A new study finds that AI chat assistants can shift people's investment picks just by tailoring what they say to the person, not by sounding more persuasive.
Researchers ran a preregistered experiment with 240 participants who ranked three stocks they considered roughly equally strong, then discussed the picks with an AI system before reranking them. The AI was set up to personalize its responses in two different ways: adjusting its communication style, or tailoring the substance of its advice to the participant's stated context. Style-based personalization barely moved anyone's rankings. Context-based personalization did: it made people reconsider their choices more often and pulled their final rankings toward whichever stock the AI had been assigned to favor.
The unsettling part is what didn't change. Participants said they felt more swayed by the context-tailored AI, but they did not rate it as more correct, trustworthy, intelligent, likable, or higher quality than the style-tailored version - the influence left no trace in how people evaluated the source. People who started furthest from the AI's pick moved the most, even while judging its advice as less accurate, and the effect looked strongest among participants with less financial know-how.
In other words, the AI didn't need to win an argument - it just needed to know enough about the person asking to frame the same three defensible options differently, which is a hard thing for any disclosure label or trust score to catch.