Evaluating the quality of educational content is essential before deploying it in online learning environments. However, traditional validation methods often face obstacles such as time limitations, limited student involvement, and difficulties in collecting relevant feedback. This study explores a new approach that uses virtual student simulations based on large language models (LLMs), particularly GPT, to test learning materials automatically. Each virtual student is given a different level of access to the course content and asked to answer a quiz. The results show that students with more content access tend to perform better, but some with limited access still achieve high scores. This suggests that certain parts of the content are more influential than others, and that the model can sometimes infer correct answers from partial information. This approach explores the possibility of quickly testing and improving course materials without involving real learners. It can also help identify which resources are most important for understanding. In future work, we plan to explore how to simulate different student profiles and learning strategies to make the method even more realistic.