Mapping artificial intelligence integration in higher education: A systematic review using the FACETS and SAMR frameworks

Synthesis: Mapping artificial intelligence integration in higher education — a PRISMA 2020 systematic review by AlSheikh et al. (2026) that screened 959 records across eight databases to map how AI is being integrated into undergraduate higher education, classifying 22 included intervention studies with the FACETS framework (Form, AI use case, Context, Education focus, Technology, SAMR) and grading their depth of educational transformation with the SAMR model (Substitution, Augmentation, Modification, Redefinition). The review finds that AI integration remains largely incremental — clustering at Substitution and Augmentation rather than transformation — that it is dominated by generative AI applied to assessment automation and personalized learning support, and that equity, ethics, academic integrity, and faculty readiness remain undertheorized.

Overview

Generative AI tools like ChatGPT are reshaping postsecondary teaching, learning, assessment, and curriculum design, yet the literature on AI integration in higher education remains fragmented and lacks structured frameworks for comparing and evaluating integration. This review addresses that gap by combining two complementary lenses: the FACETS framework (a coding schema that characterizes AI interventions by Form, AI use case, Context, Education focus, Technology, and SAMR level) to enable systematic, comparative description, and the SAMR model to gauge the extent of educational transformation each intervention affords. Rather than appraising effectiveness, the review maps the landscape — documenting what has been implemented, where evidence concentrates, and where it is missing.

Method

• Design: A systematic review conducted per PRISMA 2020 and registered with PRISMA-P, using narrative synthesis.
• Sources: Eight databases (PubMed, Medline, Web of Science, ProQuest, Scopus, Dimensions, OpenAlex, IEEE Xplore) were searched for English-language studies published between January 2015 and July 2025, yielding 959 records.
• Included studies: After duplicate exclusion and title/abstract/full-text screening, 22 intervention-focused studies met the inclusion criteria, most from North America and Asia, spanning medicine, nursing, engineering, computer science, and co-curricular wellness.
• Coding: Studies were classified with the FACETS framework and graded on the SAMR model; outcomes and ethical and equity considerations were extracted and synthesized thematically.
• Quality appraisal: No formal risk-of-bias instrument was used, consistent with the review's mapping (rather than efficacy) aim.

Key findings

• Generative AI dominates the landscape. Large language models — especially ChatGPT, GPT-3/4, and writing-assistance tools — were the most frequently reported platforms, while predictive analytics, LMS integrations, robotics, and Multimodal AI or affective computing remained peripheral.
• Assessment automation and personalized support lead the use cases. Applications clustered around automated grading, plagiarism detection, and formative feedback, plus adaptive pathways and recommender systems for personalized learning support.
• Integration is largely incremental, not transformative. Most studies sat at the SAMR Substitution or Augmentation level, with fewer at Modification and only one nearing Redefinition — AI was usually introduced to improve existing practices rather than reconceptualize curricula or learning outcomes.
• Two Redefinition exemplars. Zhao et al.'s whole-curriculum biomedical-informatics course used AI as a co-creator across course design, assignments, presentations, reflection, and debate, while Jeong's embodied social robot delivered 24/7 Well-Being coaching previously unfeasible at scale — the only interventions approaching genuinely novel educational practices.
• Benefits versus challenges. Reported benefits were efficiency, personalization for learning and evaluation, engagement, and extended educator reach; challenges centered on equity, Ethics, and academic integrity, alongside risks to critical thinking and over-reliance.
• Disciplines and geography are uneven. Medicine, nursing, engineering, and computer science led integration, with humanities, social sciences, and interdisciplinary programs scarce; research concentrated in North America, East Asia, and Europe, with minimal representation from the Global South.
• Five documented evidence gaps. The review flags limited evidence of pedagogical transformation, under-explored equity and access implications, scarce research on faculty readiness and institutional support, inadequate ethical and governance frameworks, and a pronounced over-reliance on generative AI to the neglect of robotics, multimodal, and hybrid human–AI collaboration.

What this means for practice

• Instructors. Use AI as a learning partner rather than a threat: have students draft solutions or explanations with an AI tutor, then critique and improve them, so routine knowledge work is offloaded and class time goes to deeper discussion — while deliberately preserving the mentorship, empathy, and nuanced feedback AI cannot replicate.
• Instructors. Redesign assessments for AI-rich contexts. Because generative AI can already solve standard questions, shift weight to critical thinking, personal input, open-ended reflection, oral defenses, and applied projects, and permit transparent AI use (for example, "draft with AI, then document how you improved it") so the tool becomes part of the learning rather than a cheating route.
• Instructors. Cultivate AI Literacy and ethical use directly. Teach the limits of the technology — confident-but-wrong answers, hallucination, bias — require students to verify outputs and cite sources, and set guidelines that separate permitted exploratory use from plagiarism.
• Administrators. Invest in faculty development and institutional strategy, not just micro-level tool adoption: prompting and pedagogical redesign for staff, clear honor-code language on AI, and communities of practice for educators.
• Policymakers. Close the equity and governance gaps this review documents. Only 22 intervention studies met inclusion, equity and access implications are under-explored, and ethical and institutional frameworks remain thin, so require governance and faculty-readiness work alongside tool procurement.

Limitations

• The review included only 22 intervention studies after screening 959 records from eight databases, and these cluster in medicine, nursing, engineering, and computer science from North America and Asia, with minimal representation from the Global South and scarce coverage of the humanities and social sciences.
• The search was restricted to English-language publications, and gray literature and unpublished research were excluded, so relevant work may have been missed.
• No formal risk-of-bias instrument was used, consistent with the review's mapping rather than efficacy aim; integration depth was graded on the SAMR model and synthesized thematically, not appraised for effect size.
• Synthesis was narrative, which the authors state restricts causal or generalizable conclusions, and the evidence base is time-sensitive: many included studies were exploratory, limited in scale, or pilots, with a single Redefinition-level example.

Citation

AlSheikh, M. H., Zaini, R., ALmulhem, M. A., & Ahmad, S. (2026). Mapping artificial intelligence integration in higher education: A systematic review using the FACETS and SAMR frameworks. Frontiers in Education, 11, 1871468.