From digital competence to AI-responsive pedagogy: a scoping review of technology-supported teacher education

Synthesis: Patiño Hernández, Chiappe and Ortega González report a PRISMA-ScR scoping review of technology-supported teacher education drawn from two analytically distinct Scopus samples: 155 studies on digitally supported teacher education (RQ1 and RQ2) and 40 AI-focused studies (RQ3), 195 publications in total, covering 2006 through 2025. Coding combined primary-category classification, descriptive frequency analysis and, for the AI subset, multi-label coding. Competence and ICT-integration frameworks were the most frequent strategy in the larger subset (46.5%), digital competence development the most reported advantage (76.1%) and infrastructure or access the most reported barrier (45.2%); in the AI subset, professional development (35.0%) and instructional design and assessment (32.5%) led the approaches, with pedagogical innovation or redesign (35.0%) the leading reported benefit and superficial integration (35.0%) the leading concern. The key qualification is the authors' own: the RQ1/RQ2 search string named competence-oriented terms explicitly but carried no equivalent descriptors for instructional design or assessment, so the frequencies describe the retrieved corpus rather than the field.

Key Findings

• Competence frameworks dominate the retrieved corpus, partly by search design. Frameworks and models for digital competence and ICT integration accounted for 46.5% (n = 72) of the 155-study subset, ahead of digital modalities and courses at 12.9% (n = 20), specific digital resources and environments at 11.6% (n = 18), and instructional design and assessment for learning at 1.9% (n = 3). The authors note that digital competence, TPACK and DigCompEdu were explicit retrieval descriptors, so this is the composition of the retrieved evidence base, not a prevalence estimate.
• Digital competence development is the advantage the literature reports most. It was the primary advantage in 76.1% (n = 118) of the subset, against flexibility, access and continuity of training at 8.4% (n = 13) and pedagogical innovation and redesign at 5.2% (n = 8). Only 3.9% (n = 6) named transfer to practice and professional development.
• Barriers cluster on material conditions and on transfer. Infrastructure, connectivity or limited access was the most frequent primary barrier at 45.2% (n = 70), followed by limited transfer to practice or sustainability at 20.6% (n = 32) and insufficient training or digital competence at 17.4% (n = 27); time, design or management overload accounted for 6.5% (n = 10).
• The dominant advantage was stable across strategy types; the dominant barrier was not. Digital competence development led within frameworks and models in 72 of 72 studies (100.0%) but limited transfer to practice or sustainability was the leading barrier there in 27 of 72 (37.5%). Digital modalities and courses showed infrastructure or access as the leading barrier in 18 of 20 (90.0%), teacher professional development showed time, design and management overload in 7 of 14 (50.0%), and active and collaborative methodologies showed insufficient training or digital competence in 5 of 9 (55.6%). Instructional design and assessment was the only category where pedagogical innovation and redesign was dominant (2 of 3; 66.7%).
• AI enters teacher education through several professional-learning functions rather than one model. In the 40-study AI subset, AI-supported professional development and teacher accompaniment was the most frequent primary approach at 35.0% (n = 14), closely followed by instructional design and assessment for learning at 32.5% (n = 13); diagnosis of AI adoption, perceptions, literacy or competence accounted for 22.5% (n = 9) and active, collaborative and practice-based approaches for 10.0% (n = 4).
• The leading AI concern is pedagogical, not technical. Pedagogical limitations or superficial forms of AI integration was the most frequent primary barrier at 35.0% (n = 14), ahead of insufficient AI-related preparation or professional competence at 27.5% (n = 11) and weak institutional, curricular or policy support at 15.0% (n = 6); privacy, ethical or security concerns and problems of transfer or sustainability each accounted for 10.0% (n = 4).

What the AI studies show about competence becoming practice

The AI subset moves the review's question from what teachers know to what they can evaluate. Fan et al., 2025 reported an 18-hour AIPACK program for in-service and preservice elementary teachers with gains across AI knowledge, AI content knowledge, AI pedagogical knowledge and integrated AIPACK. Huynh et al., 2025 compared tool-focused training with training that also addressed AI ethics, human-centered education and pedagogical reflection: the tools-only condition produced stronger gains in perceived Self-Efficacy, while the mindset-oriented condition produced more nuanced consideration of pedagogical and ethical risks, so technical confidence and reflective competence did not develop in parallel.

AI competence also depended on prior disciplinary knowledge. Kuzu et al., 2025 found that prospective teachers with both subject-matter preparation and prompt-engineering training produced more viable complex tasks with ChatGPT than participants lacking that preparation, and Kang et al., 2025, working with 39 preservice teachers on the BROKE prompting framework, found that disciplinary literacy partially mediated the relationship between prompt engineering and instructional design. Feedback studies point the same way: Pargmann et al., 2025, in a longitudinal study with 103 student teachers, found that an analytical AI platform could match or exceed human feedback in some dimensions of lesson planning, yet exposure to AI feedback did not automatically improve AI literacy, attitudes or motivation.

Across the multi-label coding, critical AI literacy, ethical judgment, human oversight, professional agency, disciplinary knowledge and institutional governance recurred as conditions for defensible use rather than as separate applications. Practice-based designs (Yu et al., 2025; Huang et al., 2025) embedded AI inside pedagogical routines where outputs had to be discussed, tested and justified; their small number (10.0%) suggests these forms remain less developed than professional development, design and diagnostic approaches.

What this means for practice

• Faculty developers. Build AI preparation around evaluation, not operation. Technical self-efficacy and ethical or reflective competence did not move together in Huynh et al., 2025, and prompting was educationally meaningful only when it mobilized disciplinary knowledge and judgment (Kang et al., 2025).
• Instructors and teacher educators. Treat competence frameworks as descriptive architectures, not evidence of transfer. The review reports limited transfer to practice or sustainability as a barrier in 20.6% of the larger subset, and dominant within the framework category itself in 37.5%.
• Curriculum designers. Fund the thin end of the evidence. Instructional design and assessment for learning appeared in only 1.9% of the RQ1/RQ2 corpus, yet the AI subset shows design, feedback and reflection as the functions where AI is most productively embedded.
• Administrators and policymakers. Budget for infrastructure and the institutional conditions AI work depends on. Infrastructure, connectivity or limited access was the leading barrier at 45.2% in the larger subset, and weak institutional, curricular or policy support was a primary concern in the AI subset.
• Assessment designers. Keep humans accountable for interpretation: AI feedback matched or exceeded human feedback in some planning dimensions in Pargmann et al., 2025 but did not by itself raise AI literacy.

Limitations

• The architecture of the RQ1/RQ2 search string is a primary limitation: competence-oriented terms were explicit retrieval descriptors while instructional design and assessment had no equivalent descriptors, so the 46.5% and 1.9% frequencies are properties of the retrieved corpus, not population-level estimates.
• Scopus was the only bibliographic database, so studies indexed elsewhere may have been omitted, and the probabilistically selected screening samples were progressively reduced to the final subsets of 155 and 40, so the 95% confidence level and 5% margin used to size the initial screening volume should not be read as inferential precision estimates.
• Coding assigned one primary strategy, advantage and barrier per study, and the review did not run formal independent double coding across the whole corpus, so no inter-coder reliability coefficient is reported; the cross-category patterns are descriptive co-classifications, not causal associations.
• The AI evidence is a rapidly evolving domain: the search was run in March 2026 but delimited to 2006 through 2025, so the 40-study subset maps that window, and the AI-responsive pedagogy construct is offered as open to refinement as new evidence accumulates.

Citation

Patiño Hernández, J. F., Chiappe, A., & Ortega González, E. (2026). From digital competence to AI-responsive pedagogy: a scoping review of technology-supported teacher education. Frontiers in Education, 11, 1949296.