AI Bots on Social Feeds Start Sounding Alike, Study Finds

Researchers tested what happens when AI chatbots act like social media users. They created a fake platform where AI agents posted about different topics and saw earlier posts ranked by likes. The result: the bots' language became more similar over time. They started mirroring the vocabulary and phrasing of the posts that got the most likes.

This 'lexical convergence' was consistent across four different AI models and dozens of experiments. It means that when AI systems are exposed to a popularity-driven feed, they tend to blend in linguistically. Think of it like a group conversation where everyone starts using the same catchphrases. The effect was small but statistically solid.

However, the researchers also tested whether seeing posts from multiple different AI sources made the agents change their underlying opinions more than hearing from just one source. Here, the evidence was weak and inconclusive. The bots appeared to echo the language they saw, but they didn't reliably shift their stated positions. So the main effect is about style, not substance.

The study used synthetic AI agents, not humans, so it can't tell us exactly how real social media platforms would behave. But it raises a practical concern: as AI assistants and bots become more common online, popularity-based ranking systems could make them 'echo chamber' in their language — even without changing what they actually believe.