<![CDATA[Artificial intelligence (AI) has expanded opportunities for supporting Arabic writing through automated grammatical error correction (GEC). However, existing research remains fragmented, with limited synthesis connecting learner error patterns, AI-based correction models, and their pedagogical implications for Inshā’ instruction. This study systematically reviews AI-based Arabic GEC research to identify dominant nahwu–ṣarf error patterns, evaluate the evolution of AI correction models, and examine their educational implications for non-native Arabic learners. Following the PRISMA framework, eleven studies published between 2020 and 2025 were synthesized using qualitative thematic analysis and VOSviewer-assisted keyword mapping. The review identifies recurrent learner difficulties in agreement, definiteness, morphological ambiguity, syntactic dependencies, orthographic variation, and diacritical accuracy. Arabic GEC has progressed from rule-based and hybrid approaches to Transformer- and large language model (LLM)-based systems that provide more context-sensitive corrections. Nevertheless, current models remain constrained by challenges related to i‘rāb, learner-specific error profiles, explainable grammatical feedback, and limited annotated corpora. The review proposes a human-in-the-loop pedagogical framework integrating AI-generated feedback with teacher validation to foster formative assessment, learner autonomy, and reflective revision in Arabic writing instruction]]>
