Enhancing the learnability and reliability of machine learning algorithms remains a critical research objective. As a learnable framework, Conformal Prediction (CP) leverages historical data to generate statistically valid predictions for new instances at a specified confidence level. For any predictive model, CP produces prediction sets or regions that are guaranteed to contain the true label with at least the prespecified confidence level under the assumption of exchangeability. This review highlights pioneering contributions from Chinese research communities to both theoretical advancements and practical applications of CP. Based on a systematic review of the existing literature, we analyse CP’s core theoretical strengths and objectively discuss its inherent limitations, such as computational inefficiency and the curse of dimensionality. Furthermore, we evaluate CP’s current potential across a range of real-world domains, including biomedicine, cybersecurity, industry, and environmental science. Notably, novel applications such as tea analysis and the identification of traditional Chinese medicine using electronic noses demonstrate CP’s versatility and significant practical value.