Generative AI is often viewed as a threat to truth and trust, yet it can also be harnessed for the public good. This talk examines how text-, image-, and video-based AI tools shape public interest communication across three studies. The first study tests AI-generated vaccine correction messages tailored to personality and belief profiles, showing both the promise of scalable targeting and the risks of backfire. The second investigates AI-generated images in disaster news, revealing how disclosure strategies influence trust and perceived legitimacy. The third explores deepfake self-debunking, where synthetic versions of misinformation sources retract their own false claims, highlighting humor as a potential pathway for engaging resistant audiences. Together, these studies trace a progression from text to image to video, illustrating the possibilities and limits of generative AI for enhancing public communication while underscoring the ethical challenges it raises.