The Educational Effects of Generative AI Use in International Undergraduate Students' Self-Introduction Writing : Pre-Post Text Changes and Error Patterns
TL;DR
This study presented a practical instructional model that positions generative AI not as a ghostwriting tool, but as a scaffold within the writing process, suggesting that optimal outcomes emerge from the collaboration between generative AI and human instructors.
Abstract
This study implemented a three-stage writing instruction model—No AI, With AI, and Without AI—for international undergraduate students’ employment-oriented Korean self-introduction writing. Repeated- measures analysis across the three stages showed that increases in text length and lexical diversity occurred mainly in stage 2, whereas syntactic complexity declined across both stages 2 and 3. However, stage 2 texts also featured factual inaccuracies and an increased use of Sino-Korean vocabulary, which were subsequently addressed through individualized instructor feedback in stage 3. Thus, this study presented a practical instructional model that positions generative AI not as a ghostwriting tool, but as a scaffold within the writing process. The analysis of authentic learner texts empirically demonstrated its effectiveness, suggesting that optimal outcomes emerge from the collaboration between generative AI and human instructors.