Most existing explainable fake news detection systems rely on fact-checking reports to generate truth explanations. However, the low efficiency of verification leads to debunking delays. To address this, researchers have explored using crowd wisdom for detection by extracting supporting and opposing arguments and selecting the higher-quality side as evidence. Yet, crowd wisdom may contain misleading or biased information due to its unverified nature. This paper proposes CWCD, a model based on competitive wisdom and conflict debate, to handle biases in crowd wisdom. The model first divides crowd opinions into supportive and opposing camps. By comparing both sides, conflicting facts are identified as potential misinformation. A multi-agent debate mechanism then adjudicates each conflict, selecting the correct side as valid evidence. This process helps eliminate biases and improve detection accuracy. Extensive experiments on two real-world benchmarks show that CWCD achieves state-of-the-art performance, especially in fine-grained detection capabilities.