The introduction of generative AI (GenAI, e.g., ChatGPT) has transformed second language (L2) writing practices. Writers need to learn how to critically employ the tools’ affordances to maximise their learning and L2 instructors face the challenge of guiding students through these novel contexts (cf. Sasaki, 2023). Also collaborative writing (CW) interventions have to find ways how to include GenAI as another technology that impacts the effectiveness of CW as an instructional L2 writing tool (cf. Storch, 2022; Su et al., 2023). In this study, we present an in-depth analysis of how GenAI influences revision and deliberation processes during CW within four writing pairs (N = 8; aged 18–23) engaging in a classroom-based intervention that tasked them to compare their own writing with GenAI models. We analysed screen recordings of evolving revision behaviours, changes in text quality as well as joint discussions among pairs. Complementing Strobl, Menke-Bazhutkina, Abel, and Michel (2024), the current paper (i) relates writers’ revision behaviour to text quality evaluated for functional adequacy (Kuiken & Vedder, 2017); and (ii) presents a fine-grained review of dialogues to identify patterns of CW revisions in interaction with GenAI models. Findings shed new light on CW models (Storch, 2009) in the digital era where GenAI is implemented in L2 writing and revision instruction.