Unpacking student interactions in GenAI ‐assisted social annotation: An epistemic network analysis of cognitive and social presence

Abstract Cognitive presence and social presence are essential for computer‐supported collaborative learning (CSCL). Yet, monitoring and providing personalised feedback to support these presences remain a persistent challenge, particularly in the context of social annotation (SA). Generative Artificial Intelligence (GenAI) has the potential to address this challenge by enabling timely and personalised support. However, how GenAI shapes cognitive and social presence during SA processes remains underexplored. This study bridges this gap by investigating how GenAI‐assisted social annotation (GASA) alters the dynamics of cognitive and social presence. Using a quasi‐experimental design, 63 Chinese undergraduate students from two classes participated in a 5‐week collaborative reading programme. One class (experimental: N = 33) used an SA platform integrated with a GPT‐4o mini‐powered chatbot that provided scaffolding for both cognitive presence (e.g. through monitoring and personalised suggestions) and social presence (e.g. using vocatives and expressing acknowledgement), while the other class (control: N = 30) used the same platform without chatbot assistance. Data were collected from collaborative reading engagement surveys, reading performance scores and platform logged data (annotation and chatbot interaction logs). Content analysis was employed to analyse log data for indicators of cognitive and social presence based on the community of inquiry framework. ENA was subsequently employed to model the structure of co‐occurring cognitive and social presence across the two conditions to understand the effect of GenAI scaffolding, and between groups with high and low reading performance within the experimental condition to reveal nuanced variations in response to GenAI scaffolding. The results from Mann–Whitney U ‐tests indicated GASA significantly enhanced students' cognitive engagement, emotional engagement and reading performance, as well as altered the dynamics of cognitive and social presence. ENA results further revealed that the experimental class exhibited a strong connection between social presence and cognitive presence indicators of integration and resolution. Additionally, within the experimental class, students with higher reading performance effectively utilised the GenAI for cognitive integration, resolution and social presence, while lower performing students exhibited a weaker integration of GenAI scaffolding into their collaborative cognitive and social presence. This study provides a process‐oriented understanding of students' use of GenAI support for social annotation and offers practical insights for designing GenAI‐enhanced collaborative learning environments. Practitioner notes What is already known about the topic Cognitive and social presence are crucial for effective computer‐supported collaborative learning (CSCL), particularly in the context of social annotation (SA). Learners encounter challenges in maintaining cognitive and social presence within SA environments. Meanwhile, instructors face time‐consuming and labour‐intensive demands when implementing strategies to assist learners in enhancing these presences. Generative artificial intelligence (GenAI) presents a promising solution by offering timely and personalised support during the SA process. What this paper adds This study implemented GenAI‐assisted social annotation (GASA) by designing a collaborative reading platform, CollaboRead, which incorporates a GPT‐powered chatbot to provide scaffolding for both cognitive and social presence. This study examined whether GASA could alter the collaborative dynamics of cognitive and social presence and its association with learning outcomes (collaborative reading engagement and reading performance), and crucially, how differential responses to GenAI scaffolding were associated with reading performance. This study offers a process‐oriented understanding of GenAI‐augmented collaborative learning through epistemic network analysis, suggesting the mechanisms by which GenAI scaffolding may affect the interplay of cognitive and social presence during SA, potentially enhancing cognitive engagement, emotional engagement and reading performance. Implications for practice and/or policy Educators should consider integrating GenAI tools with configuring scaffolded support into collaborative learning activities to foster deeper engagement and improve learning outcomes. Educators should be aware of the importance of mechanisms that promote learner agency in engaging with GenAI scaffolding and that support AI literacy for enabling learners to effectively interact and act on GenAI suggestions in their collaborative knowledge co‐construction, thereby maximising the effectiveness of GenAI in SA and potentially other CSCL environments.