Artificial Intelligence in ESL/EFL Education: Evidence from Recent Reviews (2024–2025)

The advent of generative artificial intelligence (AI), exemplified by ChatGPT, has triggered a paradigm shift in English language education. This study adopts a “review of reviews” design, synthesizing 14 systematic reviews, meta-analyses, and meta-syntheses (2024–2025). Articles were identified through Scopus and Web of Science using a five-stage selection procedure and analyzed through thematic synthesis. A central contribution is the articulation of a pedagogy-first, ethically grounded research agenda, which distinguishes this study from prior work. The analysis highlights three dimensions: (1) the overall effectiveness of AI-enhanced instruction, (2) applications in writing and speaking, and (3) evolving learner roles in engagement, self-regulation, and emotional experiences. The findings confirm AI’s strong potential to enhance productive skills: automated feedback systems improve accuracy, cohesion, and revision processes in writing, while dialogue-based chatbots strengthen speaking fluency, confidence, and willingness to communicate. These tools serve not only as technological aids but also as cognitive scaffolds and interactive partners that reshape how learners engage with language. At the same time, the benefits are conditional, shaped by factors such as intervention duration, interface design, learner characteristics, and educational context. Risks¾including overreliance, plagiarism, privacy concerns, and uneven attention to the K-12 contexts¾temper these gains. Current research remains concentrated in higher education, limiting cross-level generalizability. Unlike earlier reviews that cataloged applications or emphasized single-skill outcomes, this synthesis integrates effectiveness data with pedagogical, affective, and ethical perspectives. It underscores the need for future research that is longitudinal, context-sensitive, and multi-skilled, while ensuring innovation aligns with pedagogy and robust ethical safeguards.