Extreme content uses unusually intense, radical, or socially divisive language or viewpoints. A short video, slogan, or post often attracts attention more easily than a careful explanation of competing evidence, so nuance disappears while the strongest claim stands out. section 7.5.1
Recommender systems can make this pattern grow. They learn from what people click, watch, or otherwise engage with, then show similar material more often. As an audience expands, competition for attention increases; emotionally charged language can help content enter the most-promoted slice. Repeated exposure may therefore shift a person’s feed toward more extreme positions without clearly signaling that the feed is unusual. section 1.3.3 section 7.5.1
From a narrow feed to a distorted public picture
Suppose only 10% of posts receive interaction. As the network grows, creators compete for that small visible share, and stronger language may improve their chances. A user repeatedly shown conspiracy claims may conclude that mainstream journalists are hiding the truth, even though the feed is reflecting personalization rather than the whole information landscape. Similar systems can expand both constructive movements such as #BlackLivesMatter and #MeToo and harmful misinformation, harassment, or conspiracy theories. section 1.3.3 section 7.6 section 7.5.1
The consequences extend beyond individual feeds. Emotionally charged recommendations can contribute to polarization, a loss of shared reality, and harmful mobilization; in Myanmar, unchecked misinformation and harmful content were linked to an extreme case of social damage. Extreme content is not automatically harmful, but its visibility depends on platform incentives, user behavior, and surrounding institutions. section 7.6 section 7.5.1
