Do users trust extroverted AI advisors despite worse advice?

In a lab-based user study with 64 participants, LLM-advisors configured with different personality traits were compared on a financial advisory task built around scripted investor profiles. The excerpt states that "users reported a preference for and increased satisfaction as well as emotional trust with LLMs adopting an extroverted persona, even though those agents provided worse advice." The conclusion restates the finding directly: "even though the extroverted advisor provided lower-quality advice, participants trusted this advisor more than the conscientious one."

The study's method split each advisor interaction into a preference-elicitation stage and an advisory-discussion stage, and compared personalized against non-personalized advisors as well as an extroverted against a conscientious personality layered onto the personalized condition. The paper frames the personality result as evidence that "users appear insensitive to the quality of advice being given, or worse these can have an inverse relationship" — across the personality conditions tested, perceived trustworthiness and actual advice quality moved in opposite directions.

This sharpens the trade-off described in Does chatbot personalization build trust or expose privacy risks?: that note shows personalization raises trust and anthropomorphism together with privacy concern, but this study adds that trust can be decoupled entirely from advice quality, with the advisor's personality presentation — not personalization accuracy or informational content — driving perceived trustworthiness. It also sits alongside Can a seller infer what a buyer really wants to pay?, which shows personas leaking information to the other side of a transaction; here the persona effect instead runs through the advisor's own presentation toward the user receiving advice.

The excerpt does not name which underlying model or model family was used, and the finding comes from a single lab study in which participants role-played three scripted investor narratives rather than making real financial decisions with real money — the paper does not establish that the trust/quality inversion would hold outside this controlled setup or in other high-stakes advisory domains. If it does generalize, it implies that the advisor persona least likely to be challenged by users may not be the one giving the best advice, raising a fiduciary-duty question about whose interest a personality-tuned advisor actually serves.