In contemporary education, accurately predicting student performance and delivering prompt feedback is paramount for fostering a comprehensive grasp of academic progress and adopting strategies for enhancing the quality of student learning. This review examines the studies on virtual tutoring systems utilising Generative Pre-trained Transformer 3 (GPT-3) to enhance student performance and feedback. It explored machine learning integration in Virtual Learning Environments (VLE), analysing datasets and associated challenges. A systematic review was conducted using a keyword-based search strategy, with terms such as "virtual learning environment," "e-learning," "virtual tutor," "online exams," and "feedback systems." Articles from IEEE, Science Direct, Springer, Wiley, Google Scholar, and SCOPUS, published between 2017 and 2024, were considered. Of the identified studies, 130 were excluded for focusing solely on student performance evaluation outside the e-learning context, leaving six relevant articles for data extraction. This study reviews the effectiveness of GPT-3-based virtual tutoring's effectiveness compared to traditional methods, highlighting their advantages and limitations. The findings underscore GPT-3's potential for personalising learning experiences, improving student outcomes, and advancing educational technology. This research provides insights that support the development of data-driven virtual tutoring systems and empower educators with actionable feedback to promote student success.