Dalarna University's logo and link to the university's website

This study examines how perceived ad personalization triggers a dual cognitive pathway: perceived relevance and perceived intrusiveness, and how these pathways shape ad attitude and ultimately click-through intention, with platform trust moderating the personalization-intrusiveness relationship. A further aim was to establish whether this mechanism operates equivalently across social media and conversational AI advertising environments. 

Drawing on the Elaboration Likelihood Model (Petty & Cacioppo, 1986), the Hierarchy of Effects (Lavidge & Steiner, 1961), and Privacy Calculus Theory (Culnan & Armstrong, 1999), a quantitative experimental vignette survey was administered to 400 respondents distributed across Social Media (n = 219) and ChatGPT (n = 181) platform conditions. A multi-group covariance-based structural equation model was estimated using lavaan in R. 

Findings largely supported the proposed dual-pathway model. Perceived ad personalization significantly triggered both relevance (β = .696) and intrusiveness (β = .118), with the relevance strongly demonstrating more path activation. Ad attitude was shaped by both pathways: positively through relevance (β = .401) and negatively through intrusiveness (β = −.564), with intrusiveness carrying the heavier affective weight despite its weaker activation. Attitude positively predicted click-through intention (β = .567). Serial mediation through the relevance chain was observed, but the intrusiveness-based chain was partially supported. Platform trust demonstrated as a significant moderating effect of the personalization-intrusiveness relationship, with stronger buffering effects observed in the ChatGPT condition. 

Multi-group analysis revealed structural differences across platforms. The personalization-intrusiveness path was significant in the Social Media condition but non-significant in ChatGPT, which indicates that goal-directed conversational engagement attenuates surveillance signal processing. The affective and conative stages of the Hierarchy of Effects operated equivalently across both platforms. This suggests platform sensitivity in cognitive triggering but platform neutrality in affective resolution. 

Theoretically, this study contributes to advertising literature by establishing platform context as a boundary condition of the dual-pathway model and validating the Elaboration Likelihood Model in a generative AI advertising context, with a clear empirical distinction between activation strength and affective weight of competing perceptual pathways. In practice, findings suggest that advertisers may benefit from transparency in early engagement stages, careful ad designs with minimal intrusiveness signals, and platform-tailored personalization strategies that prioritize relevance in AI contexts. 

Limitations include cross-sectional design, sample heterogeneity, single-item operationalization of click-through intention, and the novelty of ChatGPT advertising at the time of data collection. Future research may consider longitudinal designs, true latent moderation estimation, ad avoidance as an alternative behavioral outcome, and cross-cultural replication to further delineate the boundary conditions of the dual-pathway mechanism across the expanding landscape of digital advertising platforms.