Despite their impressive generative capabilities, Large Language Models (LLMs) are often prone to hallucination—generating factually incorrect information—which poses a significant challenge to ensuring the reliability of their outputs. This study re-evaluates the potential of LLMs as tools for fact verification by leveraging their inherent reasoning capabilities. The proposed method combines prompting techniques and exploits reputable reference data sources from Wikipedia. Specifically, for each statement that needs to be verified, the system will retrieve relevant evidence from Wikipedia, then build a prompt that combines the original statement with evidence to request the model to evaluate the authenticity. Experimental results on the GPT-3.5, GPT-4o-Mini and T5 Instruct models show the clear effectiveness of the method, which is evaluated based on indicators such as Expected Calibration Error, Accuracy, Area Under Rejection curve and Factuality Score. In particular, GPT-4o-Mini is significantly superior in its ability to verify the correctness of information as well as ensure reliability for this model. These results confirm that, although LLMs are susceptible to hallucinations when left unchecked, when provided with adequate evidence and appropriate guidance, they can be effective and reliable information verification tools.