Described as a semio-pedagogical activity, synchronous online language learning requires socio-affective, pedagogical, semiotic, and technological competencies, including the ability to provide “clear and concise instructions” (Guichon, 2009: 169). Procedural instructions for language tasks are important for several reasons. First, successful task completion “is often predicated on the effectiveness of [the] instructions” (Watson Todd, Chaiyasuk, & Tantisawetrat, 2008: 26). Second, language learning happens in meaning-focused interaction (Nunan, 2004) and instructions offer opportunities for authentic communication (Watson Todd, Chaiyasuk, & Tantisawetrat, 2008). Third, instruction-giving is part of task-based teaching competencies (Raith & Hegelheimer, 2010). In face-to-face contexts, studies have largely drawn on Conversation Analysis to investigate different components of instructions (Markee, 2015), their interactive nature (Somuncu & Sert, in press), and teachers’ verbalisations of written instructions (Ha & Wanphet, 2016). Our understanding, however, of teachers' use of semiotic resources within the complex multimodal affordances of videoconferencing is limited. Indeed, from a multimodal perspective, earlier studies have analysed online teachers’ semio-pedagogical competence (Develotte, Vincent & Guichon, 2010), the way in which a single resource is employed (e.g. gaze, gestures, Wigham, 2017) or largely explored the social or interactive aspects of language learning (Cappellini & Azaoui, 2017, Satar, 2016). The few studies that have examined task instruction-giving from a multimodal perspective have largely been limited to trainee teachers (Codreanu & Combe-Celik, 2012; Cappellini & Combe, 2017; Satar & Wigham, 2017) In this paper, we will present a multimodal (inter)action analysis (Norris, 2019) of screen-recorded lessons to understand variance in an experienced online teacher’s instruction-giving (inter)actions as a result of task repetition. We explore whether higher- and lower-level actions (HLAs and LLAs) in task instructions-as-process differ when the same teacher repeats the same task with two different dyads of University-age learners of B1-B2 CEFR level (Council of Europe, 2001). Lessons were conducted via the videoconferencing platform Skype and adopted TBLT approach to engage learners in authentic language use during a meaningful task and elicit linguistic output (Ellis, 2000). Our analysis for this paper explores a convergent task (information-gap). Our analysis demonstrates that task instructions appear to be more efficient during a second iteration: fewer higher-level actions are employed and, overall, instructions are shorter in length. In terms of lower-level actions, our findings indicate that the teacher uses the same gestures or gesture types across iterations. We will conclude with a discussion of (1) how semiotic misalignment may fragment or distort the shared interactional space and (2) how modal density misalignment can lead to different foregrounding of actions for the teacher and the learners.