Enhanced learning through smart reflection

Reflective writing is more than a diary exercise; it is a structured task that helps learners transform experience into knowledge. In the Process Management lecture for the Bachelor in Information Systems from Thiemo Wambsganss, students were asked to complete a short learning journal entry at the end of each week. The task was simple: “Think about the previous lecture unit: What content or experiences particularly appealed to you or challenged you?” Forty students wrote their reflection for six consecutive weeks – and, for the first time at BFH, they did so with digital scaffolding from Rflect.ch’s AI-enhanced coaching engine. 

Introduction

Reflective writing is a powerful tool for students. It helps them understand their learning experiences better and improves their ability to reflect critically. But reflecting in general is not easy. Many find it challenging to reflect deeply without structured guidance. This is exactly where our recent collaboration between the Human-Centered AI Learning Systems (HAIS) Lab [1], of Bern University of Applied Sciences (with Léane Wettstein, Thiemo Wambsganss and Roman Rietsche) and the startup Rflect [2] comes in. Following a cantonal-funded BeLearn project last year [3], in our recent Innosuisse-funded project, we sought to explore the impact of AI-enhanced coaching on students’ reflective writing process on their feeling of self-efficacy and perceived reflective skill learning. [4]  

Why Reflective Support Matters

Writing about what we learn turns experience into lasting knowledge. Decades of research show that structured reflection strengthens metacognition and professional growth. Reflection is strongly linked to metacognition and self-regulated learning. Yet students often struggle to move beyond surface descriptions toward deeper analysis. Prior work shows that personalized feedback helps (for review, see Chan and Lee 2022), but lecturers rarely have the capacity to coach every learner individually. AI offers a scalable solution but only if its prompts are pedagogically sound and well embedded into the reflective writing process of students (e.g., Neshaei et al. 2025). 

What We Wanted to Find Out – and How

We wanted to explore how AI-enhanced digital tools could support students in structured reflective writing. Specifically, we used the reflective writing app from Rflect in our “Process Management” lecture in the Bachelor Digital Business and AI, which involved around 40 students over a period of 6 weeks. Students regularly wrote reflective texts about their learning experiences using the Gibbs Reflective Cycle—a well-known method for structured reflection (Gibbs 1988). The Gibbs cycle helps students to systematically explore their experiences by guiding them through six stages: description, feelings, evaluation, analysis, conclusion, and action plan. 

To find out more about the impact of AI-enhanced reflective writing on students’ learning process, we design a pre-/post-test field experimental study. Before the first reflection, we asked students to conduct a pre-survey. We measured two main constructs (1) self-efficacy (Bandura, 1997) and (2) perceived reflective writing competences according to the Gibbs Cycle (Gibbs 1988). Each week thereafter, students wrote a reflection in the Reflect app,  about their learning experience of the past lecture. After writing their reflective text, learners had the option to click the so-called “Go Deeper” button to receive feedback. The feedback was automatically generated through a chain of agentic prompts, carefully and learner-centered designed over several iterations. The automatic feedback asked open questions rather than giving directive responses – for example, “What alternative explanation could there be for the outcome you observed?” – and encouraged students to revise their text. After six weeks (and six reflective exercises), we measured the two main constructs (self-efficacy and perceived reflective writing competencies) again. We also asked for qualitative feedback from the students to learn more about the impact of AI-enhanced feedback on students’ reflective writing process and their learning journey.  

How did the AI Coaching work?

Instead of providing “corrective” feedback, the system used a large language model (GPT 4o mini) to emulate a human coach. A first agent identified which Gibbs steps were present in the student text, then a second agent formulated an open question aimed to scaffold their current reflective writing journey. To iterate on the prompts we followed a human-centered prompting methodology, which involved user interviews, expert workshops, data collection and iterative modelling.  

What did we find?

After the six weeks, we found that students rated their self-efficacy significantly higher than before the intervention. The self-efficacy increased by 26.15% (p <.001). This means that students felt more confident that they could learn effectively through reflection. Also, we found that students perceived their reflective competence to be higher than before the AI-enhanced reflective writing task 15.11% (p = .017). Also, the qualitative comments provided many interesting insights. Some participants wished for more formatting options to structure their texts more freely (“That it has some more functions like in a OneNote for freehand text”). Further, while many valued the tool (“I think it’s very good and complete”; “It was actually all pretty clear.”) some wished for more transparency on the AI-functions implemented (“It is not transparent what was ‘positive’ and how exactly I can improve the score.“; “Explain the functions such as ‘go deeper’ in more detail.”). 

What This Means for Educators

AI coaching does not replace human mentorship, but it can free teaching time and provide every student with immediate, individualized scaffolding. When the feedback is given in the form of coaching and open-ended, learners can retain ownership of their thinking while still benefiting from external scaffolding. For larger bachelor courses, that can be a helpful game changer without bearing the risk of providing wrong corrective feedback (e.g., if an open question is wrongly posed, students might just ignore it). As such, students can maintain their agency over their learning process. 

To sum up, we derive the following three recommendations for educators:  

1. Reflective writing is a powerful exercise fitting almost any context, domain, and subject. AI-enhanced coaching can help to make this more interactive and increase students’ self-efficacy.  2.
2. Integrate it in a class’s deliverables. We decided to give students time in class to reflect. Using the tool weekly at the end of our lectures and reserving classroom time for students to reflect helped them stay motivated and engaged. We combined personal student reflections with collaborative reflections at the beginning of each lecture. This eventually helped not only to start the lecture with the most recent content, but also to deepen the reflection through a collaborative discussion. Another option is for such a continuous reflection journey to replace, for example, a longer end-of-semester reflection paper. If reflection occurs instantly and is brief, learning is enhanced, students are less inclined to cheat, and it reduces the corrective workload for the teacher.
3. The ‘How’ matters: bring it close to how students learn. Students utilize a multitude of apps to structure their learning in their personal lives. Working with an app that, to the students, feels like something they would use alongside Insta or WhatsApp made it much more natural to reflect ‘on their terms’.  

References

[1] www.haislab.com 
[2] www.rflect.ch
[3] https://belearn.swiss/en/projekt/exploring-the-effects-of-conversational-agents-on-learners-reflective-writing/
[4] This study is part of a larger Innosuisse project led by Roman Rietsche and Thiemo Wambsganss called “Reflective Writing for Personal Development: Leveraging AI to go from pilot to scale” funded by the Swiss Innovation Agency.

Bandura, A. (1997). Self-efficacy: The exercise of control. W H Freeman/Times Books/ Henry Holt & Co. 

Cecilia KY Chan and Katherine WT Lee. (2021). Reflection literacy: A multilevel perspective on the challenges of using reflections in higher education through a comprehensive literature review. Educational Research Review 32 , 100376. https://doi.org/10.1016j.edurev.2020.100376 

Graham Gibbs. (1988). Learning by doing: A guide to teaching and learning methods. Further Education Unit Publisher: Oxford Polytechnic. 

Seyed Parsa Neshaei, Thiemo Wambsganss, Hind El Bouchrifi, and Tanja Käser. (2025). MindMate: Exploring the Effect of Conversational Agents on Reflective Writing. In Proceedings of the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems (CHI EA ’25). Association for Computing Machinery, New York, NY, USA, Article 397, 1-9. https://doi.org/10.1145/3706599.3720029