AI-supported learning gains ground as students co-create with ChatGPT

A key question explored in the study was the extent to which students edited ChatGPT’s outputs. On average, students modified AI responses by 50%, indicating a healthy balance between leveraging AI assistance and applying human judgment. Interestingly, most students (44%) reported modifying responses at a rate of 60%, suggesting active involvement in the refinement process. Only a small minority used ChatGPT’s output without significant changes.

AI-supported collaborative learning is gaining momentum as universities explore how students and emerging technologies like ChatGPT can co-create knowledge in team-based environments, navigate complex tasks, and reshape traditional learning models

A new study published in the journal Computers highlights both the promise and complexity of integrating generative tools into academic teamwork. Titled “Students Collaboratively Prompting ChatGPT”, the study offers one of the most detailed empirical investigations yet into this phenomenon. 

The research sheds light on which collaborative models students prefer, how they adapt AI-generated content, and the ways in which digital tools support or hinder teamwork. With generative AI tools becoming standard in many educational settings, this study provides important guidance for educators, institutions, and policymakers aiming to preserve academic integrity while enhancing collaborative learning outcomes.

What collaborative structures do students prefer when using AI?

The study organized students into 51 teams of three and randomly assigned each team one of five collaborative prompting modes. These ranged from tightly coordinated team efforts (Mode #2) to independent work later merged into a unified product (Mode #5). Despite the structured experimental design, researchers found no overwhelming student preference for any one collaborative mode. Students slightly favored Mode #2 (team-oriented prompting), Mode #4 (team-to-member prompting), and Mode #5 (independent cooperation), primarily due to the clear division of roles and smoother workflows.

Importantly, students reported initial disagreements in about 30% of collaborative decisions, suggesting that AI-assisted group work still requires critical thinking, negotiation, and interpersonal skills. These disagreements were typically resolved through dialogue, reinforcing the importance of peer interaction in knowledge construction. Mode #3 (member-then-team) was the least liked, possibly because it combined both independent and collaborative components, creating ambiguity in task ownership.

Each collaborative style came with its own strengths and limitations. Mode #2 facilitated seamless collaboration but was less ideal for students who preferred working independently. Mode #5 allowed autonomy and encouraged critical thinking but resulted in difficulties merging individually created work. This variability reinforces the need for adaptive instructional design, there is no one-size-fits-all approach when it comes to collaborative AI use in classrooms.

How do students engage with and modify AI-generated content?

A key question explored in the study was the extent to which students edited ChatGPT’s outputs. On average, students modified AI responses by 50%, indicating a healthy balance between leveraging AI assistance and applying human judgment. Interestingly, most students (44%) reported modifying responses at a rate of 60%, suggesting active involvement in the refinement process. Only a small minority used ChatGPT’s output without significant changes.

Students’ final deliverables reflected this engagement. On average, 44% of their submitted work was based on ChatGPT’s responses. This suggests moderate reliance on the tool, with significant variation: some teams leaned heavily on the AI, while others maintained higher levels of original authorship. This variation underscores the importance of training students in responsible AI use - knowing when to trust, question, or discard an AI-generated response is now a crucial academic skill.

These findings align with cognitive apprenticeship and constructionist theories of learning, where students actively transform knowledge rather than passively consume it. By modifying and contextualizing ChatGPT’s responses, students demonstrated engagement with course content and higher-order thinking skills such as analysis, synthesis, and evaluation.

The study also found a statistically significant correlation between students’ perceived learning and their actual project grades. Students who felt they learned a lot, either individually or as a group, tended to perform better academically. This alignment supports the value of structured collaborative learning and reinforces self-assessment as a viable metric for evaluating educational outcomes.

What role do communication tools play in AI-supported collaboration?

The research further investigated how students communicated while collaborating on AI-assisted assignments. A wide variety of tools were used, including phone calls, Instagram, Google Drive, Zoom, email, Messenger, and even FaceTime and Viber. Most students favored a hybrid communication model, combining synchronous interaction for real-time discussion with asynchronous tools for file sharing and document collaboration.

Students appreciated ease of use, quick feedback, and the ability to co-edit documents in real-time. However, they also reported common obstacles such as scheduling conflicts, connectivity issues, and the lack of direct, in-person communication. These barriers often led to misinterpretations and lower engagement levels, especially when team members couldn’t align their work schedules.

Despite these challenges, many students expressed a strong preference for at least some face-to-face interaction, citing its effectiveness for clarifying misunderstandings and fostering group cohesion. This finding aligns with media richness theory and the concept of transactional distance, highlighting the importance of using diverse communication channels to bridge the gaps in remote collaboration.

As for the recommendations, students suggested improving technical support for online collaboration platforms, allowing more time for live meetings, and integrating features like scheduling tools into communication apps. These insights are crucial for institutions seeking to optimize their digital learning environments.