The stakes are higher than they might appear. Curriculum evaluation is the backbone of quality assurance in higher education, assessing whether a programme’s objectives, course structure, content sequencing, teaching strategies and assessment practices align with educational goals, industry needs and academic standards. Yet the evidence suggests many institutions still struggle with this task. The review cites data from the Association of University Programs in Health Administration indicating that roughly one in four health administration programmes in the United States fails to fully meet certification criteria related to annual assessment of student learning outcomes or the systematic use of those outcomes for programme improvement. Traditional evaluation models, many of which date back decades, were simply not designed for the volume, diversity and complexity of data that modern universities generate.

The dominant classical framework, Tyler’s objective model, emphasises alignment among learning objectives, instructional processes and outcomes, and remains widely used in accountability-driven education systems. Its successor, the CIPP model developed by Daniel Stufflebeam, broadened evaluation into a decision-oriented process spanning Context, Input, Process and Product, later extended to include Outcome, producing the CIPP-O framework. But as the review notes, both CIPP and CIPP-O were fundamentally conceived for evaluation environments dominated by manual processes and conventional analytics. They depend heavily on evaluators’ subjective judgement and struggle to manage the growing data intensity of contemporary higher education. This is precisely the gap the authors argue AI can fill.

From the 28 studies analysed, four major opportunities emerged. The first is learning personalisation: AI can detect individual learning needs, inform the design of personalised curricular structures, tailor materials to students’ preferences and provide real-time formative feedback that helps educators adjust their teaching strategies. The second is predictive analytics. Machine learning algorithms such as Random Forest, Support Vector Machines, k-Nearest Neighbours, Naïve Bayes and Logistic Regression can mine historical data, including which learning materials students accessed, how often they failed tests and their performance scores at different stages, to produce predictions that support early intervention and data-driven curriculum reform. The third and fourth opportunities, evaluation automation and administrative efficiency, proved the most dominant themes in the literature, represented in 25 and 18 of the reviewed articles respectively.

The technical machinery behind these gains is advancing quickly. Natural language processing can now map learning outcomes and analyse curricula with improved objectivity, as demonstrated in competency-based curriculum development for industrial engineering education. Large language models are being deployed to process vast volumes of textual data, from academic documents to student feedback, identifying curriculum weaknesses and generating evaluative recommendations. Studies of LLM-based curricular analytics in the United States show these systems can sift through granular learning activity data captured by modern learning management systems, including access timestamps, forum interactions and assignment completion, to detect problematic areas and offer evidence-based recommendations. In China, researchers exploring LLM-generated course evaluations report notable levels of rationality and interpretability, while fairness-enhanced LLM frameworks apply pre-processing, in-training and post-processing techniques to mitigate algorithmic bias.

Real-world institutional experiments are following. The University of Florida’s AI Across the Curriculum initiative requires every student to complete a foundational AI course alongside coursework in ethics, information and technology, and pairs the programme with industry collaboration on real-world problems. Edtech companies now offer AI-driven systems that perform automated curriculum mapping, aligning learning outcomes with competency standards or industry requirements, and universities in high-income countries have begun adopting these tools to keep programmes current. Across Asia, the growing volume of AI-related higher education research signals increasing institutional engagement with AI for teaching, assessment and curriculum quality assurance.

Yet the review is emphatic that the risks are equally real. The most serious challenge is data bias. When an AI system is trained on data drawn from a non-representative student population or limited, homogeneous sources, the model can reproduce and reinforce those biases in curriculum evaluation, disadvantaging students from underrepresented backgrounds. Biased datasets may also fail to capture the complex contextual factors that shape educational success, which is especially problematic in settings requiring deep contextual understanding of curriculum effectiveness. The authors warn that such opacity breeds distrust, because decision-makers often cannot understand or verify how AI systems generate their evaluative outputs.

Data privacy presents a second major concern. AI-driven curriculum evaluation demands stringent data-governance policies, transparent oversight mechanisms and robust protection of student personal information, with failures exposing institutions to serious ethical and legal consequences. Beyond privacy, the literature documents cultural resistance from conservative academic cultures unprepared for automated evaluation, limited educator understanding of how AI can be meaningfully applied, and human resource gaps intertwined with ethical risks. Excessive reliance on AI has been linked to reduced critical thinking and cognitive engagement, alongside concerns about plagiarism and misinformation. Infrastructure readiness compounds the problem: institutions lacking technological capacity face substantial integration difficulties, and the cost of acquiring, deploying and maintaining AI systems can be prohibitive for resource-constrained universities, potentially widening existing inequalities in educational quality.

To bridge these opportunities and challenges, the review proposes a conceptual framework called AI for Curriculum Evaluation aligned to CIPP-O, or AICE-CIPP-O. The framework positions AI as an analytical support layer across every stage of evaluation: needs analysis and policy document analysis at the Context stage; alignment assessment between programme and course learning outcomes at the Input stage; monitoring of curriculum implementation through learning analytics at the Process stage; analysis of assessment results and feedback supported by large language models at the Product stage; and data-driven recommendations at the Outcome stage. Crucially, the framework extends evaluation from a predominantly descriptive orientation towards diagnostic and predictive capability while preserving a human-in-the-loop mechanism, because the literature consistently finds that AI models cannot yet fully replace human evaluators and that final interpretation must remain under human oversight to protect academic validity and integrity.

The policy implications are far-reaching. The authors argue that successful AI adoption requires transparent governance structures of the kind emerging in Australia and the European Union, adequate digital infrastructure, ethical data governance with strong student-privacy safeguards, algorithmic transparency, and sustained capacity building for educators, potentially accelerated through train-the-trainer approaches. Curriculum developers are encouraged to pursue continuous formative evaluation through learning analytics in collaboration with AI specialists and data analysts, while students themselves need education in AI ethics embedded within the curriculum. The review also identifies significant research gaps: few empirical studies have examined the long-term effectiveness of AI within the curriculum evaluation cycle, cross-country comparative studies remain scarce, standardisation within accreditation frameworks is limited, and no systematic benchmarking has compared AI-based evaluation outputs with those of human experts. The review’s limitations are candidly acknowledged, including the predominance of conceptual rather than empirical studies in the field, with 18 of the 28 analysed articles offering no empirical data, and the concentration of research in contexts with advanced digital infrastructure. The overall message, however, is clear: AI will not replace the academic judgement that curriculum evaluation depends on, but universities that pair intelligent tools with clear governance, human oversight and genuine institutional readiness stand to make that judgement faster, fairer and far better informed.

Subject of Research: The use of artificial intelligence to support curriculum evaluation and quality assurance in higher education

Article Title: Artificial intelligence in curriculum evaluation through a narrative systematic review of opportunities and challenges in higher education

Article References: Aryani, L. N., Widiana, I. W., Suharta, I. G. P., Lasmawan, I. W., & Arnyana, I. B. P. (2026). Artificial intelligence in curriculum evaluation through a narrative systematic review of opportunities and challenges in higher education. Discover Education, 5(1), Article 989. https://doi.org/10.1007/s44217-026-02177-3

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02177-3

Keywords: artificial intelligence, curriculum evaluation, higher education, quality assurance, predictive analytics, learning personalisation, large language models, algorithmic bias, CIPP model, data privacy, institutional readiness, human-in-the-loop