Collaborative knowledge construction with generative AI: Exploring argumentative co-writing processes through n-gram and cluster analysis

Abstract Since the beginning of computer-supported collaborative learning (CSCL) research, collaborative writing has been playing a pivotal role as a tool for learning and knowledge construction. In the study presented here, we ask to what extent large language models may not only assist individuals in their writing processes but also serve as a collaboration partner. For this purpose, we analyzed the writing process of individuals supported by ChatGPT. We introduce the use of recurring n-grams as a means for textual uptake, that is, the extent and granularity with which human writers adopt and adapt artificial intelligence (AI)-generated text. On the basis of the overlaps between the ChatGPT output and participants’ final texts, we identified clusters of text reproducers, integrators, and reconstructors. Participants in these clusters differed not only in their subjective contributions and authorship but also in their prior use of ChatGPT and their affinity toward technology interaction. Referring to the conceptualization of interindividual interactions as uptake events, we suggest that n-grams are adequate means to analyze the uptake process in AI-supported human writing. Our findings show that AI-supported writing comprises distinct uptake patterns that differ systematically in the degree of textual reuse and perceived authorship, thereby revealing varying modes of engagement in human–AI co-writing, ranging from passive uptake of AI-generated text to more active and integrative forms of collaboration.