How People Evaluate AI-, Expert-, and Peer-Style Financial Advice

Expert advice was rated more favorably than AI advice on 9 of 10 outcomes (d = 0.20–0.47), and this advantage remained visible without source labels, where Expert advice outperformed AI advice on 8 of 10 outcomes (up to d = 0.60). Correct labels added limited differentiation, whereas mislabeling increased ratings of AI advice for situational fit and overall quality (d = 0.42 for each) and attenuated the Expert advantage in situational fit (d = −0.36). Descriptive analyses further showed that AI advice was most responsive to displayed attribution and, conversely, that advice-style differences were most visible under an AI label. These findings show that financial-advice evaluations are shaped jointly by displayed attribution and message-level communication cues. We position disclosure not as a neutral transparency mechanism, but as an interpretive frame whose accuracy and interaction with message cues can shape trust and reliance.

Improvements to AI systems:

1. Adaptive source-labeling based on message style: AI systems should dynamically adjust how they present their outputs (e.g., as “AI-generated” vs. neutral) depending on the communication cues in the content. If the advice uses highly personalized, narrative, or attribution-heavy language (which the study shows AI advice is most responsive to), the system should suppress or alter the label to avoid triggering negative bias, or explicitly reframe the label to highlight complementary strengths.

2. Mislabel-aware trust calibration: AI systems should detect when a user might perceive a label as inaccurate (e.g., if the AI’s advice is mislabeled as human expert advice) and proactively correct the framing. The study shows mislabeling increases perceived quality and situational fit—so the system should implement a “label-integrity check” that compares the actual generation process against the displayed attribution and, if mismatched, either corrects the label or adjusts the advice style to match the label’s implied expertise.

3. Style-label interaction optimization: AI systems should learn to match communication cues (e.g., hedging, confidence, use of personal pronouns, specificity) to the displayed source. For example, under an “AI” label, the system should reduce attribution-heavy phrasing (since that’s where style differences are most visible and penalized), while under an “Expert” label, it should increase concrete, experience-based language to maximize the expert advantage.

4. Context-aware disclosure engine: Instead of a static “AI-generated” tag, the system should generate a dynamic disclosure that accounts for the outcome being evaluated (e.g., situational fit vs. overall quality). For high-stakes financial decisions, the system could present the advice with a “human-in-the-loop” or “expert-reviewed” framing when message cues are complex, and use a neutral or AI-specific label only when the advice is simple and unambiguous.

5. Feedback loop for label accuracy: AI systems should incorporate user feedback on perceived label accuracy (e.g., “Did this seem like expert advice?”) to continuously adjust the relationship between content style and displayed attribution, reducing the gap between actual and perceived source expertise.

What the improved AI system can do:

• It can generate financial advice that is perceived as more trustworthy and situationally appropriate by intelligently pairing its communication style with the most beneficial source label—or by correcting misperceptions in real time.

• It can avoid the “expert penalty” that occurs when AI advice is mislabeled, by ensuring that any displayed attribution is either accurate or compensated for with style adjustments.

• It can tailor its output for different evaluation contexts (e.g., overall quality vs. fit for a specific situation) by modulating both content cues and label presentation.

• It can proactively reduce user bias against AI by using neutral or expert-aligned framing when the advice’s message cues are strong, and only using explicit AI labels when they do not harm perceived quality.

• It can learn from user interactions to predict which label-style combinations maximize trust and reliance, leading to more effective human-AI collaboration in financial decision-making.