Short-term load forecasting (STLF) plays a key role in the operation, stability, and economic aspects of microgrids. Precise STLF helps in scheduling the generation of power, managing its storage, controlling demand response, and ensuring the reliability of distributed energy resources. Recent advancements in the field of artificial intelligence (AI) have revolutionized forecasting techniques, providing more accurate predictions even for complex and changing load profiles. This paper presents a review of the state-of-the-art AI techniques applied to short-term load forecasting in microgrids along with specific challenges. It covers a range of AI-based approaches—from conventional machine learning methods such as Artificial Neural Networks (ANNs) and Support Vector Machines (SVMs) to advanced deep learning architectures like Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNNs), as well as emerging hybrid solutions. However, integrating AI into STLF for microgrids introduces notable challenges, including data scarcity, high load variability, overfitting risks, limited model interpretability, and computational constraints in real-time settings. Addressing these limitations is essential not only to enhance forecasting accuracy but also to strengthen overall microgrid management. Accurate STLF directly impacts energy resource allocation, cost reduction, and system resiliency which are the key metrics that define microgrid performance. Ensuring high data quality, computational feasibility, and cybersecurity will play a key role for broad adoption of AI techniques. Finally, this paper discusses future research directions and the potential for AI-driven STLF to improve operational efficiency, adaptability, and resilience in ever-evolving microgrid environments.