Artificial Intelligence (AI) has rapidly reshaped residential building energy research; however, the field lacks a data‐driven synthesis of its global structure and evolving themes. This study addresses this gap by combining bibliometric analysis with latent Dirichlet allocation (LDA) topic modelling to analyse 1368 articles (2020–2025) from Web of Science Core Collection (WoSCC) and Scopus. Results show a dramatic increase in publications within a global but uneven landscape: China and the United States (US) lead in volume, while Vietnam and Denmark produce the most influential outputs per paper. The United Kingdom (UK) maintains a balanced profile, and several EU member states, including Italy, Spain and Germany, rank among the leading contributors. Collaboration networks are primarily academic, with limited industry involvement. LDA identifies four main themes: energy performance modelling, HVAC optimisation, load forecasting and management and smart‐home energy management. Within these, research centres on operational applications and shows a methodological progression from conventional machine learning (ML) toward hybrid deep learning (DL), including emerging transformer‐based and graph‐based models. The analysis scale is expanding from single‐buildings to district‐level and interconnected systems. The comprehensive review identifies a persistent gap in occupant‐centric modelling and highlights the need for real‐time data integration and improved transparency to deliver practical, human‐responsive, and scalable energy efficiency solutions.