Prompting Techniques and AI Feedback: A Study of University Students’ Perceptions and Efficacy in Academic Writing

Abstract

Purpose: The purpose of this study is to explore how students can best leverage AI for academic support. Further, this study examines how AI-generated feedback on student writing changes when applying different prompting techniques and how students perceive this feedback across different prompting techniques. Additionally, it seeks to identify which approach students perceive to be the most effective for receiving useful, high-quality feedback. Theory: This study is grounded in two theoretical perspectives that help explain how students interact with and evaluate AI-generated feedback in academic writing: Sociocultural Theory, particularly Scaffolding and the Zone of Proximal Development, and the Technology Acceptance Model (TAM). These frameworks complement each other, as one offers a learning-oriented lens to understand how AI can support students’ writing development, and the other helps to understand perceptions of usefulness and usability of prompting techniques. Together, they offer a lens for interpreting how different prompting techniques shape students’ experiences with generative AI tools. Method: This study employed a mixed-methods, within-subjects interventional design grounded in an interpretivist perspective. Eleven students participated in a workshop where they tested three prompting techniques: zero-shot, roleplaying, and chain-of-thought, to receive AI-generated feedback on a prewritten academic text. Data were collected through a questionnaire and post-workshop semi structured group interviews. The questionnaire provided quantitative demographic and contextual information about participants’ AI tool experience, while the interviews explored students’ perceptions of feedback quality and the usefulness of prompting techniques, providing qualitative data.. Results: The results of the study show that students’ perceptions of AI-generated feedback varied depending on the prompting technique used. While zero-shot prompting was seen as simple and intuitive, it often led to vague and general feedback. In contrast, roleplaying and chain-of-thought (CoT) techniques were perceived as more effective in generating useful, specific, and structured feedback. However, no single prompting technique stood out as universally superior. Although roleplaying was generally viewed as easier to use and useful, students’ experiences highlighted a more practical insight: effective feedback is less about selecting the “right” technique and more about knowing how to communicate clearly with the AI and staying critically engaged throughout the process. Participants who received the most useful feedback were those who provided context, iterated on their prompts, and followed up when initial responses were insufficient.