Large Language Models (LLMs) like ChatGPT are becoming part of everyday life, in education, mental health, and personal decision-making, often replacing human professionals as a first point of contact. Users frequently describe the responses they receive as surprisingly accurate, insightful, or even tailored to their personal needs. Why do these AI-generated replies feel so compelling? This study explores whether part of the answer lies in the Barnum effect—a well-known psychological bias where people interpret vague, generally positive statements as personally meaningful, especially when they believe the source is authoritative.


While the Barnum effect has been widely studied in contexts like horoscopes or personality feedback, its role in shaping responses to generative AI has not been systematically examined. To investigate this, I designed two studies. In the first, participants will complete measures of cognitive reflection, so-called bullshit receptivity, and trust in AI. They will then submit an open-ended advice-seeking question and receive a response generated by an LLM. They will rate its accuracy, personal relevance, helpfulness, and insightfulness. In the second study, participants will evaluate general advice-seeking responses either generated by an LLM or taken from real psychological self-help materials. Using a 2×2 design, half of the responses will be labeled as AI-generated and half as human-written, regardless of their actual source. This will allow us to disentangle the effects of content quality from source attribution.


Together, the studies will test whether LLM-generated responses are consistently rated as more meaningful—even when generic or misattributed. Findings may suggest that their persuasive power stems not from true personalization, but from broad phrasing, confident tone, and the illusion of expertise. I will discuss implications for trust in AI and the cognitive biases that shape our interactions with it.