Effectiveness of Artificial Intelligence Interventions in Higher Education: A PRISMA 2020 Systematic Review with Structured Narrative Synthesis
- Pornsawan Chaimeerang
- Piyathida Sripol
- Pongsuk Dornpinij
- Phanuwat Ruangkulsap
- Thada Jantakoon
- Wantipa Unarat
Abstract
Artificial intelligence (AI) is being adopted in university teaching faster than evidence on its instructional value is being consolidated. This systematic review examined whether AI-based interventions improve achievement, engagement, satisfaction, and knowledge retention among undergraduate and graduate students relative to conventional instruction. Following PRISMA 2020, seven queries in SciSpace, Google Scholar, and PubMed, run on September 23, 2026, identified 902 records. After deduplication and a relevance-ranking cut-off, 500 records were screened, 82 reports were assessed for eligibility, and the 80 provisionally eligible reports were verified against publisher, index, or repository records. Reports that could not be verified (n = 52), proved ineligible (n = 4), or existed only as non-peer-reviewed preprints (n = 2) were excluded, leaving 22 peer-reviewed reports: 14 primary comparative studies and eight meta-analyses. Evidence was integrated through a structured narrative synthesis without meta-analysis, using design-based robustness indicators. Five primary studies were randomised. Meta-analyses restricted to higher-education students reported moderate effects on achievement in college populations (Hedges’ g = 0.32–0.53) and inconsistent effects in medical education (SMD = 0.27–2.06); broader-population syntheses were treated as contextual only. Primary studies were favourable when AI was embedded in structured pedagogical designs, whereas comparisons with well-designed conventional or human instruction often showed no advantage, and unguided generative AI use improved immediate performance more than retention or transfer. Given the small, largely non-randomised evidence base and declared departures from standard search practice, the evidence supports only tentative, design-dependent claims about AI in higher education.
- Full Text:
PDF
- DOI:10.5539/hes.v17n1p66
Index
Contact
- Sherry LinEditorial Assistant
- hes@ccsenet.org