From Answer-Seeking to Reasoning: A Request Protocol for Sustainable Integration of Generative AI in Engineering Graphics Education

Sustainable Development Goal 4 (SDG 4) emphasizes inclusive and equitable quality education and lifelong learning opportunities for all. Within this broader framework, the sustainable integration of generative artificial intelligence (AI) in engineering education raises a specific question: whether students can retain and apply the competencies developed during AI-supported learning when the tool is no longer used. Generative AI can provide learning support beyond limited classroom hours, but it can also perform the cognitive operations that students are expected to master. Using an engineering graphics course, this study compared three strategies of AI use for their effects on how long student competence lasts. The three strategies were no AI, unrestricted AI use, and a request protocol. Under the protocol, students may ask questions about work they have already produced, but they may not ask for the answer. Competence was measured four times under an instruction not to use the tool, and the last measurement fell four weeks after the course ended. The ranking changed after withdrawal. Immediately after instruction, the unrestricted group scored highest before adjustment, but the adjusted between-group differences were not significant. At the delayed test the protocol group held the largest advantage of any contrast and the unrestricted group fell to last. Both adjusted differences were significant at p < 0.001. The advantage was largest among students who started lowest and was no longer distinguishable from zero above a baseline score of 64.95. The difference between the two tool conditions sat in classification rather than drafting. Curve-type misjudgment was about six times as frequent in the unrestricted condition as under the protocol, while the drafting conventions did not differ detectably at the end of term. Protocol adherence acted as a graded skill rather than a compliance record and tracked delayed competence (r = 0.63, n = 33). In this 48 h foundation course, regulating what students asked, not how much they used the tool, was associated with competence that persisted after withdrawal.