A Systematic Review of AI-Based and Teacher-Based Writing Assessment

Abstract

Establishing valid and reliable writing assessments in education remains a persistent challenge, especially with the emergence of artificial intelligence (AI) as a tool in evaluation. Although previous studies have investigated both AI-based and teacher-based writing assessments, few have addressed them through a systematic lens, leading to fragmented insights and inconsistent frameworks. This study aims to investigate the current state-of-the-art of AI-based and teacher-based writing assessments and to identify emerging debates and research directions in the field. A total of 258 articles were collected from Scopus, ScienceDirect, JSTOR, Emerald, and Ebscohost, and filtered using PRISMA guidelines. Eight articles met the inclusion criteria. The findings reveal that AI-based assessments offer high consistency and efficiency in evaluating surface-level language features, but struggle with assessing higher-order discourse aspects such as coherence, argumentation, and rhetorical structure. Conversely, teacher-based assessments provide richer, context-aware feedback, yet are limited by issues of subjectivity and scalability. A hybrid model that integrates AI efficiency with human insight emerges as a promising solution to balance reliability and validity in writing assessment. Nevertheless, key debates remain regarding AI's scoring authority, construct validity, ethical concerns and, implementation across diverse educational contexts. This study calls for the development of unified frameworks and teacher training  to support equitable and effective AI-human collaboration in writing assessment.