University Teachers Feel Least Prepared for the Digital Skills AI Teaching Demands Most

University teachers in China rate themselves strongest at the routine digital tasks of teaching and weakest at exactly the evaluative, design-oriented and ethical competences that artificial intelligence in higher education increasingly demands, according to a new descriptive study published in Discover Artificial Intelligence. The paper, led by Jie He, proposes a six-dimension conceptual framework called University Teacher Digital Literacy–Six Dimensions, or UTDL-6D, and reports how 520 teachers across 12 Chinese higher education institutions rated their own competence against it. The author is explicit that the framework is preliminary and descriptive rather than a validated diagnostic instrument, and that the survey measured general digital literacy, not AI literacy itself.

The study used a sequential multi-phase design. It began with a structured integrative review of 68 peer-reviewed articles published between 2018 and 2025, identified through searches of Web of Science, Scopus, ERIC, EBSCOhost, CNKI and Google Scholar. It then moved to an adapted Delphi consultation with 12 specialists drawn from six Chinese universities, including educational technology researchers, senior faculty developers and institutional digital transformation leaders. Finally, a cross-sectional survey of 520 university teachers described their self-reported competence and the barriers they perceive. The review was deliberately not called a systematic review: screening and quality appraisal were carried out by a single reviewer, and database-specific search records were not retained, so the search can be approximated but not exactly reproduced.

UTDL-6D comprises six dimensions: technological operational competence, information and data literacy, digital content creation, digital communication and collaboration, innovation consciousness and pedagogical integration, and digital ethics and critical engagement. The framework was built by mapping established models such as the European DigCompEdu framework, TPACK, the UNESCO ICT Competency Framework for Teachers and the DCS-UT scale, combined with themes emerging from the reviewed literature. Eight candidate dimensions were initially produced and narrowed to six after expert consultation, with two pairs merged: innovation consciousness with pedagogical integration, and digital ethics with critical engagement. Notably, those two merged dimensions drew the weakest clarity agreement in the first Delphi round, at 67 percent and 58 percent respectively, and the author acknowledges that separating the disposition to innovate from the capability to integrate technology pedagogically remains a defensible alternative for future testing.

The survey results reveal a distinctly uneven competence profile. Teachers rated themselves highest on technological operational competence, with a mean of 3.72 on a five-point scale (standard deviation 0.68), and on digital communication and collaboration, with a mean of 3.58 (SD 0.72). Information and data literacy followed at 3.41 (SD 0.74). Three dimensions fell below the scale midpoint of 3.0: digital content creation at 2.89 (SD 0.91), digital ethics and critical engagement at 2.78 (SD 0.85), and innovation consciousness and pedagogical integration at 2.65 (SD 0.88), the lowest-rated dimension of all. In other words, respondents expressed the greatest confidence in routine operational and communicative practice and the least confidence in the dimensions that call for design judgement, pedagogical redesign and ethical evaluation.

This pattern carries particular weight for AI-enhanced teaching. The competences that AI-mediated education draws on most heavily, such as judging whether a generative output is accurate and appropriate, recognising bias in an algorithmic recommendation, protecting student data in a learning-analytics environment, and redesigning assessment when generative tools are freely available, sit predominantly within the two dimensions the sample rated lowest. The author frames this as a hypothesis the study generates rather than a finding it establishes: if self-reported competence tracks capability even loosely, the areas of weakest perceived general competence coincide with the areas on which responsible AI integration most depends. The survey contained no items naming an AI system, asking about generative-AI use, or requiring judgement about algorithmic output, so no score in the study can be read as a measure of AI capability.

Alongside the competence profile, the study documented the barriers teachers perceive. The most frequently endorsed were lack of protected training time, reported by 78.3 percent of respondents, rapid technological change at 72.5 percent, insufficient institutional support at 68.1 percent, and misalignment of incentives and rewards at 64.2 percent. These were followed by lack of discipline-specific training at 61.5 percent and inadequate infrastructure at 57.3 percent. The four most frequently endorsed barriers are all structural or institutional rather than dispositional, a pattern consistent with prior work locating teacher digital competence in the interaction between individual capability and organisational conditions. The author cautions that these are perceptions of prevalence: respondents indicated which barriers they recognised, not which most constrained them, so the ordering carries no information about causal priority.

The six dimension scores were moderately and uniformly intercorrelated, with all 15 pairwise correlations positive and statistically significant, ranging from r = 0.42 to r = 0.62 (N = 520, p < 0.001). The strongest association was between technological operational competence and information and data literacy at r = 0.62, and the weakest was between technological operational competence and digital ethics and critical engagement at r = 0.42. The author is careful to note what these correlations cannot establish: correlations between summed scale scores cannot demonstrate discriminant validity, which is a property of the relationship between latent factors and their indicators assessed at the item level. The moderate observed correlations are equally compatible with a smaller number of broader factors.

The psychometric status of the instrument is stated with unusual candour. A confirmatory factor analysis specifying six correlated factors returned acceptable global fit indices, including χ²/df = 2.14, CFI = 0.94, TLI = 0.93, RMSEA = 0.048 and SRMR = 0.042, and Cronbach’s alpha ranged from 0.83 to 0.91 across the six dimensions. However, the item-level output from which standardised loadings, composite reliability and average variance extracted would be computed was not retained, and no alternative models, such as a one-factor or higher-order specification, were estimated. The author notes that a six-factor solution that has not been compared against a one-factor solution has not been shown to be six-dimensional at all. The participant-level response file, analysis syntax and AMOS model file were also not preserved, so none of the analyses can be re-estimated.

Comparisons by academic rank are reported as exploratory for two reasons. First, the 520 participants were nested within 12 institutions, and because per-institution identifiers were not retained, clustering could not be modelled. The author illustrates the potential distortion: under an intraclass correlation of 0.05, the effective sample size would shrink from 520 to roughly 167, and the documented F statistic of 8.42 for the innovation and pedagogical integration dimension would fall to about 2.7, near the conventional significance threshold. Second, academic rank is not an explanatory variable, being bound up with age, experience, resources and opportunity. The one well-documented comparison showed full professors (M = 2.87) rating themselves above lecturers (M = 2.38) on innovation consciousness and pedagogical integration, with Cohen’s d = 0.56, a moderate difference, but the author withdrew an earlier claim that rank differences held across all dimensions because only one comparison is documented in sufficient detail.

The study closes with a proposed Five-Pillar Enhancement Framework addressing infrastructure, professional development, collaborative ecosystems, incentive mechanisms and innovation culture. Every entry in this roadmap is explicitly a hypothesis: no strategy was implemented and no outcome was measured, and the same expert panel that helped generate the framework also appraised it, so its ratings cannot count as independent validation. The author also acknowledges that two endorsed barriers, conceptual ambiguity about digital literacy and privacy and ethical concerns, lack corresponding strategies in the roadmap. Future priorities include a properly powered validation study in a new sample with ordinal-appropriate estimation and item-level reporting, testing the six-factor structure against simpler and seven-factor alternatives, developing AI-specific items from the framework’s theoretical mapping to AI-literacy components, comparing self-reported with demonstrated competence, and evaluating the roadmap through longitudinal implementation studies. The study’s central descriptive message stands: the competences most needed for responsible AI-enhanced teaching are those university teachers feel least prepared for, and the obstacles they identify are ones institutions, rather than individuals, are positioned to remove.

Subject of Research: University teacher digital literacy in AI-enhanced higher education

Article Title: A preliminary conceptual framework and descriptive survey of university teacher digital literacy in AI-enhanced higher education

Article References: He, J. (2026). A preliminary conceptual framework and descriptive survey of university teacher digital literacy in AI-enhanced higher education. Discover Artificial Intelligence, 6(1), Article 1661. https://doi.org/10.1007/s44163-026-02609-x

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02609-x

Keywords: digital literacy, higher education, artificial intelligence, university teachers, UTDL-6D, Delphi method, cross-sectional survey, professional development, digital ethics, China, generative AI, psychometrics