Fake news detection, which is typically based on text classification of news content, is crucial for the accurate dissemination of information on the Internet. Previous methods have demonstrated that incorporating auxiliary information, such as user comments, can enhance detection performance. In this paper, we argue that related content retrieved from the Internet is also a considerable source of information, which aids in verifying the authenticity of news. Therefore, we introduce an adaptive multi-source information integration framework that selectively incorporates news content, user comments, and retrieval information, which is generated by Retrieval-based Knowledge Consistency Validation (R-KCV). Specifically, we begin by searching the Internet for richer external information and refining the search results using a knowledge-based information collection strategy to reduce redundancy. Next, we introduce an inconsistency critique generation strategy and a self-check mechanism, which reorganize the retrieved information to generate critically analyzed supplementary insights. Finally, we propose an adaptive attention mechanism to dynamically balance the importance of information from different sources, enabling the generation of a fused multi-source detection result. Experiments on three representative datasets demonstrate the effectiveness of our model.