What is GDELT? A practical guide to the world's largest open news database (2026)

GDELT (the Global Database of Events, Language and Tone) is an open project that reads the world's online news around the clock, in more than 100 languages, and turns it into structured data: who is mentioned, where, about what, and in what tone. It is free, updates every 15 minutes and is one of the largest open datasets about the news ever built.

This guide explains what is inside GDELT, how to use it without writing much code, and where it can mislead you. We use GDELT every day: every number on Mirada starts there.

What GDELT is, in one paragraph

GDELT was started by data scientist Kalev Leetaru and has been running in its current form (GDELT 2.0) since 2015. Every 15 minutes it collects newly published articles from tens of thousands of news sites worldwide, machine-translates articles in 65 languages into English, and extracts people, organisations, places, themes and emotional tone from each one. The result is published as open files and through free APIs.

What data is inside

GDELT is not one table but several related datasets:

• GKG (Global Knowledge Graph) — one record per article: URL, outlet, date, the people, organisations and locations mentioned, themes (such as "protest" or "economy"), and tone scores. This is the part most people mean by "GDELT".
• Events — "who did what to whom" records coded in the CAMEO scheme (for example, a government threatening another government), with locations. Historical event data goes back to 1979.
• Mentions — every time an event is mentioned by a new article, which lets you measure how much attention an event gets.
• Specialised collections — television news transcripts (TV API), global geography of coverage (GEO API) and others.

How often it updates and how big it is

New files are published every 15 minutes. On a typical day GDELT processes hundreds of thousands of articles; on Mirada we see about 300,000 articles a day from roughly 11,000 outlets after filtering. Within 15–20 minutes of publication an article can already be counted.

The card above is built from GDELT data in real time: it shows how many articles about the war in Ukraine appeared in the last 30 days and how coverage changed this week.

How to use GDELT without code

The quickest way to explore GDELT is through tools built on top of it:

1. Mirada — type any topic, person or country into the search and get a ready page: articles per day, countries, languages, outlets, tone and the latest headlines, in four languages. No setup and no SQL.
2. GDELT's own explorers — the project publishes web interfaces for searching news and TV coverage; they are powerful but designed for researchers.
3. Spreadsheets — the DOC API (below) returns CSV, so you can paste an API link into Google Sheets with IMPORTDATA.

The DOC API: free queries in a browser

The DOC 2.0 API searches the last three months of coverage and returns articles or timelines. Paste this into a browser:

https://api.gdeltproject.org/api/v2/doc/doc?query=%22climate%20change%22&mode=timelinevol&format=json&timespan=30d

Useful parameters:

• query — keywords, phrases in quotes, sourcecountry:, sourcelang:, domain: and tone<-5 style filters;
• mode — artlist (articles), timelinevol (share of coverage over time), timelinetone (average tone), tonechart;
• timespan — for example 24h, 7d, 30d, up to about three months;
• format — json, csv or html.

The API is rate-limited and meant for light use. For heavier work use BigQuery or the raw files.

BigQuery and raw files: the full archive

For research at scale, GDELT is mirrored in Google BigQuery (public dataset gdelt-bq). A query like this counts daily articles mentioning a person:

SELECT DATE(_PARTITIONTIME) AS day, COUNT(*) AS articles
FROM `gdelt-bq.gdeltv2.gkg_partitioned`
WHERE _PARTITIONTIME >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL 30 DAY)
  AND V2Persons LIKE '%elon musk%'
GROUP BY day ORDER BY day

BigQuery bills by the data scanned, so always filter by partition date. The raw 15-minute CSV files are also downloadable for free; this is what Mirada ingests into its own database.

How tone is measured

Every article gets a tone score from a dictionary of emotionally loaded words: positive words push it up, negative words down. Scores usually range from −10 to +10; most news is slightly negative because conflict, crime and disasters dominate coverage. Tone is useful for comparisons — between countries, outlets or weeks — and much less useful as an absolute verdict on a single article. We explain the method in detail in our guide to news sentiment analysis.

Limits you should know about

• Coverage is uneven. Large international outlets are well covered; small regional and paywalled media much less.
• Machine translation and entity extraction make mistakes, especially with rare languages, transliterated names and namesakes. Expect fragments like photo credits being read as names.
• Duplicates. The same agency story republished by hundreds of sites inflates counts unless you filter by unique outlets — Mirada only creates a page for a person once at least 25 different outlets cover them in a week.
• Headlines, not full text. GDELT shares metadata and links, not the articles themselves.

FAQ

Is GDELT free?

Yes. The data, the raw files and the APIs are free. Querying the BigQuery mirror costs money beyond Google's free monthly quota.

How far back does GDELT go?

Event data goes back to 1979; the article-level Global Knowledge Graph starts in 2015.

Does GDELT include social media?

Mostly no: it focuses on online news sites and broadcast transcripts. Mirada adds public Telegram channels on top.

What is the easiest way to see GDELT data for a topic?

Search the topic on Mirada: you get a ready-made page with charts, countries, tone and the latest articles, updated every 15 minutes.