THE USE OF ARTIFICIAL INTELLIGENCE IN DEVELOPING SPEAKING AND WRITING SKILLS FOR LANGUAGE TESTS: A SYSTEMATIC REVIEW

Abstract

Integrating generative artificial intelligence (AI) into language education brings about specific changes in teaching methods, particularly with regard to practical skills such as speaking and writing. This systematic review evaluated the empirical impact of AI on language output, following the PRISMA 2020 guidelines. As the focus was on qualitative pedagogical changes, the eligibility framework bypassed traditional clinical comparison models in favour of strict publication parameters. A targeted search of Web of Science, Scopus and ERIC yielded an initial pool of 20 records. After applying exclusion criteria relating to date, language and empirical rigour, six peerreviewed studies published between 2020 and 2026 remained for the final qualitative synthesis. The extracted data suggest that AI tools provide practical support; they reduce psychological barriers during oral practice and alleviate the cognitive burden of mechanical writing errors. However, the literature consistently highlights significant risks. Uncritical reliance on automated assistance can lead to cognitive offloading, and technical flaws can expose learners to factual hallucinations and algorithmic bias. Broad restrictions on these tools are largely impractical. Instead, instructional designs must pivot towards explicit training in AI literacy and adopt human – AI hybrid models. By requiring students to evaluate and correct flawed machine outputs, educators can transform algorithmic weaknesses into valuable analytical exercises.