AI-powered chatbots become increasingly embedded in everyday digital interactions. But the role of anthropomorphic design features in shaping user trust remains underexplored across contexts with differing stakes. This study investigates how anthropomorphic cues influence trust – measured through perceived benevolence, integrity, and ability – as well as the intention to follow AI-generated advice, in both high-stakes (healthcare) and low-stakes (retail) environments. Using a 2 × 2 between-subjects experimental design (N = 163), we find that anthropomorphism significantly increases all three trust dimensions and the intention to follow chatbot advice in low-stakes contexts, with trust levels increasing by over 30% compared to non-anthropomorphic versions. In contrast, in high-stakes settings, anthropomorphism improves perceived benevolence but has no significant effect on integrity, ability, or behavioural trust. These findings highlight the boundary conditions of anthropomorphic design: while beneficial in domains with comparatively limited consequences, such as retail, related cues do not increase professionalism or competence in more serious domains with a higher risk profile. Our findings offer actionable guidance for tailoring chatbot design to context, informing the deployment of trustworthy AI systems in both commercial and sensitive domains.