AI credibility labels can make a crowd's agreement meaningless.

Social media sites often show AI scores to help people spot misinformation. But when users share or like posts, later users see both the AI score and previous people's choices. This makes it hard to tell why a crowd agrees. The researchers studied this process by modeling how people learn from both an AI indicator and earlier peer judgments. They found that an AI score changes what crowd agreement means. A crowd agreeing might mean many people independently checked the facts, or simply that everyone copied the same AI. This creates a big trade-off. Relying heavily on an AI can preserve correct calls, but it can also trap the crowd into an error by silencing private doubts. Testing real users showed that people usually trust their own view more than the AI, but trust the AI over several peers. Finally, diversifying AI signals across people keeps the crowd smarter.

Why it matters

When you see a warning flag on a social media post, you might think it helps you decide. But this paper shows that if everyone sees the same AI flag, the crowd's agreement on a post's truth might just mean everyone copied the AI, not that they independently verified it. That makes it harder to spot misinformation, and it means platforms need to design these tools differently.

Questions answered