Schedule network diagrams serve as an important basis for detailed project planning, progress control, and schedule-related dispute resolution. However, current schedule network review practices still rely heavily on human expertise, while existing automated tools mainly focus on quantitative indicators and rule-based checks. Their capability remains limited in addressing non-quantitative issues such as construction logic, activity semantics, contractual consistency, and engineering reasonableness. This study aims to investigate the feasibility and limitations of applying Large Language Models (LLMs) to schedule review tasks. Through a literature review, this study examines existing schedule network review standards, automated review cases, and the challenges of applying LLMs in professional domains, with a focus on the recognition of schedule network elements and the assessment of scheduling logic. Excel data exported from Microsoft Project are used as model input to evaluate the basic capability of LLMs in interpreting schedule network fields, activity relationships, and logical reasonableness. In addition, a customized GPT in ChatGPT integrated with Retrieval-Augmented Generation (RAG) is adopted to enhance domain knowledge and improve the review process. The results indicate that, after incorporating RAG, the model can refer to review standards and project data, thereby improving its performance in professional judgment, data consistency checking, and explanation of review results. Overall, although LLMs are not yet capable of replacing scheduling software or human review, they demonstrate potential in supporting activity semantic interpretation, standardizing review procedures, and integrating information across multiple documents.