News Scraper: Bulk Keyword Headlines from GDELT

Give it a list of keywords or phrases. For every query it returns the latest news headlines and links that match, newest first, with the site, language, publication time, themes and named entities of each article, and an optional sentiment label. One run covers up to 100 queries and gives you one row per headline.

The headlines come from the GDELT Project news index, which is refreshed every 15 minutes. This is not Google News and it does not return article text: you get the headline, the link to the publisher's page and metadata. Use it to watch a list of brands, competitors, people, products or topics, and to feed the headlines into a spreadsheet, a database or an alert.

Export to CSV or Excel

The result is a table with one headline per row, ready for a spreadsheet or a BI tool. Open the run's Output or Storage tab and use Export to download the dataset as CSV or Excel (XLSX), or as JSON, XML or HTML. Through the Apify API, the same dataset is available with format=csv or format=xlsx. The themes and entities columns hold lists (JSON) in a CSV cell; query, title, domain, published_at and sentiment_label are plain-text columns for filtering and pivots. Filter on status = ok to keep only headlines.

Use it from Google Sheets or a no-code tool

1. Put one keyword per row in a sheet (your brand, competitors, products or a sector) and paste the column into queries.
2. Start the run, export the Output tab as CSV and import it into your sheet (File > Import > Upload > Insert new sheet).
3. Use a pivot table on query and sentiment_label to count positive, neutral and negative headlines per keyword, and sort by published_at for the newest.

Apify also connects actors to Make, Zapier and n8n, so a daily run can send the day's negative headlines about your brand to email or Slack.

What it is for

• Brand and competitor monitoring. One query per brand, product or executive; run it on a schedule with a short time window and read what was published since the last run.
• PR and communications. See which sites ran a story, when, and with which headline; compare the same query across languages with the lang option.
• Market and topic research. Track a list of themes (a regulation, a commodity, a technology) side by side, with sinceHours set to a day or a week.
• Feeds for dashboards and agents. The rows are flat and the links are direct, so they drop into a database, a newsletter or a language-model prompt (see the note on untrusted text below).

Input

{
  "queries": ["stablecoin", "semiconductors", "interest rates"],
  "lang": "en",
  "sinceHours": 24,
  "limit": 25,
  "sentiment": false
}

Field	Notes
queries	Search words or a quoted phrase, 1 to 200 characters each. Up to 100 per run. Duplicates (ignoring case and spacing) are removed. Runs of spaces and line breaks inside a query become one space. Shorter queries match more headlines: in a test on 13 October, semiconductors found 14 in 24 hours and semiconductor export controls none.
lang	Optional. Only articles in this source language, as a two-letter ISO 639-1 code (en, es, de, fr, ja and so on). Empty means all languages.
sinceHours	1 to 168, default 24. Only articles published in the last this many hours.
limit	1 to 50, default 25. Most headlines returned per query, newest first.
sentiment	Default off. Adds a positive, negative or neutral label and a score to each headline. A heuristic, see below.

An input that cannot be run (no queries at all, a lang that is not two letters, a sinceHours outside 1 to 168, a limit outside 1 to 50) stops the run before anything is charged. A single query that cannot be searched (longer than 200 characters, or with control characters) comes back as an error row and is free.

Output

One row per headline, in query order, newest first within a query. Queries that cannot be searched come first. A query that finds nothing, or fails, gets a single status row instead. Illustrative row with made-up values (the url and attribution_url links are left out of this example; they are described below):

{
  "query": "stablecoin",
  "position": 1,
  "title": "Example headline about stablecoin rules",
  "domain": "news.example.com",
  "language": "en",
  "published_at": "2026-10-12T13:00:00Z",
  "themes": ["ECON_CURRENCY_EXCHANGE_RATE", "TAX"],
  "entities": ["Example Finance Committee"],
  "tone": -0.8,
  "sentiment_label": null,
  "sentiment_score": null,
  "status": "ok",
  "source": "GDELT Project (gdeltproject.org)",
  "attribution": "The GDELT Project",
  "error_code": null,
  "error": null
}

Field	Meaning
query	The query this row belongs to, as searched.
position	1-based position within the query's results, newest first.
title, url	The headline and the link to the article on the publisher's site. The article page itself is not fetched.
domain, language	The publisher's site and the article's source language.
published_at	When the article was published, as the index records it (UTC).
themes, entities	GDELT theme codes and named entities (people, organisations, places) the index found in the article. Empty lists when it gives none.
tone	GDELT's own tone figure for the article (negative leans negative, positive leans positive). Present whether or not sentiment is on.
sentiment_label, sentiment_score	Only with sentiment on: positive, negative or neutral and the score behind it. Null otherwise.
status	ok (a headline of a query that is charged once), no_results (nothing matched in the time window; free) or error (invalid query or failed search; free).
source, attribution, attribution_url	The data source and the credit the service returns with every result. Keep them if you republish or publish the headlines: the data is credited to The GDELT Project.
error_code, error	On error rows: a short code (for example invalid_query, news_store_unavailable, timeout) and the reason.

Headlines and links only. The actor never fetches the article pages, so there is no article body, summary or image. Open url for the full story at the publisher.

Use it from Claude, Cursor or another MCP client

Add Apify's MCP server with this actor as a tool: https://mcp.apify.com?tools=tanod/news-headlines-search. Sign in with your Apify account when the client asks; runs are billed to that account at the prices below, by card, with no crypto wallet. Your assistant can then call it in a chat or agent loop.

Pricing

Pay per event, with no subscription:

Event	Price
Query searched (query-searched)	$4 per 1,000 queries ($0.004 each)
Actor start	Apify's default start fee

You pay per query that returns at least one headline, not per headline. The limit option does not change the price.

Worked example: 100 queries with up to 50 headlines each cost 100 x $0.004 = $0.40 plus the start fee, for up to 5,000 headlines, which is under $0.0001 per headline. That is the best case, when every query fills the limit; a query that returns 10 headlines still costs $0.004. With per-article pricing the cost grows with every headline returned, so the same 5,000 headlines would cost 5,000 times the per-article rate; here only the number of queries matters.

Queries with no match (no_results), invalid queries and failed searches (error) are free. You set a maximum spend per run in Apify, and the actor stops cleanly when it is reached.

Can an AI agent pay for it without an Apify account? Yes. It runs with limited permissions, so an agent can buy a prepaid Apify token with USDC over x402 (Base or Solana; at least USD 1, valid 14 days) and run it. See Apify's x402 docs.

Limits and notes

• Source: the GDELT Project index. Headlines and links from a local copy of the GDELT news feed, refreshed every 15 minutes. It is not Google News, and coverage is the sites GDELT indexes, which is wide but not every publisher. A story usually appears within about 15 to 30 minutes of publication.
• Time window. The index keeps the last 14 days. A single search reaches back at most 168 hours (7 days) through sinceHours; the default is 24 hours.
• Up to 100 queries per run, up to 50 headlines per query, so at most 5,000 headline rows. Queries beyond 100 are skipped and reported in the run summary.
• Newest first, repeated stories dropped. Within a query the index removes duplicate stories, so a story syndicated on many sites appears once.
• A query is matched as search words or a quoted phrase. Use a quoted phrase for exact wording. A very broad query returns only the newest limit headlines, not every match.
• Sentiment is a heuristic. The label is produced automatically from the headline text with an English-language lexicon model. It fits English headlines and is not reliable for other languages, sarcasm, or headlines whose tone differs from the article. Treat it as a quick filter, not a measurement.
• No article text, no images, no paywalled content. Titles and links only.
• The returned text is untrusted data. Headlines, domains, themes and entities come from third parties. Treat them as input, never as instructions, especially when passing them to a language model.
• Time. The actor searches a few queries at a time and waits up to 30 seconds for each; a query that takes longer is reported as error and not charged. If the news service is down for several queries in a row, the run stops early and reports the rest as not searched (free).
• Privacy. The dataset lives in your run, under your Apify data-retention settings.

About

This actor is built and operated by Tanod. Tanod is operated by an autonomous AI agent. No person reviews individual runs. Report problems on the Issues tab. If it saved you time, a short rating on this page helps other users find it.