Exploring the Dunning-Kruger Effect in AI-Assisted EFL Writing among University Students

Abstract

The growing dependence of AI writing tools — including ChatGPT, Grammarly, Gemini, QuillBot, and DeepSeek among university EFL learners increases an essential yet often overlooked question: are students actually getting better at writing, or are they merely getting better at using AI tools? This study explored that question by examining the role of the Dunning-Kruger Effect in the writing experiences of ten EFL university students at Universitas Muhammadiyah Palu who regularly use AI tools in their academic work. Data were collected through written asynchronous interviews and analyzed using Interpretive Phenomenological Analysis (IPA). Four themes emerged from the data: students showing confidence that outpaced their actual independent ability; students struggling to acknowledge their own contributions from AI-generated ones; a visible gap between how students assessed themselves and what they could actually do on their own; and perhaps most strikingly, students who recognized their over-reliance on AI but made no real effort to change it. Collectively, these findings suggest that AI tools affect more than the way students write. They gradually change students' views of their own writing skills, often in ways that do not match their actual writing ability. The study offers practical insights for EFL teachers, curriculum developers, and university administrators thinking through how AI should and should not be integrated into academic writing instruction.