The Two News Feeds That Teach AI About the World

Every time an AI chatbot answers a question about current events, it's drawing on enormous piles of news articles that researchers collected ahead of time. Two of the biggest piles are GDELT (short for Global Database of Events, Language, and Tone) and Common Crawl News. A new paper from researchers Ameir El Ouadi and David Beskow put them side by side to see how they actually differ.

The two collections are built in very different ways. GDELT gathers news from broadcasts, print publications, and websites across the globe — it's designed to track who did what to whom, in real time. Common Crawl builds its collection by automatically visiting news websites around the world and copying what it finds. That sounds similar, but the researchers found the two end up pulling from surprisingly different parts of the world.

Why does that matter to you? Because these datasets are the raw material behind AI tools you may already use — search engines, news summarizers, translation apps, and chatbots that answer questions about politics or world events. If one collection leans heavily on certain countries or certain types of outlets, the AI trained on it inherits that blind spot. A story that's front-page news in one region might be nearly invisible to a system fed by the other.

The takeaway isn't that one dataset is better. It's that 'the news' as AI sees it isn't neutral or complete — it's a snapshot shaped by how it was collected. The authors argue researchers should understand each source's strengths and gaps before using either. For the rest of us, it's a reminder that when AI tells you what's happening in the world, someone chose the ingredients.