Brand relationships in conversation: a review of conversational agents using bibliometrics and language models

Purpose This paper aims to synthesise research on conversational agents in the context of brand relationships to map the current state of knowledge, propose an integrative framework and develop a future research agenda. Design/methodology/approach This study analyses 293 publications from Scopus and Web of Science (2004–2025), combining bibliometric analysis, topic modelling and a multi-step validation protocol based on embedding analysis and expert consensus. Findings Five thematic clusters explain how conversational agents shape brand relationships through functional competence, social simulation, anthropomorphic projection, technological affordances and emotional resonance. An integrative framework structures these findings into an explanatory model linking configurable design features, psychological mediators and brand relationship outcomes. The current boundaries of knowledge are defined by three key tensions: the anthropomorphism paradox, the stage-contingency problem and the intensity asymmetry between positive and negative outcomes. Originality/value To the best of the authors’ knowledge, this is among the first reviews to examine conversational agent research through the brand relationship perspective, combining bibliometric mapping with language-model topic modelling and expert validation. The proposed framework bridges descriptive synthesis and mechanism-based modelling, and the expert validation protocol helps address the subjectivity concerns associated with traditional literature reviews.