As a result of the swift development of the power industry, the number of power-related documents has increased sharply. Analyzing these massive power documents manually is often time-consuming and laborious, with low efficiency. However, since power documents are written in human language, which is complex and polysemous, this poses challenges for the processing of large models. When large models handle them, whether the references to words and sentences in the documents are accurate becomes a key issue. This study aims to deeply explore the limitations of manual analysis, the potential of large model applications, and the influencing factors of the accuracy of reference processing by large models in the processing of power documents. Through the analysis of a large number of actual power documents and related case studies, it is found that although manual analysis can ensure a certain degree of accuracy, it is difficult to cope with the growth in the number of documents.