The use of AI for the characterization of fiber-reinforced soils (FRS) has dramatically improved geotechnical engineering by offering data-driven alternatives to traditional experimental methods. This review provides a comprehensive appraisal of AI methods, including machine learning (ML), deep learning (DL), and hybrid learning (HL) techniques, for the prediction of im-portant mechanical properties (unconfined compressive strength, shear strength, tensile strength, and deformation behavior) of FRS. In general, studies have shown that artificial neural networks (ANNs) and support vector machines (SVMs) tend to have better prediction capacities with coefficient of determination (R2) values of greater than 0.99, suggesting they provide a more valid representation of nonlinear soil-fiber interactions than classical empirical models. Hybrid models such as GA-ANN and WFLSSVR also demonstrated improved prediction accuracy due to optimized metaheuristic algorithms with observed mean absolute percentage errors (MAPE) of less than 3%. Important complexities in model prediction can arise through fiber type (natural vs. synthetic), fiber length, and fiber and soil properties. Challenges still remain with regard to data availability for natural fibers, interpretability of models, and transferability of results to field scale. Future developments could be boosted by implementing approaches such as physics-informed machine learning, explainable AI (XAI), and IoT-based monitoring of site factors to link numerical models with real-world observations. Overall, AI presents transformative potential for optimizing FRS design and deployment, supporting sustainable geotechnical solutions through the use of standardized datasets and hybrid experimental-computational frameworks.