One of the challenges in online education is ensuring that feedback leads to meaningful learning. While formative feedback is valuable, it often fails its implementation due to the high workload for instructors, which raises concerns about its sustainability. For feedback to be effective, students must internalize it through self-regulation. Self-feedback—comparing one’s own learning with external sources—can support this, but requires explicit instruction. This study examines how different comparison sources foster self-feedback in asynchronous online learning.

We conducted a pilot quasi-experimental study at the Open University of Catalonia (UOC), a fully online institution where formative assessment typically involves a sequence of progressive asynchronous assignments. The study took place in a master’s course on children’s language evaluation for educators. Students completed a case-based task requiring them to design an assessment plan for a multilingual child with potential language difficulties. The study focused on the second of four course assignments, which counted for 35% of the final grade.

A total of 136 students and three teachers participated. First, all students submitted an initial version of the assignment. They then accessed a designated online space to revise and resubmit their work using one of three randomly assigned comparison sources:
(1) a model solution (n = 44),
(2) a rubric (n = 44), or
(3) ChatGPT (n = 45).

Students in the AI condition received example prompts, but were free to use the tool independently. Teachers did blind assessed both versions.
All students followed a guided self-feedback protocol structured around four phases: self-evaluation, comparison, changes implementation, and final reflection. Questions such as “Which elements are missing or underdeveloped?” helped activate reflective thinking. After the reflection process, students resubmitted the task with the new changes. The entire process was time-bound (three hours), and no other sources were allowed, enhancing the robustness of the design.

The results showed students’ scores increased from a mean of 25.21 points (SD = 3.92) in the first submission to 27.45 points (SD = 4.09) in the resubmission, indicating overall improvement. Descriptive statistics reveal a gradual increase in mean improvement depending on the comparison tool. The ChatGPT group achieved an average gain of 1.61 points (SD = 1.34), with a maximum individual improvement of 5.2 points. The rubric group improved by 1.91 points on average (SD = 1.53), reaching a maximum gain of 6.4 points. Finally, the model solution group showed the highest mean improvement, with an average gain of 2.24 points (SD = 1.72) and a maximum observed improvement of 7.4 points. These results suggest that all three types of comparison sources supported performance improvement, with the model solution showing the greatest effectiveness in enhancing students’ initial assignment.

These findings underscore the need to guide self-feedback explicitly and show how each comparison source has distinct strengths and limits. They also reveal the strengths and limitations of each comparison source, offering guidance on their use to trigger self-feedback, enhance formative feedback, and support its sustainability in online higher education. This analysis forms part of a wider study that will add qualitative data and a larger sample.

Keywords: Feedback, online, self-feedback, higher education.