Retrieval-augmented Generation in Marketing: an Integrative Literature Review of Architectures, Applications and Strategic Implications

Purpose This study synthesises and critically interprets the emerging literature on retrieval-augmented generation (RAG) in marketing. It examines how RAG architectures, applications and governance conditions shape interactive marketing, customer engagement, market sensing and responsible AI deployment. Design/methodology/approach An integrative literature review guided by the SPIDER framework was conducted across Scopus, Web of Science, IEEE Xplore, ACM Digital Library, Google Scholar and Emerald Insight for publications from 2020 to 2025. A final corpus of 25 marketing-relevant RAG studies was analysed through qualitative coding and thematic synthesis. Adjacent peer-reviewed literature on AI in marketing, recommender systems, customer analytics, data-driven marketing capability, AI-augmented co-creation and immersive digital environments was used to contextualise, rather than expand, the primary corpus. Findings Five themes emerge: adaptive and knowledge-graph-enhanced RAG architectures; multilingual, real-time and hyper-personalised customer interaction; intelligent and context-aware retrieval; marketing intelligence and decision support; and robustness, hallucination mitigation and responsible deployment. The review shows that RAG can support retrieval-grounded interaction and market sensing, but evidence is stronger for technical and prototype demonstrations than for longitudinal customer, conversion or strategic outcomes. Responsible deployment requires transparency, privacy safeguards, fairness controls, human oversight and auditable knowledge sources. Originality/value The study advances a marketing-specific interpretation of RAG as a socio-technical marketing capability. It contributes by shifting personalisation from predictive targeting to retrieval-grounded interaction, linking market sensing to dynamic knowledge grounding and positioning governance as intrinsic to the capability rather than a downstream safeguard.