AI detection tools

The emergence of generative artificial intelligence (Generative AI), such as ChatGPT, has brought significant changes to higher education, including Arabic language learning for non-native speakers (Arabic as a Foreign Language). Writing skills (Maharah al-Kitabah) are considered a relatively complex productive skill. Learners are required not only to master word formation (Sarf) and Arabic grammar (Nahwu), but also to organize ideas coherently and logically. This article employs a Systematic Literature Review (SLR) approach based on the PRISMA 2020 standards to map the use of ChatGPT in Arabic writing, examine the reliability of AI in Arabic grammar, and identify research gaps that remain underexplored. The review findings indicate that ChatGPT can serve as a useful writing partner (human-AI co-writer), particularly for generating ideas, enriching vocabulary, and correcting basic grammatical errors. However, AI still frequently produces errors or “hallucinations” when dealing with morphologically complex words, variations in Arabic dialects, and unvocalized texts. As a result, AI-generated writing may appear linguistically polished on the surface while containing grammatical inaccuracies. Furthermore, excessive reliance on AI may pose risks to academic integrity, while AI detection tools themselves often produce inaccurate results when identifying Arabic-language texts (false positives). This review also finds that existing research remains overly focused on English, tends to examine only the final writing outcomes without investigating students’ cognitive processes, and has paid limited attention to AI models specifically developed for the Arab region, such as ALLaM and Jais. Therefore, the approach to AI-assisted learning needs to shift from merely using AI toward purposeful and reflective utilization through frameworks such as the Spiral Learning Model and Genre-Based Pedagogy, so that AI functions as a scaffold without diminishing students’ autonomy and critical thinking.

Higher education has historically operated under a paradigm of information scarcity, positioning universities as premier institutions for curating and transmitting knowledge. The convergence of digital hyper-abundance, algorithmic curation, and generative artificial intelligence (GenAI) has disrupted this model, producing what educators increasingly describe as a crisis of critical thinking in higher education. The purpose of this review is to examine how information abundance, algorithmic curation, and GenAI use may influence university students’ critical thinking, and to synthesise what this evidence implies for what universities teach and how they assess. Using a critical narrative review design with thematic synthesis, the review analyses 45 verified sources (empirical studies, systematic reviews and meta-analyses, conceptual works, and policy guidance) identified through citation chaining and targeted searches; empirical GenAI studies date from November 2022 to 2026, and foundational theoretical works from 1976 onward. The synthesis is anchored in an integrative framework that combines cognitive load and dual-process theory, cognitive offloading, scaffolding theory, self-regulated learning, and epistemic cognition, with critical thinking as the principal outcome. Five themes emerged: (1) information abundance appears to shift the central cognitive challenge from retrieval to evaluation; (2) unguided GenAI use is associated with reduced cognitive engagement and, in several experiments, with weaker independent performance once AI support is removed, although guardrailed designs appear to mitigate this; (3) fluent AI output can be mistaken for validated knowledge, and AI-detection tools are unreliable and biased; (4) lateral reading, guided AI use, and process-visible assessment have promising but uneven support; and (5) workload, equity, and policy constraints limit adoption. From these themes, the review derives the Fade–Visibility Matrix, which crosses AI support structure (persistent versus fading) with assessment design (product-only versus process-visible) to yield four diagnostic states. The evidence is early and methodologically uneven, but it is most consistent with a structural rather than individual explanation. Universities should therefore teach evaluation, verification, and metacognitive regulation explicitly, design AI support to fade, and redesign assessment so that the learning process, not only the finished product, can be observed.