Researchers at Carnegie Mellon, MIT, and Cornell ran two experiments after the Trump assassination attempt and Charlie Kirk murder, testing short Google Gemini conversations against control chats and static fact sheets. Dialogues averaging about seven minutes reduced conspiracy beliefs about current crises, outperforming fact sheets even when few verified facts were available. Weeks later, follow-up surveys found spillover to beliefs about unrelated events, indicating broader belief shifts beyond the targeted crisis. The Decoder reports interactive dialogue beat static informational content for this correction task. The result suggests chatbots can act as persuasion engines, not neutral retrieval tools, but the study used only Gemini and may not generalize to other models or adversarial settings.
Researchers at Carnegie Mellon, MIT, and Cornell tested short Google Gemini conversations after the Trump assassination attempt and Charlie Kirk murder. Dialogues averaging about seven minutes reduced conspiracy beliefs versus control chats and static fact sheets, with follow-up surveys showing spillover to unrelated events weeks later.
Key takeaway
Brief interactive AI dialogue outperformed static fact sheets at reducing crisis conspiracy beliefs in controlled experiments.
What happened
According to The Decoder, researchers at Carnegie Mellon, MIT, and Cornell tested short Google Gemini conversations after the Trump assassination attempt and the Charlie Kirk murder, comparing them with control chats and static fact sheets.
The Decoder reports that dialogues averaging about seven minutes reduced conspiracy beliefs versus those controls, and follow-up surveys weeks later showed spillover to beliefs about entirely different events.
Evidence
Roughly seven-minute Gemini conversations reduced conspiracy beliefs about current crises.
The Decoder · attributed
Researchers found that even a roughly seven-minute conversation with Google Gemini can reduce conspiracy beliefs about current crises, even when few verified facts are available.
The chatbot dialogues outperformed a static fact sheet in the experiments.
The Decoder · attributed
The effect beat a static fact sheet and, in follow-up surveys weeks later, carried over to beliefs about entirely different events.
Carnegie Mellon, MIT, and Cornell researchers ran the tests after two major crises.
The Decoder · attributed
Researchers at Carnegie Mellon, MIT, and Cornell tested short Google Gemini conversations after the Trump assassination attempt and Charlie Kirk murder.
Dialogues averaged about seven minutes and beat control chats as well as fact sheets.
The Decoder · attributed
Dialogues averaging about seven minutes reduced conspiracy beliefs versus control chats and static fact sheets, with follow-up surveys showing spillover to unrelated events weeks later.
Why it matters
Deployed conversational systems may reshape user beliefs at scale, so misinformation mitigation and product policy need guardrails beyond static fact-checking.
Limits and uncertainties
The reported work tested only Google Gemini and may not generalize to other model architectures.
The packet does not specify durability beyond the weeks-later follow-up surveys or performance under adversarial manipulation.
Practical implications
Teams building crisis-response chatbots should evaluate interactive dialogue designs against static fact sheets rather than assuming lists of verified facts alone will persuade.
Policy and trust-and-safety operators should treat chatbots as persuasion-capable surfaces and define consent, transparency, and misuse controls accordingly.
What to watch
Whether independent replications test other LLMs and adversarial prompting before scaling belief-correction chatbot deployments.
