Artificial Intelligence, Building Information Modeling, and Digital Twins in Smart Cities: A Systematic Bibliometric and Thematic Review of Research Trends and Integration Maturity

The methodology employed for this review paper on the integration of AI–BIM–DT in smart cities is a combination of Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)-based systematic review, bibliometric analysis, and thematic synthesis. This multifaceted method is used to ensure an organized and clear mechanism for the identification, screening, and inclusion of relevant studies. The literature search and selection process are based on the systematic review, which uses the PRISMA protocol. The next step would then be bibliometric analysis, which is used to quantitatively evaluate patterns of publication, citation networks, and trends of research in the field to get the information about how the research changes and what impacts it has over the years. Lastly, thematic synthesis will be applied to qualitatively analyze and synthesize the contents of a set of chosen studies to determine essential themes, concepts, and relationships. Such a combined approach enables a comprehensive understanding of the research advancement in AI, BIM, and DT technologies applied to smart cities, which encompasses both the quantitative and qualitative aspects of the existing literature.

As shown in Figure 2, the hybrid review workflow proceeds through four linked stages: a PRISMA-guided systematic search and screening of the literature, a bibliometric mapping of publication and citation patterns, a thematic synthesis of the screened studies, and the consolidation of findings into a conceptual framework. This staged workflow allows the quantitative mapping of the AI–BIM–DT research landscape (bibliometric analysis) to be interpreted alongside a qualitative understanding of how these technologies are applied (thematic analysis), with each stage building on the inclusion criteria established by the PRISMA process. However, as with any literature-based synthesis, the approach remains subject to data-integration, privacy, and standardization limitations that are discussed further in the Conclusion.

Figure 2. Workflow of hybrid systematic review and analysis for artificial intelligence (AI)–building information modeling (BIM)–digital twin (DT) integration

2.1 PRISMA systematic review

This review’s focus is on the integration of AI–BIM–DT in smart cities and its applicability, and on identifying how far the current research has achieved its goal, to draw a research agenda that can lead future research on the integration of AI–BIM–DT in smart cities. Thus, the search was performed on papers that addressed the AI–BIM–DT on a city scale in three major search engines, which are among the largest databases for multidisciplinary research: Scopus, Web of Science, and IEEE Xplore. Thus, the search strings were widened to include built environment, urban environment, and district. Using the topic filter in Web of Science, which leads to finding data in Title, Abstract, and Keywords, the search strings for Scopus, Web of Science, and IEEE Xplore were identical. Therefore, we utilized the same string that could generate as many outputs as possible that were related to our main topic. As a result, 342 published article outputs were generated. The search covered the publication period from 2020 to 2025 and was restricted to the Title, Abstract, and Keywords fields (or the closest equivalent field available in each database); the exact search-execution date is not restated here but will be confirmed in the search-methods documentation to be supplied during revision.

2.1.1 Inclusion and exclusion criteria

The inclusion criteria were: (i) peer-reviewed journal articles; (ii) published in English; (iii) reporting empirical studies, systematic reviews, or frameworks addressing the integration of AI, BIM, and DT for a smart city or urban digital context. The exclusion criteria were: (i) editorials, conference papers, and preprints; (ii) non-English-language publications; (iii) purely theoretical work without empirical or applied validation; (iv) studies that addressed the development of an AI–BIM–DT solution for a single building or a small infrastructure project in isolation, rather than at a smart city or urban digital scale. These criteria were applied consistently across Scopus, Web of Science, and IEEE Xplore. The systematic literature search was conducted through the University's electronic library/discovery platform, covering the period 2020–2025. The following Boolean search string was consistently applied across the databases as shown in Table 1, with minor syntax adjustments where required by the respective database: (("Digital Twin" OR "Digital Twins") AND ("BIM" OR "Building Information Modeling") AND ("AI" OR "Artificial Intelligence" OR "Machine Learning" OR "Deep Learning") AND ("Smart City" OR "Smart Cities" OR "Urban Digital")). Searches were performed within the Title, Abstract, and Keywords fields, or the nearest equivalent fields available in each database. The filtering functions available within the university library platform were subsequently applied to refine the search results according to publication year, subject area, document type, language, source type, and publication stage, where applicable. Records retrieved from the three databases were consolidated, and duplicate publications arising from overlapping database coverage were identified by comparing bibliographic information and removed manually, with only one instance of each publication retained; this removed 184 of the 342 records, leaving 158 unique records. Following deduplication, a two-stage manual screening procedure was performed: title/abstract screening against the eligibility criteria, followed by full-text examination of potentially relevant articles. The title/abstract and full-text screening was conducted by one reviewer using the predefined eligibility criteria; consequently, procedures for resolving disagreements between multiple reviewers were not applicable, and this single-reviewer design is acknowledged as a limitation of the review process. During title/abstract screening, 110 of the 158 records were excluded: 58 were conference papers, editorials, or preprints; 29 focused exclusively on a single building or small infrastructure project; 18 were purely theoretical studies without validation; and 5 were non-English-language publications. This left 48 articles for full-text assessment, during which 4 further studies were excluded: 2 were purely theoretical studies without validation, and 2 were limited to single-building or small-scale infrastructure applications. This review includes papers that integrate all three technologies (AI, BIM, and DT) within the context of smart cities, while excluding those that address only two of the three technologies (e.g., AI with BIM, or BIM with DT). Ultimately, 44 studies satisfied the eligibility requirements and were included in the final qualitative and quantitative synthesis, as shown in Figure 3.

Table 1. Data collection and search keywords

Figure 3. Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram illustrating identification, screening, eligibility, and inclusion stages

Table 2 synthesizes 44 studies drawn from the reviewed corpus and classifies each according to the depth of AI–BIM–DT data integration it reports, using a four-level integration tier taxonomy: (i) conceptual discussion, where AI, BIM, or DT concepts are discussed narratively without an operational data pipeline (ii) one-way data flow, where a single data source (e.g., a BIM model, point cloud, or geographic information system (GIS) layer) feeds a downstream process without feedback (iii) bidirectional synchronization, where the physical asset and its digital representation exchange data continuously in both directions and (iv) full twin with AI-driven feedback, the most advanced tier, in which an AI/ML layer closes the loop by generating predictive or prescriptive actions back onto the physical system.

In terms of primary data sources, the corpus is dominated by literature-corpus and bibliometric/citation-record sources (consistent with the high share of conceptual-discussion papers), followed by a smaller but technically important cluster of sensor- and IoT-derived sources (building energy, air quality, indoor temperature, and hydrological sensors) and geospatial sources (GIS, remote sensing, and point-cloud/UAV data) that underpin the one-way and bidirectional tiers. This pattern suggests a maturity gradient in the field: geospatial and BIM-based inputs are comparatively well established as one-way feeds into DTs, whereas real-time, bidirectionally synchronized sensor networks and especially AI-driven feedback loops remain comparatively nascent, pointing to a clear direction for future empirical work in this review.

2.2 Bibliometric analysis

In this quantitative bibliometric analysis, 44 articles have been selected for analysis, as illustrated in Table 1, and were chosen based on the PRISMA inclusion criteria after the filtering process. In this analysis, the high citation publication trend, the top ten authors, and the highly cited journals have been explained. With AI–BIM–DT technology for smart cities, the year-wise publication trend graph of articles from 2020 to 2025 has also been presented. Within the 2020–2025 search window and inclusion criteria applied here, no articles meeting the full AI–BIM–DT integration criterion were retrieved before 2020; earlier conceptual work relevant to individual components of this convergence exists but falls outside the defined search period and inclusion criteria, so its absence from the 44-paper sample reflects the review’s scope rather than an absence of prior research. After 2020, several papers were published, though with more limitations and research gaps, which are addressed in this review.

The bibliometric analysis was performed using VOSviewer (version 1.6.20). Bibliographic records exported from the combined Scopus, Web of Science, and IEEE Xplore search results were used as the citation database. Author keywords were selected as the unit of analysis for a co-occurrence analysis, applying a minimum occurrence threshold of 3, full counting, and association-strength normalization; the clustering resolution was left at the VOSviewer default (1.00), and no thesaurus file was applied to merge synonymous or variant terms. Figure 4 presents the resulting keyword co-occurrence network, showing four thematic clusters (digital technologies; smart city/urban planning; sustainability/energy; and computational methods). The interpretation of these clusters, and of the citation and authorship patterns in Figure 5, is presented in Section 3 (Bibliometric Findings) rather than repeated here, to keep this subsection focused on methods.

An interconnected research landscape encompassing smart cities, digital technologies, and sustainable urban development has been presented through the bibliometric analysis. Four major thematic clusters have been identified in the VOSviewer visualization, each represented by distinct colors as shown in Figure 4. The blue cluster is identified as the largest and most central, based on research concerning digital technologies like DT, IoT, 3D modeling, and BIM, which have been highlighted as afforded by its contribution to data-driven urban systems. Smart cities, urban planning, and GIS have been related to the green cluster; practical applications and strategic development have been mentioned. Sustainability, energy conservation, life cycle management, and decision-making have been targeted in the red cluster based on long-term and environmental planning. The yellow cluster has been associated with new ways of computing and analysis, implying that they are becoming more and more used in the urban development processes. Close connections between these clusters, especially those related to DT, smart city, and BIM, have been witnessed, and these connections show the interdisciplinary character of this research field. The relative node sizes have also denoted the core role of DT, smart city, and BIM, representing a change to more integrated and technology-focused solutions to the challenges of complex city life.

Figure 4. Keyword co-occurrence network map showing thematic clusters

Figure 5. Bibliometric publication trend graphs

Figures 5(a)–(c) summarize the publication/citation trend, the leading journals, and the most-cited contributing authors identified in the bibliometric analysis, using the same VOSviewer export (Scopus + Web of Science + IEEE Xplore combined) described above. These results are interpreted in Section 3 rather than repeated here. As Figure 5(b) indicates, sustainability, energy efficiency, and spatial information systems are prominent topics of interest in this research community.

Figure 5 presents a comprehensive bibliometric analysis highlighting the relationship between publication output and citation impact within the research domain. The most used publishing platforms are MDPI and the publishing companies of (MDPI AG and MDPI AG) Indexing, which demonstrates its great contribution and citation influence in this sphere. A similar pattern can be observed with other publishers like Elsevier B. V., Tongji University, and Springer Nature, which also have moderate citation influence, with the highest citation numbers obtained by Buildings, Energies, and the ISPRS International Journal of Geo-Information, as illustrated in Figure 5(a). As Figure 5(b) indicates, sustainability, energy efficiency, and spatial information systems are prominent topics of interest in this research community, with the two most notable contributors being M.E. Sepasgozar and M.T. Shahzad, who will most likely contribute significantly to the field and be cited frequently in other related sources. All in all, these analyses help to understand that the research in this area is mainly focused on open-access platforms due to several highly productive authors and journals that determine the existing picture of knowledge, as Figure 5(c) demonstrates.

2.3 Thematic analysis

The thematic analysis followed an inductive coding approach in which recurring concepts, applications, and challenges across the 44 included studies were identified through repeated reading of the full texts, grouped into provisional codes, and then consolidated into broader analytic categories. Some studies contributed evidence to more than one theme where their content spanned multiple domains. A theme-definition table (giving the inclusion rule, representative studies, and number of papers per theme) and the inter-coder verification procedure will be provided as supplementary material in the next revision. This process identified five main themes: data-driven decision-making and real-time monitoring, sustainable urban development and energy efficiency, interoperability and integration of technologies, smart infrastructure management and maintenance, and visualization and simulation of urban planning illustrated in Figure 6. The themes assist in knowing the actual and expected possibilities and use of AI–BIM–DT in smart cities, which in turn can be incorporated in a research agenda that could further contribute to the benefit of smart cities.

Figure 6. Key thematic relationships enabling integrated smart city development

The integration of AI, BIM, and DT technologies within smart cities is a rapidly evolving field, characterized by several interconnected themes that highlight its potential and ongoing challenges, as shown in Table 3. The thematic synthesis of the literature in detail demonstrates that there are five major areas: Data-driven decision making and real-time monitoring, sustainable urban development and energy efficiency, interoperability and integration of technologies, smart infrastructure management and maintenance, and visualization and simulation of urban planning [44, 45]. All these themes collectively lead to a sound research agenda that will add value to such convergent technologies in urban settings [46]. Data-Driven Decision Making and Real-Time Monitoring. This theme is seminal and, as such, the use of AI, BIM, and DT technology allows advanced data collection, processing, and analysis to underpin urban management and decision-making. DTs are dynamic and virtual models of real-life objects or systems, constantly recording more information from a variety of sensors, such as environmental conditions such as temperature and weather and operational indicators such as energy production [47]. This real-time information, which is frequently provided by IoT devices, is combined with static BIM models that present a very rich semantic and geometrical background. The large and heterogeneous data will be essential to process using AI algorithms, extracting insights and predicting future states, as well as finding anomalies or areas that could be improved [48]. As an example, artificial intelligence-based DTs may be used to observe urban infrastructure and predict the need to maintain it, as well as to streamline operational parameters, thus helping urban planners and managers make proactive decisions [49].

This is particularly crucial in overcoming the mayhem of rapidly expanding urban ecosystems. Future studies in this field ought to emphasize the creation of more advanced AI models that are able to process high-velocity, high-volume data streams of various urban sensors, as well as the development of robust frameworks for turning these insights into actionable urban policies and interventions [50]. BIM is a systematic data storage of building materials and their performance traits. Together with the real-time operation data of DTs and processed using AI algorithms, this data may streamline the performance of buildings and urban systems [51].

The AI models are proficient at interpreting and forecasting energy consumption trends, forecasting changes in demand, and proposing energy-saving options. They may include optimization of heating, ventilation, and air conditioning (HVAC) systems and lighting in accordance with occupancy and external conditions, or more advanced urban energy management systems [55]. Besides buildings, DTs, especially when utilized in relation to critical infrastructure, can enhance functionality and performance, resulting in more sustainable urban spaces and higher resistance to environmental-related issues. These digital technologies are also useful to the broader understanding of a circular economy in construction since the open-BIM and DTs allow integrating sustainability concepts in the form of tools such as material passports, which serve to advance the goals of countries to mitigate environmental impact and enhance energy efficiency [56].

Research directions include the following: This theme deals with the longstanding challenges and ongoing research to be able to facilitate smooth communication and data exchange between AI, BIM, and DTs. Data fragmentation and lack of standardized data exchange protocols traditionally impact the AEC industry adversely and deter any information flow [57]. DTs can be seen as a strategic attempt to bridge these gaps because the convergence of BIM and Industry 4.0 innovations like AI, IoT, big data, or extended reality (XR) has been attempted [45]. Nonetheless, the emergence of present-day DT applications in building operations supports the idea that the challenge of multi-source data integration, technological interoperability, and visualization needs to be addressed. The focus of future work is on building universal data standards and interoperability systems, including the digital twin definition language (DTDL), required to support compositional and modular analysis of complex cyber-physical systems [58].

This also implies that studies of the common data environment and semantic web technologies are required to make sure that data provided by different sources can be effectively connected and interpreted by AI systems [59]. Smart infrastructure management and maintenance. The fourth theme is concerned with the critical role of AI–BIM–DT integration in developing the field of managing and servicing smart city infrastructure. DTs offer a moving, dynamic, real-time reflection of physical assets, which allows constant monitoring and predictive analysis of their state [60]. Using sensor data provided by IoT devices, AI algorithms can identify abnormal conditions, predict possible failures, and plan preventative maintenance, thus making urban infrastructure using a variety of sensors much more resilient and efficient. It is of great importance, especially in the management of various kinds of assets, including stormwater infrastructure systems, bridges, roads, and public buildings, where AI and DT will be able to streamline performance, increase the lifespan of assets, and avoid expensive disruptions [61].

An example is the work of AI models in automated progress monitoring of construction using vision and bridge management, increasing the accuracy and safety of infrastructure development projects. Future research in this field should consider predictive maintenance models of the future, robotic inspection with the help of DTs, and AI-based decision support systems for infrastructure resilience planning. Visualization and Simulation of Urban Planning. The last theme is the ability of integrated AI–BIM–DT to provide three-dimensional and interactive visualization and simulation environments required in the contemporary urban planning process [46]. BIM models form the basis of geometric and semantic information of urban assets, which is supplemented with real-time operational information of DTs.

AI capabilities further enhance simulation aspects, enabling complex scenario analysis, predictive modeling of urban phenomena (e.g., traffic flow, pedestrian movement, environmental impacts), and optimization of urban designs [62]. Such visualizations can be turned into immersive experiences with the help of technologies such as XR, which enables urban planners, stakeholders, and even citizens to engage with DTs, compare alternative urban designs, and predict the effects of the suggested developments [63]. This interactive 3D environment leads to better communication, collaborative planning, and, finally, better and more sustainable strategies of urban planning [64]. The five themes discussed above would then point to the future research of urban DTs that should incorporate advanced human-computer interaction, the incorporation of generative AI in prediction of urban design alternatives, and the creation of a simulation model that can easily predict changes in social and economic aspects of urban interventions. They also define areas of critical focus in future research and technological development, pushing the limits of urban innovation as shown in Figure 9. This is presented as a conceptual arrangement of the technological pillars identified in this review rather than a fully dynamic framework; a redesigned version illustrating information flow, model updating, and feedback into the physical system, using bridge predictive maintenance as a concrete illustrative example, is provided below as Figure 10.

Figure 9. Key technological pillars shaping future smart infrastructure systems