This research investigates the use of various text similarity methods in automating the recognition of varied contract templates. Determining the correct template is a crucial step before the automation process proceeds to the clause-by-clause evaluation stage. This recognition process involves dynamically comparing clause text between drafts and templates without data labeling, relying on available text. Testing was conducted using traditional methods (Jaccard similarity, TF-IDF, BM25) and natural language processing methods (BERT, LaBSE, LLM). The research methodology involves acquiring contract samples from various sources, creating templates, and testing template recognition. The testing output is evaluated based on its effectiveness in capturing semantic equivalence and contextual understanding. Research results show that LLM is highly robust in recognizing templates by only learning from the first few sample clauses. These findings indicate that template recognition automation through LLM will provide the best precision and accuracy compared to traditional methods and other natural language processing methods. Thus, this research can serve as a foundation for developing a template-based contract review automation system that is more robust against contract variations.