AI Usage Reflection Companion

The Challenge

As generative AI tools become increasingly embedded in higher education, students are using them for reading, writing, coding, brainstorming, research, and problem solving. While these tools can support learning, they also make it harder for students to see how AI is shaping their thinking. Chat histories are long, messy, and centered on outputs, which means students may remember what they asked for, but not how much cognitive work they delegated, where AI appeared most often, or what patterns developed over time.

This lack of visibility matters because metacognition depends on being able to reflect on one’s own learning process. Prior research raises concerns about cognitive offloading, over-reliance, and “metacognitive laziness,” especially when students use AI mainly for lower-level tasks such as recall, explanation, or first-pass generation. However, our research also showed that AI use is not simply “good” or “bad.” Students use AI to save time, clarify difficult material, and support learning, but they need better ways to understand when AI is helping them think and when it may be quietly taking over parts of the thinking process.

Our Goal

Our team aims to support AI literacy and responsible AI use in higher education by helping students reflect on how they work with generative AI. Rather than policing or discouraging AI use, we aim to design a tool that makes student-AI collaboration more visible, interpretable, and actionable.

The project focuses on helping students understand where AI shows up in their workflow, how work is split between the student and AI, and what kinds of collaboration patterns emerge over time. By surfacing these patterns in a lightweight and non-judgmental way, the tool encourages students to maintain agency, practice metacognition, and make more intentional choices about how they use AI for learning.

Approach

We followed an iterative, research-driven design process that combined literature review, needfinding, framework exploration, prototyping, and usability testing. Our process included:

• A literature review on metacognition, cognitive offloading, AI-mediated learning, AI literacy, and educational theories
• 15 semi-structured interviews with UC Berkeley students to understand how they use AI, why they rely on it, how they verify outputs, and whether they notice patterns in their own use
• Walkthroughs of students’ real AI chat histories when possible, along with reactions to ChatGPT Study Mode and early reflection concepts
• Grounded theory analysis of interview transcripts, with codes related to AI adoption, prompting practices, trust, verification, reliance, learning impact, dashboard reactions, and Study Mode experiences
• Exploration of Bloom’s Taxonomy and the ICAP framework to evaluate whether existing educational frameworks could capture student-AI collaboration
• Iterative prototyping, beginning with a static AI usage dashboard and evolving into an embedded side panel plus longer-term dashboard
• 6 usability tests with UC Berkeley students to evaluate clarity, usefulness, tone, privacy concerns, actionability, and whether students would return to the tool after future chats

Our Solution

The final solution is the slime, an AI reflection companion for students. The slime is designed as an embedded ChatGPT reflection layer that helps students notice and interpret their AI use without feeling judged or monitored. This can be downloaded as a chrome extension. It combines two connected reflection experiences: an in-chat side panel for immediate, session-level reflection and a longer-term dashboard for broader usage patterns.

The side panel provides a quick read of the current chat, showing how the student and AI worked together across areas such as ideas, direction, research, building, problem solving, and final decisions. It also offers concise, actionable suggestions, such as how to clarify goals, deepen the conversation, or improve prompting in the next interaction. The longer-term dashboard helps students see where AI appears most often in their work, how responsibilities tend to be split between them and AI, and what kind of collaborator they may be becoming over time.

Through this design, the slime shifts reflection from a passive analytics dashboard to a more relational, embedded, and actionable companion. It helps students better understand their relationship with AI, preserve ownership over their learning, and use generative AI more intentionally.