AI Study Reveals How Telegram Groups Frame Israel-Palestine Conflict Differently

A new study looked at how pro-Israel and pro-Palestine groups talk about the same conflict on Telegram, a popular messaging app. Researchers gathered 87,617 messages from 16 channels — eight supporting each side — spanning from May 2021 to June 2026. That covers several waves of escalation, making it one of the largest direct comparisons of these communities.

They used three different AI methods to detect the stance of each message — whether it supported one side or the other. A custom-trained model worked best, getting 72% accuracy, beating simpler out-of-the-box tools by 8 to 11 percentage points. In plain terms, generic AI often misses the nuances of how real people talk about conflict, but a model tuned on that specific language can catch more.

The most interesting finding came from combining sentiment analysis (positive or negative tone), stance detection (which side), and framing (how the topic is presented). Both sides frequently used words about death and victims, but the emotional tone was very different. Pro-Israel channels tended to sound neutral and factual, like a news report. Pro-Palestine channels were markedly more negative and emotional — reflecting, the authors suggest, different positions as "acting party" vs "affected party."

Why should you care? Because it shows that the same event can be described with the same words but leave a completely different impression. As Telegram becomes a major source of news for many people — especially during conflicts — understanding these patterns helps you spot emotional manipulation and see how each side tries to shape your opinion. The study isn't perfect: it only covers 16 channels, and AI isn't perfect at detecting tone. But it's a useful reminder that what you read online is rarely neutral.