Governing the Unseen: A Systematic Review of AI Literacy among Language Teachers in Higher Education

Synthesis: Deng, Çelik and Duran (2026) systematically review 32 empirical and conceptual studies (Dec 2022 – Mar 2026) on AI Literacy among language teachers in Higher Education, framed by the "governing the unseen" lens that prefigures institutional policy. Guided by PRISMA 2020 and a critical-policy/sociomaterial thematic synthesis of ERIC, British Educational Index and Web of Science records, they find AI Literacy is overwhelmingly conceptualized through competency-based, multi-dimensional models whose critical and domain-specific dimensions remain underdeveloped. Professional Development is largely unstructured, Assessment leans on self-report tools, and weak institutional support plus unclear responsibility blunt gains such as confidence and innovative teaching. The review concludes that durable, equitable AI literacy demands structural AI Governance rather than individual upskilling.

Key Findings

Conceptualization. AI literacy among language teachers is most often modeled as a set of competencies spanning technological, pedagogical and (less so) critical-ethical dimensions. Domain-specific literacy — how AI reshapes language teaching's subject matter — is comparatively thin, echoing the knowledge base's broader finding that AI Literacy frameworks routinely under-specify disciplinary content.

Four governance patterns. Across the corpus, the authors identify four recurring "governing the unseen" patterns: (1) institutional invisibility of AI literacy in workload models, promotion policies, and resource allocation; (2) fragmented professional development rather than coherent, sustained PD; (3) ungoverned ethical ambiguity — unclear guidelines leave teachers to navigate moral and professional dilemmas individually; and (4) displaced accountability, where responsibility for AI literacy is shared or unclear across institutions, teacher-training programs and policy agencies.

Professional development. Training is frequently ad-hoc, unstructured and unevenly planned, mirroring the "individual effort and experimentation" pattern documented across Educational Development. Despite growing tool availability, the review finds little evidence of coherent, sustained PD models.

Assessment. Evaluation of teacher AI literacy relies heavily on self-report instruments, with limited attention to ethical and critical skills — consistent with Assessment Validity concerns about measuring literacy via perception rather than performance.

Structural barriers. Limited institutional support, unequal access to resources, unclear ethical guidelines, and shared/unclear responsibility reduce positive individual outcomes (confidence, innovation). The authors frame these as governance failures rather than individual deficits: unless there is a conscious effort to govern AI literacy, it will remain unevenly distributed, perpetuate existing inequalities, and place teachers in moral and professional dilemmas.

Conclusion. Sustainable AI literacy development requires an integrated, system-wide strategy — embedding AI literacy in the curriculum, formalizing it in workload and appraisal policies, and offering long-term, subject-specific professional education and ethical support — over occasional workshops and individual upskilling. The paper calls for longitudinal, multi-method, non-Anglophone, performance-based and governance-oriented research on language-teacher AI literacy.

What this means for practice

• Faculty developers. Replace one-off AI workshops with sustained, scaffolded professional development tied to language teachers' subject matter, since the corpus shows PD is routinely ad hoc and short-lived rather than coherent.
• Administrators. Formalize AI Literacy in workload models, promotion criteria and appraisal so that developing it counts as legitimate work rather than invisible individual effort.
• Instructors. Pair tool training with explicit ethical guidance; ungoverned ambiguity currently leaves teachers to resolve moral and professional dilemmas alone.
• Researchers. Move beyond self-report surveys toward performance-based, domain-specific measures of AI Literacy that capture critical and ethical dimensions.

Limitations

• The synthesis rests on 32 studies (Dec 2022 – Mar 2026), 24 of which rely on self-report measures and few of which use validated domain-specific instruments, so assessed literacy reflects perception more than performance.
• The search covered three databases (ERIC, British Educational Index, Web of Science) and was restricted to English-language publications, excluding non-Anglophone research and introducing language and publication bias; no Gray literature or conference abstracts were searched.
• The qualitative-dominant corpus allowed no meta-analysis or formal GRADE assessment; confidence was rated theme by theme using GRADE-CERQual, leaving certainty lower than a quantitative synthesis would give.
• Methodological quality was judged with a 0–8 composite checklist, and the authors flag self-report bias in survey studies, researcher–participant familiarity in case studies and confirmation bias in conceptual reviews as risks that were noted rather than eliminated.

Citation

Deng, Y., Çelik, F., & Duran, V. (2026). Governing the Unseen: A Systematic Review of AI Literacy among Language Teachers in Higher Education. Computers and Education: Artificial Intelligence, 100658. https://doi.org/10.1016/j.caeai.2026.100658