The growing adoption of artificial intelligence (AI) in buildings has created new opportunities to improve energy efficiency, intelligent control, and carbon emissions reduction. However, the existing literature has largely focused on algorithm development and predictive accuracy, while the relationship between algorithmic performance, system integration, and practical deployment in low-carbon buildings remains insufficiently addressed in a systematic manner. This study aims to identify research trends, application domains, prevailing methods, and existing research gaps, and to develop an integrated framework for the widespread application of AI in low-carbon buildings. Accordingly, 128 studies published between January 2015 and June 2025 were examined by combining systematic review, bibliometric analysis, content analysis, and critical analysis. Quantitative findings showed that energy prediction, with 42 studies (32.8%) , and optimization and control, with 36 studies (28.1%) , were the dominant application domains. Machine learning, with 48 studies (37.5%), accounted for the largest share among the identified method families. Regarding validation approaches, 47 studies (36.7%) were primarily simulation-based, while 35 studies (27.3%) relied on real building data. Qualitative findings indicated that data quality, model generalizability, system integration, implementation feasibility, security and privacy, and economic considerations were among the major challenges. The novelty of this study lies in integrating bibliometric, content-based, and critical evidence into a five-layer integrated framework linking building infrastructure, data, AI, intelligent decision-making, and performance evaluation. Furthermore, preliminary expert-based evaluation indicated the conceptual acceptability and comprehensiveness of the proposed framework; however, its effectiveness and practical implementability at scale require field validation and long-term operational studies.