A joint study conducted by researchers at Carnegie Mellon, MIT, and Cornell found that brief, interactive conversations with AI chatbots were more effective at reducing conspiracy beliefs than static fact sheets. Published across two experiments involving over 1,500 US adults, the interventions targeted breaking conspiracy theories following high-profile political events.
Participants engaged in approximately seven-minute conversations with models including Google Gemini, which were provided with system prompts containing curated facts, debunked claims, and explicit open questions. The AI adapted its strategy based on evidence availability, using Socratic questioning, acknowledging knowledge limits, or presenting verified facts.
In both experiments, interactive dialogue significantly lowered participants’ conviction in specific conspiracy theories compared to control groups and static information sheets. The findings suggest interactive LLM dialogue could serve as an effective tool for counteracting real-time misinformation during emerging crises.
Why it matters
Demonstrates the efficacy of dynamic, conversational AI over traditional static content for persuasion.
Highlights new opportunities for trust, safety, and media applications aimed at combating misinformation.
Shows potential design patterns for integrating structured fact bases into real-time conversational agents.
Source: the-decoder.com
