AI writes, we collaborate—or vice versa? Group strategies for using generative AI in collaborative writing assignments

Synthesis: Korchak, Costley, and Fanguy (2026) explore how university students use Generative AI (Gen-AI) during the preparation of a group Writing assignment in a scientific writing course at a large Korean university. Semi-structured interviews were conducted with 10 postgraduate students (ages 23–39, M = 27.7) majoring in AI research, and the data were analyzed using Braun and Clarke's (2006) thematic analysis, with two independent coders working in QDA Miner Lite software and a third coder resolving disagreements. Triangulating interview data with students' written self-reflection notes and their group writing documents (Introduction and Discussion sections of an academic paper), the analysis revealed three interconnected themes: (1) group strategy, (2) strategy characteristics, and (3) strategy considerations. The findings reveal that groups used Gen-AI either through pre-planned strategies or spontaneous, individually driven interactions coordinated through shared documents, and that Gen-AI supported not only language and ideation but also integration, helping merge parallel sections into coherent text — underscoring the need for coordination and verification in Collaborative Learning.

Key Findings

Three emergent themes. Thematic analysis revealed three themes: group strategy (how students establish and implement strategies of collaborative work with Gen-AI), strategy characteristics (the specific Gen-AI functions students use, technical and non-technical), and strategy considerations (perceived benefits, limitations, and other factors affecting group uses of Gen-AI). The coding was refined through four iterative analysis stages.

Group appropriation of Gen-AI. Students adopted varied strategies, ranging from explicitly negotiated, pre-decided group plans (e.g., jointly generating an AI draft of the Introduction, then individually rewriting paragraphs while watching others' progress) to "open," individually driven strategies coordinated through shared documents like Google Docs. Notably, students' interview accounts sometimes contradicted their written reflections (e.g., one student denied having a group strategy yet described one in writing), showing that group strategies were not always equally visible to all members.

Varied roles for Gen-AI. Gen-AI acted as a multifunctional collaborator, used for technical tasks (grammar correction, translation, style and tone enhancement, finding words/phrases) and non-technical tasks (brainstorming, idea generation, developing basic arguments, producing first drafts). A distinctive role was Gen-AI as an editorial integrator, merging parallel, individually written sections into a single coherent text — aligned with Lowry et al.'s taxonomy of parallel writing with horizontal division of labor followed by integration.

AI-first vs. human-first drafts. Whether the first draft came from Gen-AI or from students shaped the group workflow: AI-generated drafts served as writing samples and topic familiarization, while human-first drafts were fed to Gen-AI for refinement and expansion. Students also generated multiple AI drafts and selected the highest-quality one.

Benefits and limitations. Students reported that Gen-AI reduced preparation time, improved product quality (eloquence, grammatical correctness, coherence), accelerated learning, overcame writer's block, and built transferable academic writing skills. However, they emphasized limitations including AI "hallucination" and fabricated or outdated information, the risk of offloading important cognitive work, and the need for effortful interaction (e.g., asking AI to explain its changes) to actually learn — central considerations for the knowledge base's Learner Agency and Writing concepts.

Ethical considerations. Students imposed constraints even within "open" strategies (e.g., no generating or writing false information) and weighed ethical aspects of Gen-AI use in the group context. The authors caution that the findings reflect a small, AI-expert graduate sample and should be interpreted as context-specific rather than a comprehensive taxonomy, but they offer insights into more advanced patterns of Gen-AI use in collaborative writing in Higher Education.

What this means for practice

• Instructors. Have each group write down and share its Gen-AI strategy before drafting, because interview accounts sometimes contradicted students' own written reflections—one student denied having a group strategy while describing one in writing—so strategies are not equally visible to all members.
• Instructors. Give Gen-AI an explicit integration role—merging parallel, individually written sections into one coherent text—rather than leaving coordination to emerge, and attach a verification step for fabricated or outdated content.
• Instructors. Require effortful interrogation of AI edits, such as asking the tool to explain what it changed, so that Gen-AI's use for grammar, translation, and first drafts does not substitute for the cognitive work that builds students' Learner Agency over the text.
• Designers. Build shared documents into the assignment workflow: the "open," individually driven strategies in this study were coordinated through Google Docs, which made each member's AI use partly visible to the group.
• Researchers. Triangulate interviews with artifacts—reflections and the group writing documents—when studying Gen-AI use in groups, since the two sources diverged on whether a strategy existed at all.

Limitations

• Ten postgraduate students (aged 23–39, M = 27.7, SD = 5.4) majoring in AI research volunteered out of 143 students invited from one scientific writing course at a single large Korean university, so the authors caution the findings may not generalize to broader student populations.
• The sample is AI-expert by design, which the authors say gives insight into advanced patterns of Gen-AI use that may not yet be observable among less experienced users.
• Data are 30-minute interview accounts plus written reflections, which conflicted on at least one occasion, so reported strategies are self-report and the motivation behind AI use is, by the authors' own observation, more complex than students articulate.
• Three themes from exploratory thematic coding of one assignment are context-specific rather than a comprehensive taxonomy of collaborative Gen-AI use.

Citation

Korchak, A., Costley, J., & Fanguy, M. (2026). AI writes, we collaborate—or vice versa? Group strategies for using generative AI in collaborative writing assignments. Computers and Education Open, 100390. https://doi.org/10.1016/j.caeo.2026.100390