Large-scale news and events datasets useful for temporal RAG applications, current affairs, and media analysis.
3sources
News, events, and media datasets capture the world as it happens: articles, headlines, and structured records of events over time. They matter for a RAG (Retrieval-Augmented Generation) system whenever recency and chronology are the point, such as current affairs assistants, media monitoring, or any system that needs to answer “what happened, and when”. Because news is timestamped, it is also one of the better sources for building systems that reason about how a story developed.
The key distinction is between full article text and event metadata. Some datasets give you the actual body of news stories, which is what you want for detailed retrieval and summarisation. Others, such as global event databases, give you structured records of who did what, where, and when, extracted from coverage rather than the coverage itself. One is rich text you can quote; the other is a queryable index of events. Pick based on whether your users want to read the story or analyse the pattern.
The thing to watch here is licensing, more than almost anywhere else in the directory. News articles are usually copyrighted by their publishers, so a dataset may grant access for research while restricting redistribution or commercial reuse. Event and metadata sets are typically more open, precisely because they do not republish the original text. Always confirm what you are allowed to store and serve before you put news content in front of users.
The sources below range from raw article collections to global event databases and long-running web archives.
A subset of Common Crawl focused on news, containing millions of articles pulled from news websites around the world. It has fed several large language model training pipelines and gives you a ready news corpus without crawling sites yourself.
GDELT, the Global Database of Events, Language, and Tone, monitors print, broadcast, and web news in more than 100 languages from every country. It records events back to 1979, builds a Global Knowledge Graph, and adds sentiment and emotion analysis, all queryable on Google BigQuery.
A nonprofit digital library giving free access to millions of books, films, audio recordings, software, archived websites, and television news. Its Wayback Machine has saved more than 800 billion web pages, making it a deep well of historical and current text.