When AI Is Allowed to Lie: Testing the Impact of Honest vs Dishonest LLM Persuasion Across Issues

• In Event: Lightning Talk: Information, Media, and Online Political Behavior

Brief Overview

We test if allowing an LLM to lie alters its persuasive impact. In our experiment, participants talk to either an honest or dishonest model about one of four policy issues. The results show how constraints on accuracy shape AI-driven persuasion.

Author

• David Rand, Cornell University