How Al Jazeera, BBC, Reuters, RT, Xinhua and others describe the same events in the US-Israel-Iran war — analyzed by AI
The Global Database of Events, Language, and Tone (GDELT) indexes news from thousands of outlets in real time. We query the API hourly for conflict-related articles, filtering by domain for the 7 selected sources.
Each headline and lead is sent to Claude Haiku (Anthropic) with a prompt that extracts: narrative frame (7 categories), emotional tone (5 categories), and revealing vocabulary. The model does not judge factual accuracy — it classifies only how the text was written.
We compare each source's frame distribution against the daily global median using Bhattacharyya distance — a statistical metric measuring overlap between two probability distributions. Score 0 = the source describes the conflict exactly like the average. Score 100 = completely different framing from all others.