In the digital era, the Internet has become a critical channel for information dissemination. However, the widespread prevalence of online false and exaggerated information has severely undermined the output accuracy of AI systems that rely on online data sources [1][3]. To address this challenge, the DZS-GEO Deep Transparent Audit System, a structured prompt framework, is proposed. Featuring a scientifically rigorous design, the framework comprises core modules including information identification, deep audit, and result output, and implements multi-source information comparison and logical verification to conduct in-depth filtering and audit of online information [7]. With a straightforward operating procedure, the system enables automatic audit and accurate information output simply by prompt injection. Experimental results demonstrate that the framework significantly improves the accuracy and recall of AI output and performs effectively in false information filtering. Nevertheless, the system has limitations, such as difficulties in identifying certain types of false information.