Taking AI Advice in Crisis: How AI Anthropomorphism and Regulatory Focus of Advice Shape Advice-Taking

Abstract Prior evidence is inconsistent about when people follow artificial intelligence (AI) rather than human advice, and crisis contexts are rarely tested. Crises are urgent, unpredictable, and limit deliberation, which can alter advice use. We address this gap with two preregistered studies asking whether crises increase advice-taking from AI relative to humans and whether effectiveness depends on a fit between an AI’s anthropomorphic design and the advice’s regulatory focus. Both studies employed the judge-advisor system (JAS), utilizing Weight of Advice (WOA) to index advice-taking. Study 1 used a mixed design: context was within-subjects (crisis vs. non-crisis), and advisor type was between-subjects (AI vs. human). During crises, advice-taking from AI was higher than from humans. Study 2 focused on crises and used a between-subjects design varying advisors (human, anthropomorphic AI, non-anthropomorphic AI) and advice framing (prevention-focused vs. promotion-focused). An interaction emerged: prevention-focused advice was taken more from anthropomorphic AI, whereas promotion-focused advice was taken more from non-anthropomorphic AI; humans showed intermediate uptake. Results show that crises amplify reliance on external advice and tilt it toward AI, consistent with bounded rationality. AI influence is design-sensitive: aligning AI persona with message framing can strengthen crisis decision support.