Comparing transparent open and proprietary AI tutors in religious education through an empirical study

Abstract The role of transparent AI systems is being advocated increasingly in educational design, but there are very few studies on how transparency affects value-saturated lessons. This paper assesses Apertus, a proclaimed open and explainable multilingual chatbot, and compares it to GPT-5.1, responding to thirty lesson-specific tasks within Religious Education at lower secondary educational level. The generated responses were numerically coded according to a devised rubric including theological consistency, contextual specificity, doctrinal objectivity, linguistic clarity, and rhetorical consistency. Paired samples t-tests established that Apertus outperformed GPT-5.1 on each dimension, including modest but systematic positive differences (Δ = 0.15–1.04) and very large effect sizes (d p = 2.21–2.57). A comprehensive qualitative investigation into matters concerning themes pertaining to notions about faith, reason, moral obligations, and dialogical engagement established significantly more informative ideas about Apertus’ demonstrations. The study shows that open AI can facilitate dialogical processes within Religious Education lessons by making traceable rational processes and contributing to reflective educational goals. These results make sense within an increasingly acknowledged consensus about how better transparency can make more trustworthy educational artificial intelligence. The implications of this paper lie within making transparent artificial intelligence more prominent, especially within those domains which involve faith matters and moral debates to which Religious Education is pertinent.