<![CDATA[Generative artificial intelligence (GenAI) is increasingly used to support writing and revision, yet evidence regarding its effects on Arabic writing performance remains limited. This study examined differences in Arabic writing performance and linguistic error patterns between human-only and AI-assisted conditions. We used a randomized, counterbalanced, within-subject design with 72 students in an Arabic Language Education Program at UIN Sunan Gunung Djati Bandung, Indonesia. Participants were randomly assigned to two counterbalanced sequences: human-only followed by AI-assisted writing, or AI-assisted followed by human-only writing. Each participant completed two equivalent Arabic writing tasks of approximately 150–200 words under identical constraints. In the AI-assisted condition, participants used ChatGPT as a revision and feedback tool to help them identify, evaluate, and revise their writing rather than generate essays. Performance was assessed across seven dimensions: content, organization and coherence, grammar (nahwu), morphology (sharaf), vocabulary, syntax, and orthography (imlā'). Linguistic errors were classified into six categories: grammar, morphology, lexical, syntax, orthography, and coherence, while error transitions were separately examined as corrected, persistent, and AI-induced errors. Results indicated higher writing performance in the AI-assisted condition. We interpreted dimension-level improvements descriptively unless we tested them separately. Error analysis showed a reduction in linguistic errors under AI-assisted writing, although some errors persisted and a smaller number were AI-induced. These findings suggest that GenAI can support Arabic writing performance through revision and feedback, but do not establish durable gains in writing proficiency. GenAI is therefore best understood as a mediated resource that complements students' linguistic knowledge, agency, and judgment rather than replacing independent writing.]]>
Copyrights © 2026
