GDELT

The Global Database of Events, Language, and Tone (GDELT) is a continuously updated record of what the world's news media is reporting. It parses articles from print, broadcast, and web sources in more than 100 languages, and every 15 minutes it extracts the events, people, organisations, locations, and themes they mention and scores their emotional tone. The result is one of the largest open datasets about human society, growing by well over 2.5 TB a year.

Two datasets sit at its heart. The Event Database codes who did what to whom, where, and when, using the established CAMEO scheme, with records reaching back to 1979. The Global Knowledge Graph (GKG) connects the entities, themes, emotions, and narratives found across coverage, so you can trace how a single story or topic spreads and shifts across outlets and countries over time.

The easiest way in is Google BigQuery, where the whole thing is a public dataset you can query with SQL without downloading a byte. You can also download the raw files directly (they are published as CSV on a rolling schedule) or use the DOC and GEO APIs for lighter, targeted queries. For most RAG work BigQuery is the sensible route: filter to the events, dates, or entities you care about, then pull just that slice into your own store.

GDELT shines for temporal and current-affairs questions, media analysis, and tracking how sentiment around an event or organisation evolves. The important caveat is what it is: GDELT records what the news said, not verified ground truth, so it inherits media bias, duplicate coverage, and occasional coding errors. It stores metadata and links about articles rather than the full article text, so you often need to fetch and enrich the underlying pieces yourself. On licence, GDELT is free for research and commercial use with no attribution required, which makes it unusually easy to build on. Pair it with the Internet Archive or Common Crawl when you need the actual article text behind the events it catalogues.