Widespread misconceptions can be critical, especially in times of crisis. Through a field experiment, we study how to address such wrong or inaccurate beliefs using messages delivered to individual citizens using mobile phones. We focus on misinformation related to the COVID-19 pandemic in a hard-to-reach population – India’s slum residents. We randomly allocate participants to receive voice and video messages introduced by a local citizen, the messenger, and in which medical practitioners debunk misconceptions. To understand the role of targeting, we randomly vary the signaled religious identity of the messenger into either Muslim or Hindu, guaranteeing exogenous variation in religion concordance between messenger and recipient. Doctor messages are effective at increasing knowledge of, and compliance with, COVID-19 policy guidelines. Changes in misconceptions are observed only when there is religion concordance and mainly for religious-salient misconceptions. Correcting misconceptions with information requires targeting messages to specific populations and tailoring them to individual characteristics.