Purpose This article reviews the expanding literature on artificial intelligence in marketing (AIM) to address a central unresolved challenge. Although AIM is commonly associated with efficiency, personalization and engagement gains, empirical findings remain fragmented and often contradictory. Concurrent concerns related to trust, transparency, surveillance, ethics and organizational readiness further complicate evaluation for firms, consumers and scholars. This study provides an integrative synthesis that reconciles these conflicting narratives and identifies priority and feasibility pathways for future research and practice. Design/methodology/approach Following the SPAR-4-SLR protocol, this study systematically reviews 114 peer-reviewed journal articles published between 2013 and 2025. A dual-framework approach integrates the Antecedents–Decisions–Outcomes (ADO) and Theories–Contexts–Methods (TCM) frameworks. Findings The synthesis identifies key antecedents, decision processes and outcomes. Importantly, the review reveals systematic paradoxes, such as personalization functioning as both empowerment and surveillance, and anthropomorphism as both comfort and discomfort. Mapping these tensions within the ADO–TCM structure explains divergent findings across TCM. Research limitations/implications The review is limited to English-language journal articles indexed in major databases and may underrepresent nonindexed or emerging AIM themes. Practical implications The findings offer managers a decision map to align AI design choices with consumer trust and transparency expectations, and inform policymakers about governance priorities such as disclosure norms, bias audits and workforce reskilling. Originality/value By jointly applying the ADO–TCM frameworks, this study advances prior AIM reviews by transforming fragmented evidence into a conflict-reconciling synthesis and introducing a priority–feasibility agenda for research and practice.
Kukreti et al. (Wed,) studied this question.