The evolution of open educational practices: a comparative analysis of models and a new framework for the AI era

Synthesis: Duan (2026) asked how open educational practices have been modeled over roughly two decades, and whether those models still fit an era in which generative AI is embedded in teaching. The answer comes from a systematic review: searches of Web of Science, Scopus, ERIC, and Google Scholar covering 2007 to 2024, screened under PRISMA 2020, which reduced 540 records to 14 models and frameworks. No primary data were collected from teachers, institutions, or learners; the evidence base is the 14 published models plus the author's comparative matrix across six dimensions. The review reports a trajectory in three parts, from resource access to creation and curation, from resource use to open pedagogy and learner agency, and from technology as delivery tool to technology as infrastructure. It then proposes an AI-empowered Open TPACK framework in which AI sits beneath content, pedagogy, technology, and openness rather than inside technological knowledge.

Key Findings

1. The PRISMA screening reduced 540 identified records to 14 included models: 415 were excluded on title and abstract, 125 full texts were assessed for eligibility, and 111 of those were excluded for lacking a clear OEP model, not being original articles, or not being academic or English.
2. In the resource dimension, Ehlers' (2011) readiness matrices and Wild's (2012) engagement ladder with its piecemeal, strategic, and embedded levels treat OER as material to be adopted, while Stagg's (2014) continuum and Arinto et al.'s (2017) social inclusion framework move the learner into co-creation.
3. In the teaching dimension, Hegarty (2015) specifies eight attributes of open pedagogy, Paskevicius (2017) maps open practice onto outcomes, resources, activities, and assessment, and Baran and AlZoubi (2020) name six factors shaping student participation.
4. Tietjen and Asino (2021) make sustainability the organizing principle, building a five-circle framework around diversity of cultural voices, participatory pedagogy, open licensing, learners outside formal education, and a culture of collaboration.
5. In the technology dimension, the EU OpenEdu framework (2016) places technology among four transversal dimensions alongside six core dimensions, Huang et al. (2020) make enabling technologies the foundation of the OEP ecosystem, and Duan (2024) adds openness as a fourth domain to TPACK.
6. The review's own contribution is the AI-empowered Open TPACK framework, which positions AI as an enabling layer under all four domains rather than as a sub-dimension of technological knowledge or a discrete digital skill.
7. The paper periodizes the literature into three waves: early OEP work on readiness and resources, a turn toward pedagogical interaction, and contemporary scholarship on social justice and technology.

Fourteen models and how they were chosen

The review followed three phases: search and collection, screening, and comparative synthesis. The search strings combined "Open Educational Practices," "OEP," "Open Pedagogy," "OER models," and "Open Education Frameworks," joined with Boolean operators, across four databases chosen for coverage of journals, conference proceedings, and books. Reference lists of the main articles were also searched by snowballing. Screening used PRISMA 2020, with exclusions for articles that had no OEP model, no clear model, were not original, or were non-academic or non-English.

The 14 survivors were not a census. The author describes purposive sampling on relevance, representativeness, intelligibility, and novelty, and reports balancing the set across three time periods and across micro-level frameworks aimed at individual teachers and macro-level frameworks aimed at institutional policy. Two exclusions are named and both matter. Wiley's early 5R framework was left out because the study focuses on practices rather than permissions, and Global South regional models were left out because the databases were English-language. The paper reports no quality appraisal of the included models and no second coder.

Three dimensions and the direction of travel

The analytical work is a comparison across focus, key actors, core processes, underlying values, technological emphasis, and level of application, presented as a matrix of all 14 models. Read along the resource dimension, the field starts with access and adoption, thickens into reuse, and ends with students producing and curating. Read along the teaching dimension, the center of gravity shifts from the teacher's use of someone else's material to the learner's agency in a participatory process, with the instructor recast as guide and co-creator. Read along the technology dimension, technology moves from an unexamined delivery channel, as in Wild, to a named component, as in Huang et al., to a transversal condition of everything else.

Each reading is plausible and each is assembled from published models rather than from evidence about practice. That distinction matters for how the trajectory should be taken. A sequence of frameworks published between 2011 and 2024 records what scholars chose to emphasize, which is not the same as what institutions did. The review is careful about individual models and less careful about that inferential step, and its claim that the field is "evolving into clever ecosystems" is a reading of the literature's direction rather than a finding about the sector.

Why AI goes underneath TPACK

The framework proposal is the paper's real argument, and it is a claim about architecture. Duan's earlier Open TPACK added openness to Mishra and Koehler's technological, pedagogical, and content knowledge, so openness became a fourth domain a teacher had to hold. The new version responds to what the author calls a shift from digital infrastructure to generative infrastructure, in which AI changes not only the resources produced but how pedagogical processes are defined and how openness is enacted. If AI is infrastructure, it cannot sit inside technological knowledge as one more tool to master.

So the model places AI as a layer beneath the four domains, following the logic Huang et al. used for enabling technologies. The roles it sketches are a co-pilot in learning design, an automated curator of open resources, and a personalized scaffold for open pedagogy. What the paper does not provide is any test of the framework: no case of a teacher using it, no rubric derived from it, no comparison with the earlier Open TPACK, and no data on whether design or learning changes when AI is treated as infrastructure rather than as a tool. The claim is conceptual, and the paper's abstract describes its results as empirical when what was analyzed is a set of documents.

What this means for practice

The usable part for instructors is diagnostic rather than prescriptive. The three dimensions give a way to locate a course or a project: whether the open work is mainly about finding and adopting material, about who is producing it, or about the platform it runs on. Most course-level open pedagogy sits in the second dimension, and the review's finding that the models moved there is a reasonable prompt to ask whether a given assignment actually gives students agency or only gives them a free textbook. For faculty developers, the paper's recommendation is to build "smart instructional design capabilities" rather than train single tools, so that teachers analyze, contextualize, adapt, and refine open resources instead of generating replacements for them. Institutions should read the technology dimension as a warning about procurement: the framework treats the platform as part of the pedagogy, not as a neutral container.

Limitations

• The corpus is English-language only and excludes Global South regional models by design, along with Wiley's 5R framework, so the periodization describes an Anglophone literature and may flatten work that did not appear in the four databases searched.
• The review is single-authored with no reported second coder, no inter-rater agreement, and no quality appraisal of the 14 models. The comparative matrix and the three-wave periodization are one reader's judgments about texts.
• The proposed AI-empowered Open TPACK framework is not tested or operationalized. No instrument, rubric, worked example, or evaluation accompanies it, and the paper's own call for future research concedes that its practical use is unexplored.
• The search window closes in 2024, so the account of AI in open education rests on 2023 and 2024 sources and predates most empirical work on generative tools in open pedagogy.

Citation

Duan, C. (2026). The evolution of open educational practices: a comparative analysis of models and a new framework for the AI era. Smart Learning Environments, 13, 15.