The authors seek to understand, on the one hand, where and when the comment space on the YouTube platform comes to be affected by brutal or aggressive debates and, on the other hand, which factors limit or promote this aggressiveness. To achieve this goal, they employ a device positioned at the edge of automated language processing and media political economy. They base their study on 2,209,206 comments, of which 1,184,859 are found in the threads, distributed in the comment spaces of 46,090 videos. These videos come from a panel of 57 French media channels with different institutional categories and positions.