The study, conducted by Abroto Abroto, Maemonah Maemonah, and Imam Machali of Sunan Kalijaga State Islamic University in Yogyakarta, Indonesia, analyzed 547 peer-reviewed journal articles indexed in the Scopus database between 2018 and 2025. Using a rigorous screening process modeled on the PRISMA framework, the researchers began with 1,366 records and filtered them down to journal articles published in English, removing duplicates and conference papers along the way. They then applied performance analysis and science mapping techniques with the VOSviewer software, tracing collaboration networks, co-citation patterns, and keyword co-occurrences to reconstruct the intellectual architecture of a field that barely existed less than a decade ago.

The headline finding is growth. In 2018, only 23 articles on AI in primary education appeared in the database. By 2025, that figure had climbed to 132, the peak of an exponential curve that shows no sign of flattening. The corpus involved 1,876 distinct authors, an average of 3.7 researchers per paper, and stretched across 27 subject areas. Social sciences led the way with 429 articles, or 41.7 percent of the total, followed by computer science at 18.9 percent and psychology at 6.2 percent. That distribution alone tells a story: this is a field born at the intersection of pedagogy and technology, but one that is still searching for a disciplinary home.

Perhaps the most striking structural insight comes from the journals themselves. Of the 293 outlets that published at least one relevant article, nearly 69 percent published only a single paper on the topic. Just 16 journals, about 5.5 percent, published five or more. The field, in other words, is scattered across a vast landscape of periodicals, with only a small core of specialized venues. Among these, Computers and Education stood out as the most influential, amassing 914 citations, an H-index of 12, and elite bibliometric scores, while Sustainability published the largest raw number of articles. Most of the leading journals carried Q1 rankings, suggesting that despite its youth, the research is flowing disproportionately through high-impact channels.

Geographically, the map is strikingly lopsided. Spain dominates both productivity and citation impact, with the Universidad de Sevilla topping the institutional table at six publications, followed by the Universitat de València, Abai Kazakh National Pedagogical University, the Universidad de Granada, and the Universidad de Murcia. Not a single institution from North America, Africa, or most of the developing world appears among the top ten. Yet 111 countries contributed to the literature overall, with emerging contributors such as Indonesia and Taiwan gaining ground. The authors attribute the imbalance to disparities in research capacity, funding, and policy priorities, and warn that it raises uncomfortable questions about whose classrooms the current evidence base actually describes.

Individual researchers tell a similar story of concentrated influence. Raúl Cózar-Gutiérrez and José Antonio González-Calero of Spain emerged as the most productive authors, each with seven articles and 182 citations, while Tebogo Moloi and Dare Mhlanga topped the global citation rankings with 283 citations from a single collaborative paper. The lesson, the study notes, is that influence in this field is driven less by volume than by a handful of exceptionally resonant contributions, many of which focus on teacher readiness and the pedagogical implications of AI in primary contexts.

The thematic analysis is where the study delivers its most consequential findings. Keyword co-occurrence mapping of 1,669 unique author keywords revealed three major clusters. The first, centered on educational technology, teacher training, and digital competence, represents the institutional and pedagogical foundations of the field. The second groups artificial intelligence with augmented reality, e-learning, and computer-aided instruction, capturing the push toward intelligent, immersive, and personalized learning systems. The third, a green cluster of humanistic and developmental themes including curriculum, learning, and child development, sits conspicuously at the periphery of the network. Even more telling is what is missing: keywords related to ethics, data privacy, child protection, algorithmic bias, and students’ psychological well-being are largely absent from the dominant thematic structure altogether.

The longitudinal analysis deepens that concern. Between 2018 and 2020, research focused on general digital technology integration, with AI barely visible as a standalone concept. From 2021 to 2022, a transitional phase shaped partly by the COVID-19 pandemic’s forced digitization, themes multiplied and attention shifted to institutional adaptation and teacher readiness. Since 2023, the field has consolidated dramatically, narrowing to a handful of concentrated themes: artificial intelligence, personalized learning, and assessment. That convergence signals maturity and an application-oriented agenda, but the authors caution that it also reflects a field racing toward implementation while ethical and regulatory frameworks lag far behind, a pattern the pandemic era already demonstrated can accelerate adoption without building safeguards.

Why does this gap matter so much for primary education specifically? Young children are uniquely vulnerable to the risks that AI systems introduce: their data privacy can be compromised before they can understand what data is, algorithmic decisions can shape their learning opportunities in ways they cannot contest, and heavy reliance on digital tools may affect developing attention, social skills, and motivation. The study argues that effective AI integration at this level remains fundamentally teacher-mediated, and that professional development must therefore extend beyond technical training to instructional design, critical evaluation of AI-supported learning, and responsible data practices. Policy, meanwhile, needs clear guidelines on privacy, transparency, equity, and child protection before deployment, not after problems emerge.

The researchers are candid about their limitations, noting that reliance on a single database, the exclusion of conference literature, and the absence of local citation analysis all constrain the picture, and that fast-moving developments like generative AI and large language models may not yet be fully captured in the index. Still, their conclusions are hard to dismiss. AI in primary education has completed a remarkable journey from peripheral curiosity to consolidated research domain in under a decade, but it has done so while treating the children at its center largely as subjects of effectiveness studies rather than as developing humans whose rights, well-being, and long-term futures deserve equal scientific scrutiny. The next phase of the field, the authors suggest, must be built on ethically grounded, pedagogically informed, and developmentally appropriate foundations, because the stakes of getting it wrong are measured not in citation counts but in childhoods.

Subject of Research: Bibliometric mapping of artificial intelligence research trends in primary education

Article Title: Mapping artificial intelligence research in primary education: a bibliometric study

Article References: Abroto, A., Maemonah, M., & Machali, I. (2026). Mapping artificial intelligence research in primary education: a bibliometric study. Discover Education, 5(1), Article 824. https://doi.org/10.1007/s44217-026-02012-9

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02012-9

Keywords: artificial intelligence, primary education, bibliometrics, personalized learning, educational technology, ethics, child development, teacher training, VOSviewer, Scopus, thematic evolution, data privacy