Cultural and Digital Readiness for AI-Integrated Higher Education: A C-TAM-Informed Curriculum Design Framework

Synthesis: Bennett (2026) argues that generative AI integration in higher education is a curriculum design problem before it is a technology adoption problem. The article is a theory-building synthesis, not an empirical test. It reads post-2022 student studies through TAM and UTAUT, cross-cultural research, AI literacy frameworks, backward design and constructive alignment, and it explicitly refuses to relabel PISA 2018 or World Values Survey indicators as measures of GenAI acceptance. Five mechanisms — usefulness, ease and competence, trust and ethical legitimacy, social norms, and culturally patterned expectations about uncertainty and authority — are carried into testable propositions P1–P6 and a reconstruction matrix. The practical core is calibrated trust and explicit assignment-level permissions, so that students who need certainty reach the same outcomes by a clearer route.

Key Findings

1. The paper offers a framework, not results. It is a theory-building integrative evidence synthesis. No new data were collected, no structural equation model was estimated, and the article reports no path coefficients, model-fit indices or cross-national latent mean comparisons.
2. Usefulness has to be designed into the assignment. Usefulness remained the most stable acceptance mechanism across the reviewed studies, but Baek et al. (2024), surveying 1,001 U.S. college students, found use varying by task, student characteristics and institutional policy. The curriculum response is to state why AI is present and which human capability stays accountable.
3. Ease of use includes competence, access and permission. Delcker et al. (2024) linked AI competence to intended and actual use among 634 first-year students, and Strzelecki (2024) found performance expectancy predicting ChatGPT intentions among 503 higher education students. Students without paid access, reliable devices, language confidence or verification skill face higher effective effort even when the interface is simple.
4. Attitudes depend on calibrated trust. Students may value speed while distrusting accuracy, or welcome language support while worrying about authorship. The framework therefore separates positive attitude from uncritical use and asks assessment to make judgment visible through source audits, error analyses, prompt logs or oral explanation.
5. Norms are part of the curriculum, not a backdrop. Student uncertainty grows when instructors, departments and institutions send conflicting signals. The paper recommends a three-level permission label — no AI, limited or supportive AI, and integrated AI with disclosure — attached to each major assignment rather than a single syllabus statement.
6. Culture shapes interpretation, not capability. High uncertainty avoidance and high power distance are treated as predictable design risks: vague rules disadvantage students who need certainty, and silent norms disadvantage students who wait for authorization. The matrix answers both with decision trees, worked examples, modeled verification and low-stakes rehearsal.
7. The evidentiary boundary is stated explicitly. PISA 2018 measured 15-year-olds before ChatGPT's public diffusion, and WVS Wave 7 measures broad values rather than classroom behavior. Both are used only as contextual evidence about the readiness pipeline entering higher education, never as GenAI constructs.

How the framework was built

The synthesis connected three literatures that are usually kept apart: direct post-2022 studies of GenAI acceptance and use, established acceptance and cross-cultural theory, and curriculum, assessment and AI-literacy frameworks. Sources were classified as direct GenAI evidence, established theory, curriculum theory or contextual readiness evidence, then mapped onto six C-TAM domains and translated through the three stages of backward design: desired outcomes, acceptable evidence, and pedagogy and support.

The method section is unusually explicit about what was removed. An earlier version of the manuscript reported a structural equation model using PISA 2018, WVS Wave 7 and Hofstede indicators, but the item-level dataset, code, country composition, weights and construct mappings were unavailable for reproducible reanalysis. The revision drops those estimates rather than retain unverifiable statistics, and it treats the older sources as framing devices for readiness questions. The resulting claims are narrower and, the author argues, more credible.

Measurement invariance is not assumed anywhere in the paper, because no cross-cultural latent-variable model is estimated. The appendix specifies what a direct test would require: current task-specific GenAI measures, individual-level espoused cultural values, configural, metric and scalar invariance testing, multilevel modeling of students within courses and institutions, and reporting of coefficients, intervals and code.

From mechanisms to a reconstruction matrix

C-TAM keeps perceived usefulness and perceived ease of use but situates them in social, cultural, ethical and institutional conditions. Its five mechanisms are expressed as propositions P1–P6. P1 predicts that usefulness will raise responsible intention when AI activity is visibly tied to disciplinary outcomes. P2 predicts that competence, access and scaffolded practice will reduce perceived effort. P3 predicts that verification practice and ethical clarity will strengthen attitudes more than encouragement to use GenAI. P4 predicts that instructor, peer and institutional norms will matter more where authority distance is high. P5 predicts that uncertainty avoidance will lower willingness to experiment under ambiguous policy, and that the relationship will weaken with exemplars, decision rules and rehearsal. P6 adds long-term professional relevance and equitable access.

The translation into practice is the reconstruction matrix, which pairs each backward-design element with an adaptation for students combining high uncertainty avoidance with high power-distance expectations. Learning outcomes distinguish required, optional and prohibited AI uses and connect each rule to its learning purpose. Assignment instructions provide a decision tree and instructor confirmation instead of "use your judgment" without criteria. Low-stakes practice begins with instructor-modeled demonstration and guided group rehearsal before individual experimentation. Assessment requires products plus process evidence, with an exemplar of what counts. Feedback rubrics reward verification and explain the consequences of unverified claims. Participation designs use structured roles and facilitated discussion before public disagreement with peers or AI output. Academic integrity guidance distinguishes misuse from good-faith error during early practice. Career relevance makes profession-specific accountability explicit, alongside critical discussion of labor, bias, privacy and environmental costs.

The matrix is a design heuristic, not a validated intervention. The paper is candid that its value depends on future testing: whether clearer permissions, structured rehearsal and equitable access actually reduce gaps in participation, confidence, learning and integrity misunderstandings without suppressing autonomy.

What this means for practice

• Instructors: Replace general AI encouragement with assignment-level statements that name the learning purpose, the permitted level of assistance, the required evidence and the access alternative.
• Administrators: Treat policy clarity as pedagogical infrastructure. Institution-wide principles help, but students need consistent signals across syllabi, and staff need common access to tools or a no-cost equivalent.
• Designers: Start from the outcome and its evidence, then decide where AI belongs. Build verification into tasks rather than bolting it on as a warning, and assess process artifacts instead of relying on AI detection.
• Researchers: Use the validation protocol as a checklist. The framework's propositions are only testable with current GenAI measures, individual-level cultural values and explicit invariance testing.

Limitations

This is a theory-building synthesis, not a systematic review or an empirical test. Four contemporary studies illustrate convergent mechanisms but do not exhaust a fast-moving literature, and their differing samples, institutions, countries and outcome measures mean the synthesis cannot estimate an average effect or rank cultural predictors. The PISA and WVS sources remain temporally and substantively limited: PISA 2018 participants were 15-year-olds in secondary education, and its digital measures predate widespread GenAI. Culture is handled as a source of hypotheses rather than a measured variable, so no claim about a national group's acceptance is supported. The reconstruction matrix has not been evaluated, and the paper reports no evidence that its adaptations close participation gaps. No new survey or focus-group data were added, and the appendix protocol describes a study that has not yet been run.

Citation

Bennett, Jeremy. (2026). Cultural and Digital Readiness for AI-Integrated Higher Education: A C-TAM-Informed Curriculum Design Framework. Advances in Curriculum Design & Education, 2(4). https://doi.org/10.63808/acde.v2i4.391