Generative artificial intelligence in L2 speaking: a systematic review of feedback and learning outcomes (2022–2026)

Abstract Feedback is widely regarded as a core mechanism in L2 speaking development; however, recent advances in generative artificial intelligence (GAI) necessitate a systematic re-examination of GAI-generated feedback. This systematic review synthesises recent empirical research on GAI–generated feedback in L2 speaking contexts, focusing on how these studies are conducted and how feedback is conceptualised, implemented, and linked to learner outcomes. Following the PRISMA guidelines, 23 peer-reviewed empirical studies published between 2022 and 2026 were identified and coded using a multidimensional framework that encompassed basic study information, feedback characteristics, and learner outcomes. Three main patterns have emerged. First, the studies included are characterised by short-term quasi-experimental designs and a strong reliance on self-reported measures, limiting causal inference and long-term generalisability. Second, GAI-generated feedback is predominantly operationalised as supportive and practice-oriented assistance rather than as a theoretically grounded feedback construct, with limited explicit relation to established feedback theories. Third, reported learner outcomes primarily involve improvements in speaking performance, particularly in fluency and pronunciation, alongside positive non-linguistic outcomes (i.e., increased psychological well-being). However, explicit analytical links between feedback characteristics and learner outcomes are often absent. This review highlights both the pedagogical potential of GAI-generated feedback for L2 speaking development and the need for more theoretically informed and methodologically robust research that systematically examines how feedback design features shape learning outcomes.