This study investigates how people respond to human versus AI financial advice in a state-dependent investment context. In a laboratory experiment, participants are given the choice between two assets with equal expected returns but differing risk levels. Advisors, informed of the true state of the world, provide open-ended recommendations. The AI advisor (ChatGPT) is trained in human-generated messages to ensure comparable content and delivery style. The design features a 2*2 between-subjects design, varying whether participants are informed about the advisor's financial incentives and the advisor's identity, along with the within-subject variation in the type of advisor. Participants take on more risk when advice is provided, particularly when advisor incentives are undisclosed. AI advice is more persuasive than human advice, but only when the advisor is explicitly identified as AI. These findings reveal a context-sensitive trust premium for AI-generated advice, highlighting the behavioral relevance of transparency in advisory interactions and implications for the design and regulation of algorithmic decision aids.