<![CDATA[The rapid proliferation of AI-driven language learning tools has transformed the feedback ecology of EFL speaking instruction, yet systematic appraisal of their differential impact across skill levels remains limited. This systematic literature review (SLR) investigates how AI-generated feedback supports and constrains the development of EFL speaking competence, with particular attention to the micro–macro skill distinction and learner needs. Guided by the PRISMA 2020 framework (Page et al., 2021), eighteen peer-reviewed studies published between 2020 and 2025 were identified, screened, and analysed through thematic synthesis following Thomas and Harden's (2008) iterative coding procedures. The principal finding is a consistent and theoretically significant asymmetry: AI-driven systems—particularly those grounded in Automatic Speech Recognition (ASR) and Natural Language Processing (NLP)—demonstrably improve micro-level skills, including pronunciation accuracy, grammatical precision, and lexical control, because these features are discrete, measurable, and amenable to algorithmic evaluation. Macro-skills, by contrast—interactional fluency, discourse organisation, and communicative competence as theorised by Canale and Swain (1980), Hymes (1972), and Brown (2004)—remain substantially underserved by current tools. This asymmetry is not merely technical but conceptual: communicative competence is inherently relational and sociolinguistic, and no reviewed system engaged the conceptualiser stage of Levelt's (1989) speech production model, where pragmatic and sociolinguistic knowledge resides. Learner needs at the macro level—authentic interaction, contextual adaptability, and meaning negotiation as foregrounded by Long's (1996) Interaction Hypothesis—are systematically unaddressed. Grounded in Thornbury's (2005) interactional view of speaking, these findings support a blended pedagogical model in which AI serves as a rehearsal and diagnostic scaffold for micro-skills while human-facilitated interaction remains central to communicative development. The review contributes a dual-lens analytical framework integrating the micro–macro distinction with learner-needs analysis, offering theoretical and methodological direction for future research in AI-assisted EFL speaking.]]>
