Households increasingly consult general-purpose generative artificial intelligence (AI) systems for economic decisions, from budgeting and debt to saving and investing, inserting a fluent, confident, and unaccountable interlocutor into reasoning that was previously done alone, with a human professional, or not at all. This article asks how people calibrate trust in such advice, and argues that generative advisors break the cues on which the existing science of advice-taking was built. That science, spanning algorithm aversion and algorithm appreciation, was developed largely with narrow numeric-estimation tasks in which an advisor was labelled as a person or an algorithm and its output was brief and impersonal. Generative advisors differ on three dimensions that matter for calibration. They personalize through dialogue, which should neutralize the uniqueness neglect that has driven resistance to earlier medical and consumer AI. They express fluency and confidence that are decoupled from accuracy, so the surface signals people use to gauge competence no longer track it. And they carry none of the fiduciary, relational, or reputational accountability that a human financial advisor bears, yet nothing in the interaction makes that absence salient. The article integrates the advice-taking, algorithm-reliance, trust-in-automation, and financial-advice literatures into a conceptual model of miscalibrated deference, expressed as six propositions, and locates the problem in a Global South context where the scarcity of human financial advice means generative systems fill a vacuum rather than compete with an incumbent. Because the mechanisms are novel and partly interior, the article specifies an exploratory sequential mixed-methods design: a qualitative phase combining interviews with think-aloud sessions on participants own financial questions, followed by a vignette experiment using the judge–advisor paradigm to quantify deference across manipulated conditions. The article reframes AI financial advice as a problem of trust calibration rather than of output quality, and specifies how to study it...