While academics, government agencies, and industry practitioners have often proposed and relied on correction messages to reduce the spread of misinformation, recent studies indicate they have been ineffective, often amplifying the false message. The current research identifies the major factors responsible for the in/effectiveness of corrections as they interact with fake news on social media using a unique dataset obtained during two shock events. Our study revealed content- and linguistic-related features responsible for this phenomenon. We find that, for example, the more facts posted in a tweet, the less effective the correction tweet is. However, the more words and the more similar a tweet is to an original falsehood, the more effective it is, suggesting information asymmetry influences misinformation diffusion. Furthermore, we find that the probability of being effective was higher for positively charged correction tweets than negatively charged correction posts. These findings have several implications for research and practice.