Corrected by AI, Experts, or Strong Ties? Comparing the Effectiveness of AI and Human Agents in Health Misinformation Correction

AI agent in online content moderation including misinformation correction is becoming prevalent. Yet little is known about how AI compares with different types of human correctors, or whether specific corrective message formats amplify or undermine correction effectiveness. Integrating online opinion leadership theory and stereotype content model, this study investigates how correction source and format jointly influence misinformation correction outcomes. A 3 (correction source: AI-driven algorithm vs. medical expert vs. strong-tie individual) × 2 (correction format: numerical vs. narrative) between-subjects online experiment exposed initial misinformation believers (N = 576) to six different corrections. Results show that expert corrections outperform both AI and strong-tie corrections. Numerical corrections generally enhance credibility perception and sharing intention; however, a significant interaction reveals that narrative corrective messages are most persuasive when delivered by experts, whereas numerical corrective messages perform best for AI-based corrections. These findings advance theorization of AI as an emerging opinion leader and offer practical guidance for designing effective human–AI collaboration strategies in misinformation governance.