Chatbots have the potential to effect large-scale behaviour change because they are accessible through social media, flexible, scalable, and automatically gather data. Yet research on the feasibility and effectiveness of chatbot-administered behaviour change interventions is sparse. There is no guarantee that established behaviour change interventions are effective when implemented in chatbots given the unique human-machine interaction dynamics. In this pilot study, we test whether an advanced, natural language processing-based chatbot can effect behaviour change using animations embedded in conversations with users. Specifically, we evaluated whether the chatbot could increase understanding and intentions to adopt protective behaviours during the pandemic in a real world cohort. We randomised 59 participants from culturally and linguistically diverse populations to three groups, receiving either a compassion intervention, an exponential growth intervention or no intervention. We measured participants’ attitudes about leaving the house for different reasons and their intentions to get tested if they experienced symptoms. We found that the chatbot reduced uncertainty about appropriate COVID-19 protective behaviours and that the exponential growth bias intervention increased participants’ likelihood of getting tested. This study provides preliminary evidence that chatbots augmented with animations can spark behaviour change, with potential for applications in diverse and underrepresented groups, highlighting the need for further research on this topic.