ChatGPT as a Macro-Level Cognitive Scaffold for First-Language Argumentative Writing: A Quasi-Experimental Study in a Madrasah Context

Abstract

Empirical evidence on the use of generative AI for first-language (L1) writing instruction in Indonesian secondary schools remains limited, and few studies have examined which specific writing competencies benefit most from such interventions. This study addresses this gap by positioning ChatGPT not merely as a text-generating tool, but as a dialogic scaffold that supports learners within their zone of proximal development. A quasi-experimental nonequivalent pretest-posttest control group design was conducted with sixty eleventh-grade students at Madrasah Aliyah Negeri (MAN) 2 Parigi, Central Sulawesi, Indonesia, taught by the same teacher across eight instructional meetings. Argumentative writing performance was assessed using an analytical rubric adapted and validated for L1 Indonesian contexts. The findings showed that ChatGPT-assisted instruction produced a large overall effect on argumentative writing achievement (Cohen’s d = 1.30). However, after Holm-Bonferroni correction for multiple comparisons, significant effects remained only in the macro-level dimensions of organisation and content, whereas vocabulary, language use, and mechanics did not retain statistical significance. The study contributes a dimensional-specificity perspective, suggesting that the pedagogical benefits of generative AI in writing instruction are not uniformly distributed across writing components but are concentrated at the level of argumentative macrostructure. These findings extend Vygotskian scaffolding theory into human-AI interaction and deepen current understanding of how generative AI supports L1 writing development in under-researched madrasah and geographically disadvantaged educational contexts.