This paper addresses the growing use of AI assistants, particularly voice-based assistants (VBAs), for information seeking by introducing a novel integration of media trust into AI trust research. While prior studies predominantly draw on interpersonal trust frameworks rooted in the CASA paradigm, which assumes that users apply social norms to human-like technologies, this perspective has been increasingly questioned as users gain experience and adapt their interactions. The study combines the qualitative analysis of online comments on different VBAs with in-depth interviews conducted in users’ homes and finds shifting interaction norms between new and experienced users, challenging CASA’s assumption of automatic social responses to anthropomorphic cues. Instead, experienced users’ trust appears to be grounded in accumulated experience and informed by technical and media-related trustworthiness evaluations rather than by social cues. By highlighting users’ subjective perspectives, the study contributes to a more differentiated understanding of trust in VBAs.