Methods to enable comprehensive corpus analyses of multimodal data are imperative to furthering our understanding of social media and digital communication. In this study, we demonstrate how ATLAS.ti version 24.2.0 can be used to construct such corpora and conduct multimodal corpus-assisted discourse studies (MCADS). The focus for such studies may be the exploration of complex patterns of co-occurrence, both intra- and inter-mode, or iterative corpus queries, especially when unexpected patterns move the research focus beyond initial research questions. In such cases, ATLAS.ti’s functionalities facilitate a triangulation of automatic pattern recognition and in-depth manual analysis. It supports a flexible, user-defined approach to the multimodal analysis of short-form social media datasets, overcoming traditional limitations such as analyses being restricted to emojis, or pre-set thematic dimensions in current AI software. In this way, the proposed methodology enables in-depth MCADS, as illustrated in the case study presented in this paper, which analyses the co-occurrences of evaluations and visual representations of social actors.