A new method called Human-like Criteria Probing for Hallucination Detection (HCPD) has been developed to effectively identify inaccuracies in large language models without relying on internal data or external references. This approach mimics human evaluators' reasoning and improves the detection of factually incorrect content, which is crucial for the safe use of LLMs. The method has shown superior performance compared to existing solutions, potentially enhancing trust in AI-generated content. The availability of the code allows for further exploration and implementation in various applications.