Study: While Social Media Rewards Sensationalism, LLMs Steer Conversations Toward Civil Discourse

A new analysis published in the Financial Times finds that while social media platforms algorithmically reward inflammatory and sensationalist content, large language models tend to guide users toward more measured, civil communication, presenting an intriguing contrast in how the two technologies shape human discourse. Social media's engagement-optimization model has been extensively documented as a driver of polarization and misinformation. The finding that LLMs produce a different dynamic. nudging conversations toward de-escalation and nuance rather than conflict. has implications for how policymakers and researchers frame AI's social impact. The analysis notes the comparison is not straightforward: social media is optimized for engagement at scale, while LLMs are optimized for helpfulness in one-on-one interactions. Whether LLM-mediated discourse at social-media scale would retain the same properties is an open question.