Abstract
The paper substantiates a pedagogical model for integrating an AI assistant into 3D modeling instruction to improve the effectiveness of forming spatial concepts in middle school students. A typology of AI prompts (conceptual, procedural, and metacognitive) is proposed, along with pedagogical constraints that prevent replacing learning with ready-made solutions (regulated queries, mandatory decision logging, double-checking, and final reflection). Within a design-based research approach, a 6-week quasi-experimental design is outlined: the intervention group learns 3D modeling with AI support, while the comparison group follows a traditional AI-free approach. A set of assessment tools is described, including a spatial visualization test (mental rotation and assembly/disassembly tasks), exercises on views and projections, a 3D model quality rubric, a conceptual error checklist, and a cognitive load self-report scale. A key methodological product is the matrix “spatial concept → error type → AI support strategy”, which links typical errors (axis confusion, ‘planar’ depth interpretation, incorrect nets, failure to preserve invariants) to appropriate prompts and reflection questions. The expected intervention effects include gains in diagnostic outcomes, a reduction in conceptual errors, higher model quality, and increased learner autonomy supported by a controlled decrease in dependence on procedural prompts.