This study presents systematic design guidelines for AI-powered clinical decision support systems (AI-CDSS) based on a comprehensive literature review and theme analysis. The proposed guidelines are divided into two parts, comprising 25 items: "How-design methods" and "What-design content and forms." The usability of these guidelines is demonstrated through an AI-CDSS design practice for stroke diagnosis and thrombolytic risk assessment in a Chinese clinical setting. The empirical case study provides practical suggestions for similar AI-CDSS designs that require fast decision-making. Challenges concerning system transparency or explainability and the impact of bias on system output are considered key issues in designing an effective and satisfying AI-CDSS. The potential usefulness of AI-CDSS in clinical decision-making scenarios has been indicated by numerous studies, with efficient design guidelines still needed to address these challenges.