Study: AI Feedback Works Only When Students Act on It

A comprehensive investigation featured on arXiv (2608.11625) delves into the role of AI-generated feedback in educational settings. It reveals that although generative AI has the potential to deliver high-quality, timely, and personalized feedback on a large scale, students often struggle to utilize this feedback effectively. Utilizing a quasi-experimental sequential cohort design, the study analyzed three AI feedback methods involving 13,037 students and 51,296 student-created resources. The methods included: Directed Feedback (n = 3,723), which provided AI comments without structured guidance; Self-Directed Feedback (n = 3,951), allowing students to engage in optional AI dialogues; and Enacted Feedback (n = 5,363), prompting students to assess feedback relevance and interact with AI. The research emphasizes that while feedback is crucial for learning, its effectiveness hinges on overcoming two key challenges: delivering quality feedback at scale and aiding students in understanding and applying that feedback effectively. Simply offering AI-generated feedback is not enough; structured support is essential for students to act on it. This study was published on arXiv as a new paper under the identifier 2608.11625.