Semantic persuasion versus algorithmic recommendation: A journey-stage aware analysis of native advertising effectiveness

Digital advertising formats differ not only in creative execution but also in how they establish relevance within algorithmically mediated environments. However, limited research has examined how their effectiveness varies across behaviourally inferred positions in the customer journey. Drawing on theories of persuasion, information processing, and computational relevance modelling, this study compares native and catalogue advertising across operational funnel-stage categories and examines how semantic and behavioural signals are associated with user responses. Using 120,000 impression-level observations from performance-oriented e-commerce campaigns, we model click-through and post-click conversion behaviour as functions of advertising format, probabilistically inferred funnel-stage category, pre-exposure behavioural signals, and semantic alignment between advertisements and surrounding content. The TOFU, MOFU, and BOFU categories are derived from observable behavioural patterns and are treated as funnel-stage proxies rather than direct measurements of cognitive states or purchase intent. Results reveal a systematic crossover across the inferred categories. Native advertising achieves significantly higher engagement than catalogue advertising in the early-stage category, whereas catalogue advertising achieves higher click-through rates in the late-stage category. Semantic congruence is positively associated with native-advertising engagement in the early-stage category, but this association attenuates in later categories. Model diagnostics indicate that semantic features contribute more strongly to native-advertising predictions, whereas catalogue-advertising predictions rely more heavily on observable behavioural signals. Post-click conversion patterns are interpreted cautiously because the format-by-stage interaction is not statistically significant. Overall, advertising effectiveness varies across behaviourally inferred funnel-stage proxies, while psychological explanations remain theoretical because they were not directly measured.