery.

Typical uses: news monitoring for a company, product or topic; collecting headlines and article text for research or sentiment analysis; feeding a daily digest without duplicates.

Try it

Click Start with the prefilled input. It searches Google News for OpenAI in US English and returns up to 20 articles with resolved URLs and best-effort article text. The prefill is a form example only. An API call with {} returns no rows and a guidance message.

What you can search

• Search terms, for example OpenAI or climate policy, and Google News search URLs.
• Topics: WORLD, NATION, BUSINESS, TECHNOLOGY, ENTERTAINMENT, SPORTS, SCIENCE, HEALTH, or a Google News topic/section URL.
• Publisher sites, for example bbc.com.

Mix source types in one run and choose a Google News edition with country and language; source availability and locale support can vary. Recency windows and excluded words apply to search and publisher inputs, while topics keep Google's curated selection. siteFilter applies only to search inputs. Exact since/until filters apply before limits and billing, but only search and publisher inputs page back through older date windows.

Three modes

Mode	Settings	What you get
Article text (default)	resolveUrls true, extractArticleText true	Resolved publisher URL plus best-effort article text, publisher description, image, author, section and keywords when the publisher page can be read.
Resolved metadata	extractArticleText false	Resolved publisher URLs and feed metadata, no publisher page reads. Suited to large pulls.
Feed only	resolveUrls false, extractArticleText false	Google News links and feed metadata only.

Input examples

Several search terms:

{"startInputs":["OpenAI","climate policy"],"maxItems":40,"maxArticlesPerInput":20}

Topics and a publisher in one run:

{"startInputs":["TECHNOLOGY","bbc.com"],"maxItems":30}

German news for September 2026, metadata only, no per-input cap:

{"startInputs":["artificial intelligence"],"country":"DE","language":"de","since":"2026-09-01","until":"2026-09-30","maxItems":1000,"maxArticlesPerInput":0,"extractArticleText":false}

Filtered search:

{"startInputs":["energy"],"siteFilter":["bbc.com","theguardian.com"],"excludeWords":["opinion","sponsored"],"timeframe":"7d","maxItems":20}

New articles only, against a saved full snapshot:

{"startInputs":["OpenAI"],"baselineDatasetId":"YOUR_FULL_SNAPSHOT_DATASET_ID","maxItems":20}

Input fields

Field	Type	What it does
startInputs	array	Search terms, topics, Google News search or topic URLs and publisher domains. Empty means no requests.
country	string	Two-letter Google News country code. Default US.
language	string	Language code, including variants such as zh-Hant. Default en.
timeframe	string	all, 1h, 6h, 1d, 7d, 30d, 1m or 1y; default all. Applies to search and publisher inputs. Explicit dates override it. Topics keep Google's curated window.
since	string	Inclusive start date or zoned datetime. Search and publisher inputs walk back through date windows to reach it.
until	string	Inclusive end date or zoned datetime; a calendar date includes the whole UTC day.
maxItems	integer	Maximum articles delivered in the run. Default 100, accepted up to 50,000; a cap, not a promised yield.
maxArticlesPerInput	integer	Maximum distinct articles selected per input before new-only suppression, known URLs included. Default 100; 0 removes this cap within the run's time budget.
resolveUrls	boolean	Default true. False keeps Google News links and reads no publisher pages.
extractArticleText	boolean	Default true. Reads publisher pages for text and metadata. Needs resolveUrls true.
siteFilter	array	Publisher domains for search inputs. Publisher-domain inputs ignore it; it does not filter topics.
excludeWords	array	Words to exclude from search and publisher inputs. Topics keep Google's selection.
alreadySeenUrls	array	Canonical URLs you already have, from the same resolveUrls setting. Matching articles are skipped and not charged.
baselineDatasetId	string	Read-only dataset from a successful full-snapshot run. Enables new-only output. Never use a new-only run's dataset here.

Output

One row per distinct article URL, 31 fields. Abbreviated example from article-text mode (illustrative values):

{
  "title": "Example article headline",
  "source": "Example Publisher",
  "publisherDomain": "example.com",
  "url": "https://example.com/news/article",
  "googleNewsUrl": "https://news.google.com/rss/articles/EXAMPLE",
  "publishedAt": "2026-10-06T09:00:00Z",
  "publisherPublishedAt": "2026-10-06T08:59:00Z",
  "description": "Publisher description, or the feed summary.",
  "body": "Readable publisher article text when available.",
  "wordCount": 6,
  "imageUrl": "https://example.com/images/lead.jpg",
  "author": "Writer A, Writer B",
  "section": null,
  "keywords": [],
  "bodyExtracted": true,
  "bodyFailureReason": null,
  "sourceInput": "OpenAI",
  "country": "US",
  "language": "en",
  "scrapedAt": "2026-10-06T19:00:00Z"
}

Field	Meaning
rowType, sourceKind, sourceInput, sourceRssUrl	Always news-article; query, topic or publisher; the input you supplied; the feed URL that found the row.
title, source, publisherDomain	Headline, publisher name and domain from the feed.
url, publisherUrl	With resolveUrls true, both hold the resolved publisher URL. In feed-only mode, url holds the Google News link and publisherUrl is null.
canonicalUrl	The resolved publisher URL, or the Google News link in feed-only mode, with fragments, trailing slashes and common tracking parameters removed. It is the identity used for new-only output, not a verified publisher rel=canonical.
googleNewsUrl, guid, position	Google's link, feed ID and position in its feed; position can restart on another date window.
publishedAt, publishedAtRaw, publishedTimestamp	Google's publication time: UTC, the raw feed string, and milliseconds.
publisherPublishedAt, dateDisagreement	The publisher's own time when available, and whether it differs from Google's.
description	With article text on, a confirmed publisher description when the page has one; otherwise the feed summary as plain text, or null.
body, bodyHtml	Publisher article text and rebuilt paragraph HTML after quality checks, or null. Bodies under 400 characters, paywalled pages and unconfirmed selections are refused. Not a guarantee of the complete article.
wordCount	Words in body, or null.
imageUrl, author, section, keywords	From the publisher page when read and available; otherwise null or empty. Author names supplied as a list of strings are joined with ", "; other non-string names return null. This does not guarantee that every author is returned.
bodyExtracted, bodyFailureReason	Whether body extraction succeeded, and any body or metadata-identity failure reason. article_identity_unconfirmed can describe refused metadata even when bodyExtracted is true. not_requested means extraction was off; null means nothing failed.
changeStatus	new in new-only runs.

Export as JSON, CSV or Excel.

New articles only (monitoring)

1. Run once without baselineDatasetId or alreadySeenUrls. Keep the dataset from that successful full-snapshot run.
2. Pass that dataset as baselineDatasetId, or supply your accumulated alreadySeenUrls. When both are supplied, both filters apply.
3. For later runs, add every delivered canonicalUrl to your seen-URL list, or take another successful full snapshot. Never use a new-only run's dataset as the next baseline.
4. Keep the same resolveUrls setting. Google News links and publisher URLs are different identities; changing modes does not translate existing state.

Use an Apify task and schedule for recurring monitoring. Each run also writes a SNAPSHOT key-value record with the current selected window, known URLs included; known rows carry no new body. Limits, failures or the time budget can shorten that window. It is not a complete archive or a baseline dataset, and removed or changed articles are not reported.

Pricing

Pay per delivered article, plus $0.00005 per run start ($0.05 per 1,000 starts). Free and Bronze plans pay $1.50 per 1,000 articles, Silver $1.20, and Gold, Platinum and Diamond $0.54. Already-seen articles, errors and empty results have no article charge; a started run still pays the start event. A maximum-charge limit stops the run before the next batch; a batch already in flight finishes first. The Store pricing panel shows the current price.

Limits and known gaps

• Publisher text, images and descriptions are best-effort. Five cohorts of eight prefill runs on 2026-10-10, on builds with the same runtime code, each produced 160 rows: 0.1.25 at 21:58-22:07 UTC and 0.1.26 at 22:33-22:35, both at 512 MB, and 0.1.27 at 22:41-22:45, 0.1.28 at 22:54-23:08 and 0.1.29 at 23:15-23:29, all three at 1024 MB. Images appeared in 48.1% to 71.9% of rows. Descriptions appeared in 51.9% to 76.9% of rows. A description counts when it differs from the feed's headline-and-source text. Coverage depends on the share of results from publishers that refuse automated reads, such as WSJ, NYT, FT and Bloomberg, and that share varies with the news set. The 75% image target is not met. In the 0.1.28 cohort, image misses came from HTTP 401 or 403 responses, read timeouts and unconfirmed article identity. Browser retries supplied publisher fields or text on 1 of 38 attempts in the 0.1.25 cohort and 9 of 77 in the 0.1.28 cohort.
• Startup time varies. On 2026-10-10 at 512 MB, five 20-row feed-only runs had a 9.729-second median, slower than the 4.945-second median measured on 2026-10-07 with the previous, smaller image. Five more runs had a 4.043-second median; the ten-run median was 4.768 seconds. Two runs spent 14.578 and 16.521 seconds pulling the image, and the slowest took 36.758 seconds in total. Feed-only mode never starts the browser, but the browser-capable image is larger. These are samples, not a guaranteed startup time.
• Memory and timeouts. The default memory is 1024 MB. Use at least 512 MB and a 300-second timeout. Heavy publisher-text runs used most of a 512 MB allocation: on 2026-10-10, four of eight article-text runs on build 0.1.25 peaked at 511.5 to 512 million bytes. All eight started the browser fallback and succeeded. The Actor stops starting new HTTP work after 240 seconds with article text on, or 270 seconds otherwise. A stalled browser startup or shutdown can still delay storage and use up the platform timeout.
• Volume. Feeds are finite windows. On 2026-10-07 a 512 MB metadata run delivered 1,000 articles in 219.597 seconds; a run asking for 50,000 reached its time budget at 975 articles in 281.311 seconds. Size metadata requests around 1,000 rows. A 2026-10-06 text run with 40 concurrent reads delivered 271 articles, 112 with extracted bodies, in 255.109 seconds; this version allows 50. These dated examples are not guaranteed throughput.
• Not supported: full-coverage story clusters, entity extraction, inline thumbnails and Google search controls such as spelling correction or country restriction.

How it works

Publisher pages are read over HTTP first. After three failed attempts of a retryable kind, or an HTTP 401, one headless Chromium read is tried with the same article identity and field checks. HTTP 404 and other final 4xx answers, an exhausted time budget, and pages refused for quality or identity never start the browser. Metadata and feed-only modes never read publisher pages. Errors go to a separate ERRORS record and never become dataset rows; COVERAGE reports browser attempts and rescues.

Google News robots.txt disallows the RSS and resolver paths (checked 2026-10-06). The Actor reads public pages without logging in and does not claim permission from the target.

Support

Open an issue on the Actor's Issues tab with the run link.