Beyond the “Rainbow Lollipop”: Uncanny Cues in AI-WOM and Its Impact on Trust Erosion and Negative Consumer Responses

Abstract

This qualitative study examines users’ meta-perceptions of uncanny cues in unsolicited AI-generated word-of-mouth (AI-WOM) on social media and their impact on consumer responses. Since objective verification of authorship is empirically inaccessible, findings strictly reflect user beliefs rather than confirmed AI content. Analyzing data from 29 TikTok videos, the research reveals four distinct themes: (1) perceived uncanny cues characterized by overly perfect “Rainbow Lollipop” language and contextual inappropriateness; (2) users’ negative emotional responses ranging from irritation to existential anxiety; (3) trust erosion regarding authenticity and perceived manipulation; and (4) users’ negative behavioral responses including blocking, reporting, and withdrawal. Extending the Uncanny Valley Theory and CASA Paradigm to text-based interactions, these findings highlight the “dark side” of excessive anthropomorphism, where perceived AI hyper-perfection disrupts social illusions. Practically, it cautions brands against using overly polished AI content, advocating for authenticity and explicit disclosure to preserve the trust ecosystem and prevent user withdrawal.