This paper explores the suitability of ChatGPT-4o as a virtual learning guide in vocational education, specifically in the context of electrical installation technology. Through a detailed case study, the ability of ChatGPT to support trainees in solving a complex, practice-oriented task is evaluated. The task includes planning the electrical installation for a bedroom based on a floor plan, requiring both textual and graphical solutions such as circuit diagrams, installation schematics, and a bill of materials.

Using a structured prompt design, ChatGPT was queried in multiple phases of the task. The responses were evaluated according to criteria such as technical accuracy, didactic value, and the handling of graphical content. The analysis shows that ChatGPT is capable of delivering accurate and well-structured textual explanations, particularly in areas like switch functionality, line selection, safety regulations, and wiring principles. These answers often use correct terminology and provide reasoning appropriate for the cognitive level of apprentices.

However, the model demonstrates significant limitations when it comes to interpreting or generating standard-compliant technical drawings such as circuit diagrams or installation plans. Graphical responses frequently contained technical errors, misused symbols, and lacked adherence to standards like DIN EN 60617. ChatGPT also struggled to recognize mistakes in student-submitted diagrams, which limits its potential as a feedback tool in tasks involving visual representation.

From a didactic perspective, while the chatbot presents technically sound answers, it often provides full solutions without prompting learners to engage in deeper reflection or problem-solving. This contradicts constructivist learning principles, which emphasize the importance of active engagement and iterative feedback.

The case study concludes that ChatGPT-4o may be a useful supplementary tool in vocational training—particularly for textual problem-solving, explanation of core concepts, and support during independent learning phases. However, it is currently not a suitable replacement for human guidance when tasks involve graphical analysis or require adaptive, dialog-driven scaffolding. For effective implementation, its use should be embedded in a thoughtful instructional design that addresses its strengths and limitations.

Future research should examine whether these findings generalize to other trades and how prompt engineering or model improvements might enhance reliability in technical and graphical domains.

Keywords: Vocational education, artificial intelligence, learning guide.