LLM-Enabled Educational Artefacts in Computer Science Education: A Systematic Mapping Study

Abstract

The use of Large Language Models in computer science education has produced a fast-growing body of primary studies. Existing secondary studies tend to categorise this literature by external criteria, such as the model used, the course it operates in, or the benefit it reports; the educational artefact itself is discussed but not adopted as the unit of analysis. This study adopts it: the LLM-enabled educational artefact (LEEA), a deliverable built on an LLM for educational purposes, either technological, such as a tool, system, or agent, or methodological, such as a model or technique, rather than the general-purpose model that powers it. Through a review of scientific databases, 158 primary studies published between 2022 and February 2025 were analysed using a six-dimensional codification scheme and six research questions. The mapped studies concentrate on a few options: 69.6% of contributions focus on tools and systems, 84.8% rely on the GPT family, and 86.1% target undergraduate education. Feedback appears in 76.6% of studies, while other instructional stages remain underexplored. Evidence comes mainly from user perception and system performance, which rarely overlap; only 21.5% of studies measure learning gains, so the evidence on learning remains limited. The study contributes a multidimensional map of this literature and a codification scheme detailing eleven educational functions and seven families of effectiveness metrics derived from the corpus.

The authors' abstract, as published at the source. Applied Sciences, 2026 · DOI ↗