Randomized volunteer intervention reduced comments and reaction on anti-COVID-19 vaccine posts on Facebook

Social correction has been shown to be effective in reducing misperceptions and limiting the production of harmful content online, but most current evidence comes from platform-run systems or controlled experiments. We evaluate a volunteer-led intervention on Facebook in which trained healthcare workers replied to COVID-19 vaccine-related posts and comments. The intervention used two sources of random variation: randomized ordering of posts on volunteers’ worklists and random assignment of whether drafted replies were posted. We estimate effects on post-level engagement metrics, distinguishing diffusion indicators, measured by shares; interaction indicators, measured by comments; and affective indicators, measured by likes and reactions. The intervention reduced comments on treated posts, with broader declines in some affective responses in randomized-treatment analyses, especially for newer posts and posts classified as anti-vaccine. We find little evidence of reduced sharing. Our results indicate changes in some forms of engagement with treated posts; they do not directly demonstrate reductions in misinformation prevalence, belief change, or exposure patterns.