Explainable artificial intelligence (XAI) has emerged as a crucial requirement for trustworthy machine learning in drug discovery, particularly as predictive models have evolved from descriptor-based classifiers to graph neural networks and transformer architectures, increasing the demand for interpretable molecular predictions. Despite this growth, a quantitative overview of the field’s research structure, collaboration patterns, and thematic evolution has been lacking. This bibliometric review examines the XAI literature in drug discovery from 2017 to 2026, aiming to map publication trends, influential authors, institutional networks, national contributions, and major research clusters, as well as to characterize the evolution of keywords and themes from general interpretability toward specific attribution methods, molecular applications, and regulatory considerations. Publications were retrieved from PubMed, Scopus, and Web of Science using search terms related to explainability, interpretability, SHAP, LIME, attention, molecular prediction, ADMET, and drug discovery. Bibliometric indicators were computed with the bibliometrix R package, while VOSviewer and CiteSpace were employed for network visualization and burst detection, including analyses of co-authorship, country collaboration, keyword co-occurrence, citation, and co-citation networks. The retrieved corpus revealed rapid growth after 2020, with the strongest expansion occurring between 2022 and 2026, and highlighted SHAP-based explanations, molecular property prediction, toxicity modeling, and graph neural network interpretation as the largest thematic areas. Smaller yet increasingly visible themes included regulatory acceptance, trust, clinical translation, and user-centered evaluation. While XAI in drug discovery is technically mature in areas such as post-hoc attribution for molecular property prediction and ADMET modeling, the field remains methodologically concentrated around a limited set of explanation techniques, and themes related to trust, regulatory readiness, and prospective validation remain underdeveloped. The most frequent indexing terms—explainable artificial intelligence, drug discovery, interpretability, SHAP, molecular property prediction, graph neural network, and ADMET—reflect both the methodological and application-focused structure of the literature and illustrate a temporal shift from broad interpretability concepts to specific algorithmic and translational applications.