We recruited 285 U.S. adults and ran a randomized vignette study covering eight typical financial decisions. The design independently varied three advice styles—AI, expert, and online community—while keeping the underlying recommendation constant and toggling source labels.
Findings reveal that advice style most strongly drives message and safety perceptions; expert labels selectively boost perceived source knowledge; and the decision context primarily shapes risk and safety assessments.
These appraisals were linked to downstream judgments: our models accounted for 69.2% of overall quality, 75.9% of trust, and 82.9% of intended reliance. Expert‑style advice remained the most preferred even when source labels were omitted.
The results suggest that financial AI should be engineered to support grounded evaluation of advice rather than merely maximizing trust.
Review: The study provides a rigorous decomposition of how style, labeling, and context interact, offering valuable guidance for building trustworthy financial AI systems.