Source-Linked News Briefs from Public Articles

Collect public news articles and create source-linked briefings for monitoring, research and editorial triage. Each brief links to the articles it uses. Generated briefs are extractive: they preserve bounded passages from the evidence and do not invent importance scores, conclusions or missing dates.

Example input and quick start

Paste a cataloged publisher article, homepage or XML feed into Public publisher page URLs, choose a small result limit, and run. Alternatively select countries and optional publishers. Explicit page URLs override country defaults. Clear supplied JSON to [] when collecting public pages.

{
  "pageUrls": [
    "https://www.theguardian.com/world/rss"
  ],
  "maxPublicPages": 1,
  "maxArticles": 2,
  "max_results": 1
}

Coverage and useful limits

The bundled catalog contains 912 active publisher configurations across 151 country and region labels. These numbers describe configured sources, not independently verified live coverage. A small global page budget applies to all countries, redirects, feeds and discovered articles. Blocking, publisher changes and unavailable pages can produce fewer results. Collection stays within publisher HTML and XML feeds; it does not use an external news database or source APIs. Historical archive discovery is not provided.

Native article bodies and publisher feed excerpts are labeled by contentCoverage. Feeds may contain only a short excerpt. Publication dates come from the source; unknown dates are omitted. Dated daily briefs require actual publication dates and are grouped by source country and publication day. Country describes the publisher, not the place an event occurred. Results are an observed sample, not a complete daily news agenda.

Each result groups dated source articles into one country/day extractive brief. No model, translation, inferred importance score or external summary service is used.

Analyze your article evidence

Use articleRecords or articleDatasetId to generate new briefs from your own raw articles. Each article needs article_title, article_body, article_url, site_name, country and a timezone-aware date_published for dated briefs. Supplied evidence is labeled sourceVerified false. Existing providedRecords/providedDatasetId remain supported for validating finished summary exports; this mode preserves your records and writes evidence-linked analysis to EVIDENCE_ANALYSIS. Do not mix public URLs with supplied input modes.

Output

Illustrative shape; the headline, date and URLs below are examples, not a fetched news report:

{
  "summary_date": "2026-10-01",
  "country": "United Kingdom",
  "topic_headline": "Illustrative publisher headline",
  "topic_summary": "A bounded excerpt from the supporting publisher article.",
  "article_count": 1,
  "source_site_names": [
    "The Guardian"
  ],
  "summaryMethod": "extractive",
  "source_article_urls": [
    "https://www.theguardian.com/world/illustrative-article"
  ]
}

Meaningful requested rows go to the default dataset. NEWS_EVIDENCE contains the supporting collected or supplied articles. PUBLIC_COLLECTION_SUMMARY reports requests, source outcomes, eligible articles, confirmed delivery and limits. EVIDENCE_ANALYSIS describes delivered evidence. No placeholder row is emitted for an empty result. Enable require_results to make an empty run fail. Only meaningful delivered results use the existing result event; deterministic briefs do not create AI add-on charges. Check the Store pricing panel for current rates.

Python API example

import os
from apify_client import ApifyClient

client = ApifyClient(os.environ["APIFY_TOKEN"])
run = client.actor("GcM9t3rmSWqf30I7P").call(run_input={
    "pageUrls": [
        "https://www.theguardian.com/world/rss"
    ],
    "maxPublicPages": 1,
    "maxArticles": 2,
    "max_results": 1
})
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(item)

Use the default dataset for JSON/CSV export and the run output links for evidence and collection diagnostics.

Support

Open the Actor Issues tab with a run ID and a sanitized input example. The run output links include collection or validation summaries for blocked sources, invalid input and partial results.