Millions of people in Bangladesh drink arsenic-contaminated water despite increased awareness of consequences to health. Theory-based and evidence-based interventions are likely to have greater impact on people switching to existing arsenic-safe wells than providing information alone. To test this assumption, we first developed interventions based on an empirical test of the Risk, Attitudes, Norms, Abilities and Self-regulation (RANAS) model of behaviour change. In the second part of this study, a cluster-randomised controlled trial revealed that in accordance with our hypotheses, information alone showed smaller increases in switching to arsenic-safe wells than information with reminders or information with reminders and implementation intentions.