In recent years, significant momentum has built up in efforts to integrate the social with the cognitive in theoretical models of speech production/processing and phonological representation. While acknowledging these advances, we argue that what limits our ability to elaborate models of processing and representation in which social‐indexical properties of speech are effectively integrated is that we remain some way from fully understanding how these properties are manifested within spoken interaction in the first place. We explore some of these limitations, drawing on data from a study of sociophonetic variability in a population of speakers of Australian English. We discuss issues relating to methods for capturing variability in the realization of vowels and consonants, and we highlight the pivotal role of speech style and the challenges that this raises for models of production and processing.