The Effect of AI Information Flow Acceptability on Creepiness and Receptivity of Recommendations

This study examines how AI information flow design—specifically input modality (speech vs. text) and data provenance transparency (stated vs. inferred)—affects users’ perceived creepiness and receptivity to algorithmic recommendations. Grounded in privacy as contextual integrity, we argue that less acceptable information flows, such as speech-based disclosures and inference-based transparency, violate users’ privacy expectations and thereby increase feelings of creepiness. We propose that creepiness serves as the key affective mechanism linking information flow design to reduced recommendation receptivity. Two between-subject experiments using a realistic AI-driven art recommender system test these effects in situ. The study contributes to privacy and IS literature by demonstrating how contextual misalignment in AI information flows shapes emotional responses and downstream engagement with recommendations.