Reconceptualizing digital literacy in the AI era: A systematic review of validated frameworks from an ecological perspective

Synthesis: This systematic review asks whether digital literacy frameworks still fit an AI-saturated world. Screening four databases for validated instruments published between January 2014 and February 2025, the authors assemble 69 digital literacy frameworks and 11 AI literacy frameworks — 80 in all — and read them through an ecological lens at three levels: the school microsystem, the socio-economic macrosystem, and the historical chrono-system. What they find is a field split by role and by region. Student frameworks cluster on information, communication, safety and creation; educator frameworks spread across pedagogy, learner empowerment and assessment. Developed regions weight communication and collaboration, developing regions weight problem solving and operational use, and copyright and licenses appear in no developing-region framework. The chrono-level answer is that AI has expanded digital literacy from safety toward AI ethics, elevated higher-order thinking, and recast communication to include human–AI interaction — while metacognitive regulation remains thinly specified.

Key Findings

1. The review analyzes 80 validated frameworks — 69 digital literacy (67 database-derived plus two reports) and 11 AI literacy — from a PRISMA screen of 35,036 without-AI and 441 with-AI records, covering publications from January 2014 to February 2025.
2. Of the 69 digital literacy frameworks, 28 target students or citizens and 42 target educators (one counted in both roles); student frameworks most often carry information and data literacy (n = 40), communication and collaboration (n = 20), and safety and ethics (n = 19).
3. Educator frameworks distribute differently, with teaching, learning and pedagogy (n = 38), empowering learners (n = 25) and curriculum and assessment (n = 25) leading, and information and data literacy (n = 31) and technology literacy (n = 28) the prominent competencies.
4. Mapped onto DigComp 2.2, framework coverage concentrates on information and data literacy (frameworks n = 24), communication and collaboration (n = 19) and safety (n = 19), and less on digital content creation (n = 16) and problem solving (n = 15).
5. Across economies, communication and collaboration was covered by 83.3% (n = 10) of developed-region frameworks against 60.0% (n = 9) of developing-region frameworks, while problem solving was slightly more frequent in developing (60.0%, n = 9) than developed (50.0%, n = 6) regions.
6. DigComp 2.2 amends 73 novel AI examples: communication and collaboration 31.51% (n = 23), problem solving 27.40% (n = 20), information and data literacy 19.18% (n = 14), safety 12.33% (n = 9) and digital content creation 9.59% (n = 7).
7. Among the 11 AI literacy frameworks, only Zhong and Liu's (2025) KAT framework does not treat ethics as an explicit dimension; technical knowledge spans all five DigComp content areas, while metacognitive regulation and higher-order thinking remain under development.

How the review selected and appraised its frameworks

The search ran across ERIC, Web of Science, ProQuest and Scopus, with two parallel branches — one carrying a generative-AI term block, one without — using Boolean strings in title, abstract and keywords. The without-AI branch returned 35,036 records and the with-AI branch 441. After duplicate removal and screening against eligibility criteria that required questionnaire or survey designs and samples of N ≥ 100, 162 records reached quality appraisal. Appraisal followed the COSMIN checklist on a three-point scale (0 = no evidence of reliability or validity; 1 = reliability or validity but not both; 2 = both), retaining only non-zero records with multi-item coverage. Thirty-nine were excluded; from 123 retained records, 67 unique frameworks plus two reports (ACARA, 2020; Fraillon et al., 2023) yielded 69. The with-AI branch retained 11 unique frameworks. Dual coding reached Cohen's κ = .89 to .92 across the three research questions, and DigComp 2.2 served as a shared reference taxonomy for cross-framework mapping rather than a normative standard.

What the validated frameworks contain

Inductive content analysis of the 69 digital literacy frameworks produced eleven element categories, stratified by role: 28 frameworks for students or citizens and 42 for educators, with one framework counted in both. For students, the five most emphasized categories align closely with DigComp: information and data literacy (n = 40), communication and collaboration (n = 20), safety and ethics (n = 19), digital content creation (n = 13) and problem solving and Creativity (n = 13), followed by technology literacy (n = 15). Educator frameworks distribute their elements more evenly and systemically: teaching, learning and pedagogy (n = 38), empowering learners (n = 25), curriculum and assessment (n = 25) and professional development (n = 8), with information and data literacy (n = 31) and technology literacy (n = 28) as the leading competencies. The authors note that educator-facing frameworks treat assessment design as part of the role, while cultural inclusiveness appears in only one educator framework and attitudes toward ICT in 7 of the 42. Student-centered frameworks emphasize discrete individual skills, whereas educator-centered frameworks feature systemic capacities such as planning, design and management of digital resources.

The digital divide, reframed for AI

The macro-level question is the digital divide. Mapping the 28 student and citizen frameworks onto DigComp 2.2's five competence areas, information and data literacy leads (frameworks n = 24), followed by communication and collaboration (n = 19) and safety (n = 19), then digital content creation (n = 16) and problem solving (n = 15). Comparing frameworks by World Bank FY2026 income classification, information and data literacy was covered in 86.7% (n = 13) of developing-region frameworks and 91.7% (n = 11) of developed ones; communication and collaboration in 60.0% (n = 9) against 83.3% (n = 10); and problem solving slightly more in developing (60.0%, n = 9) than developed (50.0%, n = 6) regions. Copyright and licenses were represented by no developing-region framework. The authors read the pattern through sociocultural theory as developed regions emphasizing integrative capacities while developing regions privilege operational skills that AI can increasingly automate — descriptive differences within the reviewed sample, not evidence of regional deficiency. Sensitivity checks removing higher-education-only instruments left the communication and problem-solving patterns intact, though information and data literacy and safety were less stable.

How AI reshapes digital literacy

At the chrono-level the authors map DigComp 2.2 onto the 11 AI literacy frameworks. DigComp 2.2 adds 73 novel AI examples, concentrated in communication and collaboration (31.51%, n = 23), problem solving (27.40%, n = 20) and information and data literacy (19.18%, n = 14), with less attention to safety (12.33%, n = 9) and digital content creation (9.59%, n = 7). The conceptual expansion is threefold: digital safety widens into AI ethics, higher-order thinking and metacognitive components gain prominence, and communication and collaboration are reinterpreted to include human–AI interaction. Technical knowledge is the most prominent AI literacy element, spanning all five DigComp content areas; AI ethics is second, and critical evaluation third, concerned with appraising outputs for accuracy, reliability, completeness and bias. Less consistently represented are metacognitive regulation, higher-order thinking and educator-specific human–AI collaboration, which the authors describe as still under development. Two of the eleven frameworks target educators, and the AI-Technological Pedagogical Content Knowledge (TPACK) extension models how content, pedagogical and AI-technological knowledge combine.

What this means for practice

• Curriculum designers. Treat DigComp 2.2 as an adaptable reference, not a global Benchmark. The review finds frameworks differ systematically by region and role, so competence lists should be localized to infrastructure, language and governance conditions rather than imported wholesale.
• Faculty developers. Build educator-facing frameworks around AI as a collaborative partner rather than a passive tool. Only two of the eleven AI literacy frameworks target educators, and the AI-TPACK extension points to composite knowledge — particularly AI-technological pedagogical knowledge — as the professional development target.
• Assessment designers. Shore up validity where generative AI decouples observable performance from durable learning: design AI-resilient tasks with transparent staged processes and real-time verification so completed work still evidences mastery.
• Policymakers. Watch the copyright and licensing gap. No developing-region framework represented it, and AI-generated content sharpens disputes over authorship and ownership, so digital literacy policy and copyright law need to move together.

Limitations

• The review covers publications from January 2014 to February 2025; the ≤2023 and >2023 split used for the period comparison is a lag-based descriptive proxy following the late-2022 release of ChatGPT, since framework design, validation and publication typically lag tool release.
• The macro-level comparison rests on a modest corpus of validated student and citizen frameworks stratified by participants' country or region and World Bank FY2026 income classification; most are instruments validated in specific educational settings rather than national-level frameworks, and socio-economic conditions vary within countries.
• Restricting inclusion to validated frameworks may overlook emerging competences absent from current instruments, such as prompt engineering; the authors note that Levy-Nadav et al. (2024) was excluded because it offers qualitative insight without empirical validation in target populations.
• The branching search did not partition the literature as intended: the with-AI branch yielded only 1 of the 11 frameworks uniquely, while 42 AI-related records entered the pool from the without-AI branch, so the branch distinction should not be read as an analytic contrast between generative-era and earlier framework work.
• Because AI literacy is treated as the umbrella construct, the review may under-represent frameworks published under narrower adjacent labels such as Machine Learning literacy or algorithmic literacy.

Citation

Guo, Z., Li, W., Chen, L. L., Liang, Q., & Wu, Y. (2026). Reconceptualizing digital literacy in the AI era: A systematic review of validated frameworks from an ecological perspective. Computers and Education: Artificial Intelligence, 11, 100679.