In response to increasing student enrollment in computing courses, teaching assistants (TAs) now play a major role in student development. In particular, they are responsible for providing feedback to students, a critical factor in student engagement and performance. However, TAs often receive little formal training in how to provide effective feedback, resulting in inconsistent or poor-quality feedback that can hinder student learning and decrease motivation.

To address this, we present FeedbackPulse-CLT, a novel approach to automated support and training of tutors to write effective feedback. The tool is available as a Google Chrome extension and uses faded worked examples, a scaffolding technique informed by Cognitive Load Theory (CLT). By providing structured, real-time guidance that adapts to the TA’s expertise levels, FeedbackPulse-CLT offers targeted training and support in feedback writing.

Our mixed-methods randomized trial of 27 CS teaching assistants compared FeedbackPulse-CLT with faded worked examples against standard FeedbackPulse. While faded worked examples showed no significant advantage over traditional methods in reducing cognitive load, the study uncovered critical insights into what motivates TAs to use training tools. It also identified practical barriers they face, including time constraints and competing demands. These findings highlight important gaps between educational theory and real-world application, providing essential guidance for developing more effective LLM-powered feedback training systems.