Spontaneous preterm birth (sPTB) remains the leading cause of neonatal morbidity and mortality worldwide, yet its accurate prediction continues to challenge obstetric care. Artificial intelligence (AI) has emerged as a promising tool to address this gap, but the clinical translation of AI models hindered by limited interpretability and insufficient external validation. This systematic review evaluated explainable AI (XAI) approaches for predicting sPTB, focusing on model performance, clinical utility, and external validation. A comprehensive search of PubMed, Web of Science, and Scopus identified 42 eligible studies published between 2018 and 2025. Studies assessed using the PROBAST risk-of-bias tool, and performance metrics extracted for narrative synthesis. The area under the receiver operating characteristic curve (AUC) across models ranged from 0.65 to 0.93, with XGBoost and ensemble methods demonstrating superior discrimination. Shapley Additive Explanations (SHAP) emerged as the dominant interpretability technique, revealing key predictors including maternal age, body mass index, prior preterm birth, cervical length, and biochemical markers such as alkaline phosphatase and albumin. External validation performed in only 38% of studies, with AUCs typically declining by 0.05–0.12 from internal validation. Clinical utility assessment via decision curve analysis reported in 29% of studies. Substantial heterogeneity in outcome definitions, predictor sets, and validation strategies limited meta-analytic pooling. This review highlights the need for standardized reporting of XAI methods, prospective external validation, and integration of clinical utility metrics before these models reliably deployed in obstetric practice.