Artificial intelligence (AI) can enhance human decision-making by providing assistance at different levels of abstraction. This study investigates whether AI should offer broad, high-level guidance (global AI) or focused, low-level assistance (local AI) to optimise performance and learning. Using a hierarchical multi-armed bandit task where both AI types provide equally valuable recommendations, we evaluate how participants leverage AI support in making sequential decisions. Findings reveal that while participants benefited from both types of AI suggestions, global AI led to significantly greater performance improvements. These results contribute to our understanding of human-AI interaction in hierarchical problem-solving, highlighting the importance of designing AI systems that effectively support human cognitive processes.