Most open datasets are a download. You fetch a CSV, it is eighteen months old, and the person who made it moved on two grants ago.
GDELT is not that. It updates every fifteen minutes, it goes back to 1979, and the whole thing is free. I spent a week building on it, and the interesting part was not the data. It was finding out who keeps it running.
What it actually is
The GDELT Project describes itself as "a realtime network diagram and database of global human society for open research." That is a large claim, and the numbers behind it are unusually specific.
The Event Database records over 300 categories of physical activity worldwide, georeferenced down to city level, from 1 January 1979 to now. GDELT's own about page puts the archive at "over a quarter-billion event records" across "more than 100 languages." Alongside it sits the Global Knowledge Graph, which pulls people, organisations, locations, themes and emotions out of the same coverage.
The distribution model is the part that matters if you are building anything. From the data page: "The entire GDELT database is 100% free and open and you can download the raw datafiles, visualize it using the GDELT Analysis Service, or analyze it at limitless scale with Google BigQuery." All datasets are in BigQuery, and the live ones update on a fifteen-minute cycle.
No key. No tier. No seat count.
Who built it
GDELT's about page credits Kalev H. Leetaru as its creator. Wikipedia's entry on Leetaru describes him instead as "co-creator of the Global Database of Events, Language, and Tone (GDELT)," naming Philip Schrodt alongside him.
I am not going to pretend to resolve that. Both framings point at something real, and the second one is worth understanding because it explains where GDELT's event data gets its shape.
Schrodt built TABARI in 2000, an automated event-coding program, and co-developed the CAMEO scheme — Conflict and Mediation Event Observations — a taxonomy for classifying who did what to whom in international affairs. GDELT hosts the CAMEO manual on its own servers, because CAMEO is the ontology GDELT's event records are coded into. When a GDELT row says a country made a statement or a protest occurred, those categories are CAMEO categories. That work predates GDELT and outlives any one project; ACLED and the Phoenix Data Project use the same framework.
So: one person built and runs the thing, on top of a coding scheme somebody else spent years getting right. Both facts are true and the site credits both.
Leetaru's own path is less conventional than the output suggests. He joined the National Center for Supercomputing Applications at the University of Illinois in 2000 as an undergraduate. He was the Yahoo! Fellow in Residence at Georgetown University in 2013–14, and is a senior fellow at George Washington University's Center for Cyber and Homeland Security.
Who pays for it
This is the question I actually wanted answered, because "free forever" usually means "free until the grant ends."
GDELT's front page states plainly that it is "Supported by Google Jigsaw." The acknowledgements on the about page run considerably wider, naming Google Ideas, Google Cloud, Google and Google News; the Yahoo! Fellowship at Georgetown University; BBC Monitoring; the National Academies Keck Futures Program; Reed Elsevier's LexisNexis Group; JSTOR; DTIC; and the Internet Archive. Individual contributors named there include Patrick Brandt of the University of Texas at Dallas, Philip Schrodt of the Peace Research Institute Oslo, and John Beiler of Pennsylvania State University.
That is a genuinely unusual coalition: a search company, a broadcaster's monitoring service, an academic publisher, a scholarly archive, a defence technical information centre, and a digital library. Not one of them is an obvious owner of the result, which is probably why the result stayed open.
The Internet Archive collaboration
The piece I ended up leaning on hardest is the television work, and it is not GDELT alone.
The TV News Visual Explorer is a collaboration between GDELT, the Internet Archive's Television News Archive, and the Media-Data Research Consortium. GDELT's retrospective on the Archive's tenth anniversary describes a decade of it: interactive maps of the geography of television news less than a year after the Archive's public debut, then the Television Explorer, then the AI TV Explorer in 2020 offering keyword search across all onscreen text, then the Visual Explorer.
Each broadcast becomes a grid of thumbnail frames, one every four seconds. I can confirm that number from the other direction: when I pulled a BBC Newsnight broadcast apart, the contact sheet was a six-wide grid and the frames stepped four seconds each. The documentation and the bytes agree, which is rarer than it should be.
What it took to build on
I put GDELT on a working news site this week — the Event Map on Apprised, which shows where the world's news is being reported and lets you read what each country's own broadcasters said about it. Four things were worth writing down.
The bulk feed works. The convenience API mostly does not. GDELT's raw 15-minute export files returned every request I made. The document API refused something like 85 to 100 percent of mine across several days and several IPs. If you are building on GDELT, build on the raw files. The polished endpoint is the one that will rate-limit you.
The country codes are FIPS 10-4, not ISO 3166. In FIPS, CH is China. In ISO, CH is Switzerland. This shipped onto four separate surfaces of mine before I caught it, and it caught me because a digest about to be mailed said "CH: 347 events" next to a story about Beijing. If you take one operational detail from this post, take that one.
Their terms are more generous than most people assume. GDELT's terms of use state that "all datasets released by the GDELT Project are available for unlimited and unrestricted use for any academic, commercial, or governmental use of any kind without fee," and that "you may redistribute, rehost, republish, and mirror any of the GDELT datasets in any form," provided every use carries a citation and a link back. That is why Apprised now hosts its own copies of GDELT's daily channel reports rather than only linking to them: plenty of corporate and school proxies block the Google Cloud Storage bucket the originals live in, and a link that never resolves is not a link.
The daily reports are the underrated part. GDELT publishes a per-channel analysis of a full day's coverage for sixty-odd television channels, most of them non-Western state and national broadcasters — Press TV, CCTV, Russia-1, Zvezda, Telesur, Belarus 24. Reading what Iranian state television led with on the same day as the BBC is a different exercise from reading either alone. Those are on Apprised too, with their film strips and transcripts.
Why this is worth your attention
The thing I keep coming back to is the asymmetry. A quarter-billion event records, a hundred-plus languages, forty-seven years of history, updating every fifteen minutes, and the total cost to a small operation is the compute to read it.
There is no version of my building a global event map from scratch. There is not a version where anyone reading this does either. The reason it is possible at all is that one researcher, a coding scheme from the political-science literature, a search company's public-interest arm, a digital library and a handful of universities decided the useful thing was to give it away and keep it running.
That is a rarer arrangement than the internet's usual arc, and it is worth saying so out loud while it is still true.
Fact-check notes and sources
- Tagline, "Supported by Google Jigsaw," and dataset descriptions: The GDELT Project home page
- Creator credit, scale figures ("over a quarter-billion event records," "more than 100 languages," 1979 to present), and the full acknowledgements list including Google Ideas, Google Cloud, BBC Monitoring, LexisNexis, JSTOR, DTIC, the Internet Archive, Patrick Brandt, Philip Schrodt and John Beiler: The GDELT Story: About the GDELT Project
- "100% free and open," BigQuery availability and the 15-minute update cadence: GDELT data access
- Terms permitting unlimited academic, commercial and governmental use, and redistribution/rehosting/mirroring with citation: GDELT terms of use
- Co-creator framing, NCSA (2000), Georgetown Yahoo! Fellow (2013–14) and George Washington University affiliations: Kalev Leetaru, Wikipedia
- TABARI (2000) and the CAMEO coding framework: Philip A. Schrodt, Wikipedia and the CAMEO manual hosted by GDELT
- Visual Explorer as a collaboration with the Internet Archive's TV News Archive and the Media-Data Research Consortium, the decade of joint work, and the M-DRC's Google Cloud COVID-19 Research Grant: The Internet Archive's TV News Archive Turns 10
- One frame every four seconds: stated in GDELT's Visual Explorer description and independently confirmed against the published contact sheets while building the Apprised integration.
The FIPS-versus-ISO country code behaviour and the raw-feed-versus-document-API reliability gap are my own observations from building against GDELT in August 2026, not GDELT statements.
