News Stories & Trends for AI Agents — Deduplicated, Multilingual

Stop paying for 40 copies of the same headline. This Actor returns stories: every news event grouped with all the outlets that cover it, in 65 languages, with coverage trends, tone and spike detection — in a short, stable JSON format designed for AI agents, MCP tools, PR monitoring and market/risk analysts.

• 🧩 One story, not 40 headlines — articles about the same event are merged across outlets and languages, so your agent reads less duplicated text and spends fewer tokens.
• 🌍 Truly multilingual — 289 outlets from 55 countries in 26 languages (RSS feeds + 48-hour news sitemaps), multilingual topic matching ("artificial intelligence" also finds inteligencia artificial, intelligence artificielle…), plus the GDELT global news index (65 languages) when it responds.
• 📈 Trends included — articles and outlets per hour/day for each topic, % change vs. baseline, a spike flag and GDELT tone when available.
• 🔁 Only what's new — in incremental mode, scheduled runs return (and charge) only stories that are new or were picked up by more outlets since the last run.
• 🤖 Agent-ready — compact mode (~150 words per story), extractive summary by default, optional written summary with your own OpenAI/Anthropic key. Works out of the box as an MCP tool via Apify's MCP server.
• 💸 $1.50 per 1,000 stories — and each story contains every article about it.

What you get

One row per story:

{
  "type": "story",
  "storyId": "3f1c9a0d2b7e4a11",
  "topic": "artificial intelligence",
  "headline": "EU regulators issue first fines under the AI Act",
  "summary": "Regulators in Brussels have issued the first penalties under the AI Act, Europe's landmark artificial intelligence regulation.",
  "firstSeen": "2026-11-05T06:10:00Z",
  "lastSeen": "2026-11-05T11:40:00Z",
  "articleCount": 37,
  "sourceCount": 29,
  "languages": ["en", "fr", "de", "es", "it"],
  "countries": ["US", "FR", "DE", "ES"],
  "tone": -2.1,
  "trend": "rising",
  "articles": [
    {"title": "EU regulators issue first fines under the AI Act", "source": "politico.eu", "url": "https://…", "publishedAt": "2026-11-05T06:10:00Z", "language": "en"},
    {"title": "L'UE inflige les premières amendes au titre de l'AI Act", "source": "lemonde.fr", "url": "https://…", "publishedAt": "2026-11-05T07:02:00Z", "language": "fr"}
  ],
  "change": "new",
  "attribution": "GDELT Project (https://www.gdeltproject.org/)"
}

Plus one trend row per topic:

{"type": "trend", "topic": "artificial intelligence", "interval": "hour", "totalArticles": 642, "totalOutlets": 187,
 "trend": "rising", "changePct": 78, "spike": false, "avgTone": null,
 "points": [{"t": "2026-11-05T10:00:00Z", "articles": 41, "outlets": 33}, …]}

• trend (story): rising / stable / fading / new, based on how coverage is distributed over the time window.
• tone: GDELT tone (negative ≈ negative coverage, positive ≈ positive), null when unknown.
• change: new or grown (incremental mode); previousSourceCount is added for grown stories.

Use cases

• AI agents & assistants — give your agent a fresh, deduplicated news briefing on any topic in one tool call (MCP or API). Fewer tokens, more signal.
• PR & communications — track a brand, CEO or product across 65 languages; schedule hourly with incremental mode and get alerts only when a story is new or spreading.
• Market, risk & policy analysts — detect coverage spikes and tone shifts on companies, sectors, countries or regulations.
• Newsletters & dashboards — feed "top stories of the day" by outlet count, not by click-bait.

How to use

1. Enter one or more topics (in English; multi-word topics are matched as exact phrases; GDELT boolean syntax like (openai OR anthropic) works).
2. Pick a time window (1h, 24h, 7d, up to 3m).
3. Run. That's it — defaults return stories covered by at least 2 outlets, with trends.

Scheduled monitoring: turn on Only new or growing stories, create a Schedule (e.g. every hour) and add an integration (Slack, email, webhook, Make, Zapier, n8n). You're only charged for stories you haven't received yet.

As an MCP tool: add this Actor to Apify's MCP server (https://mcp.apify.com) and your agent (Claude, ChatGPT, Cursor, LangChain…) can call it with {"topics": ["…"], "timespan": "24h", "compact": true}.

API example (Python):

from apify_client import ApifyClient
client = ApifyClient("<YOUR_APIFY_TOKEN>")
run = client.actor("nacenempresas/news-stories-for-ai").call(run_input={
    "topics": ["nvidia", "\"AI Act\""], "timespan": "24h", "minSources": 3, "compact": True})
for story in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(story.get("sourceCount"), story.get("headline"))

Input highlights

Field	Default	What it does
topics	["artificial intelligence"]	Topics, brands, public figures
timespan	24h	Window back in time (15min – 3 months)
minSources	2	Minimum outlets per story
incremental	false	Only new / growing stories since last run
trends	true	Volume & tone series per topic
compact	false	Short rows for LLM context windows
languages, countries, domains, excludeDomains	all	Filters
sources	gdelt, rss	Add googlenews (off by default)
customFeeds	–	Your own RSS/Atom feeds
llmApiKey	–	Your OpenAI/Anthropic key for written summaries (optional)

Pricing

Pay per event — no subscription:

Event	Price
Story delivered	$1.50 / 1,000 stories
Trend series (per topic)	$1.00 / 1,000
Run start	$0.01

A daily briefing of 50 stories costs about $0.085. In incremental mode, unchanged stories are free. Your maximum cost per run is always capped by the limit you set in Apify.

Data sources & attribution

• GDELT Project DOC 2.0 API — global news monitoring in 65 languages, used as an enrichment (extra outlets, tone) when it responds. Used under GDELT's open terms; rows that use it carry attribution: GDELT Project, https://www.gdeltproject.org/.
• Publisher RSS/Atom feeds and Google-News-style news sitemaps — 289 outlets, 55 countries, auto-discovered and re-validated every month.
• Google News RSS — optional and off by default; enable it under your own responsibility.

The Actor delivers headlines, a short snippet, source, date and link only — never full article text. Grouping runs on a small multilingual model inside the Actor; no third-party AI service is called unless you add your own key.

FAQ

How accurate is the grouping? On our labelled multilingual test set (same event reported in up to 5 languages) grouping scores 96.7% (B-cubed F1), using a fast multilingual model that runs inside the Actor. Use similarityThreshold to make it stricter (e.g. 0.6) or looser (e.g. 0.45).

Why fewer rows than a headline scraper? Because 30 articles about one event become 1 story with 30 sources. That's the point.

Is it real-time? Feeds and sitemaps are read live at every run (typical run: under a minute).

Personal data? Only public news headlines are processed. Nothing is stored outside your own Apify storage.

Questions or feature requests? Open an issue on the Actor's Issues tab — we answer quickly.