The paper introduces a frequency-driven lexical bundle approach to enhance politeness research by addressing the limitations of traditional corpus pragmatics approaches, which assume pre-defined (im)politeness values for linguistic forms. While politeness is widely recognized as a dynamic context-dependent phenomenon, corpus pragmatics research risks overlooking its variability by focusing on pre-selected politeness formulae, such as thank you, please, sorry, etc. The lexical bundle approach builds on the concept of conventionalization, defined as the process by which linguistic expressions acquire politeness-related meaning through repeated use in a specific context. The proposed approach focuses on the identification of recurring multi-word sequences (lexical bundles), capturing how they index (im)politeness in discourse. We demonstrate the utility of this approach through a case-study of Wikipedia editors’ communication, where politeness is strategically negotiated in collaborative, consensus-driven dialogue. By analyzing discussions from Wikipedia talk pages, we categorize the extracted bundles into two groups: conventionalized politeness formulae, which consistently index collaborative politeness across Wikipedia’s editorial discourse (e.g., what do you think, is it possible to) and multifunctional expressions, that manifest variable politeness-related meanings (e.g., why don’t you, am I missing something). This distinction reveals how politeness emerges through both stable norms and context-sensitive strategies. Our findings also highlight the capacity of the new approach to bridge Terkourafi’s frame-based theory with empirical analysis, offering a replicable framework for studying (im)politeness as a co-constructed interactional practice.