This study examines the integration of generative and predictive artificial intelligence (AI) models within smart cities, focusing on how user readiness and technology adoption influence their contribution to sustainable urban development and governance.
The study applies a systematic literature review following PRISMA guidelines and synthesizes evidence from 50 peer-reviewed studies (2018–2025) indexed in Scopus and Web of Science. It combines bibliometric mapping using VOSviewer with thematic analysis to examine the drivers, barriers and governance mechanisms shaping the adoption of generative, predictive and hybrid applications in urban contexts.
Generative AI fosters participatory engagement, citizen co-design and interactive simulations, advancing SDG 11 (Sustainable Cities and Communities) and SDG 4 (Quality Education) through enhanced digital literacy and inclusive planning. Predictive AI improves operational efficiency, forecasting accuracy and data-driven policymaking, supporting SDG 9 (Industry, Innovation and Infrastructure) and SDG 13 (Climate Action) by promoting sustainable resource use and climate-resilient management. Hybrid AI integrates these strengths, addressing both social and operational aspects of smart city development and aligning with SDG 17 (Partnerships for the Goals) through cross-sector collaboration and shared governance. Collectively, these models contribute to broader sustainability goals, including SDGs 3, 7 and 12.
This review acknowledges several key limitations. Reliance on Scopus and Web of Science may exclude regionally significant or domain-specific studies not indexed in these databases. The focus on English-language publications introduces potential language bias, possibly overlooking relevant research from non-English-speaking regions. Restricting the timeframe to 2018–2025 captures recent developments but may omit earlier foundational work or the most recent studies not yet indexed. Differences in research design, policy contexts and sample characteristics also affect comparability and limit generalizability. Future research should broaden data sources, include multilingual literature and adopt mixed-methods and longitudinal approaches to enhance contextual diversity and empirical robustness.
The findings provide practical guidance for policymakers, urban planners and technology developers to design AI governance systems that are transparent, accountable and aligned with the SDGs. Integrating generative and predictive AI can enhance operational efficiency, support participatory planning and promote responsible decision-making. The findings inform the development of adaptive policy frameworks that advance SDG 9 (Industry, Innovation and Infrastructure), SDG 11 (Sustainable Cities and Communities) and SDG 13 (Climate Action) through digital literacy initiatives, cross-sector collaboration and data-informed management. Strengthening these practices enables cities to translate AI’s potential into tangible contributions to inclusive and sustainable urban transformation.
Integrating user readiness and digital literacy into AI adoption is essential for building inclusive and trustworthy smart cities. These efforts support SDG 4 (Quality Education), SDG 10 (Reduced Inequalities) and SDG 16 (Peace, Justice and Strong Institutions). Generative AI encourages citizen participation and collaborative planning, while predictive AI improves service accessibility and data-informed governance. Promoting ethical awareness and community engagement helps narrow digital divides and address bias. Collectively, these elements advance SDG 11 (Sustainable Cities and Communities) and SDG 17 (Partnerships for the Goals) by fostering socially responsive and transparent AI-driven urban development.
This review is among the first to integrate perspectives on user readiness and technology adoption with comparative insights into generative and predictive AI in smart cities. It advances understanding of how AI-driven urban innovation supports inclusivity, efficiency and sustainability, while outlining policy directions and a future research agenda for equitable and transparent AI governance.
1. Introduction
AI is an integral part of today's smart cities and will remain an important component in the future. This has made artificial intelligence (AI) the heart of the smart city transformation (Lifelo, Ding, Ning, Qurat-Ul-Ain, & Dhelim, 2024). This technology is changing the way cities are managed, infrastructure is developed, and citizens interact with their environment. Two of the most prominent applications of AI are generative models and predictive models (Wolniak & Stecuła, 2024). Generative models can create adaptive urban design solutions, such as automatically generating city layout sketches based on residents' preferences. Meanwhile, predictive models help project infrastructure needs, map mobility patterns, and estimate environmental risks, such as predicting traffic density in Singapore or flood potential in Jakarta (Lifelo et al., 2024). By leveraging large-scale, real-time data, AI-driven cities enhance service efficiency, strengthen resilience, and promote evidence-based policy, thereby contributing to several SDGs, particularly SDG 7, SDG 9, SDG 11, and SDG 13.
User readiness encompasses technical ability, willingness to use AI-based services, and level of trust in the system being used. In the literature, this concept is often explained through frameworks such as the Technology Acceptance Model (TAM) and the Unified Theory of Acceptance and Use of Technology (UTAUT), which highlight the importance of perceived usefulness, ease of use, social influence, and infrastructure support in driving technology adoption (Kelly, Kaye, & Oviedo-Trespalacios, 2023). In the context of smart cities, citizens play a dual role. They are not only recipients of services but also providers of data that shape the output of AI systems.
The integration of predictive analytics and generative artificial intelligence in urban governance presents both substantial opportunities and complex challenges. Predictive models enable policymakers to anticipate risks and allocate resources more effectively, whereas generative models facilitate participatory urban planning by allowing the simulation of multiple development scenarios (Razavi et al., 2024). However, issues such as algorithmic bias, data privacy, transparency, and equitable access can hinder public acceptance (Sanchez, Brenman, & Ye, 2025). Negative perceptions of these aspects will impact public trust and, ultimately, affect the technology's adoption rate.
Previous studies have discussed technology adoption in smart cities and the application of generative and predictive AI separately (Wang, Zhao, Gangadhari, & Li, 2021; Habbal, Ali, & Abuzaraida, 2024). However, a comprehensive synthesis that combines user readiness and technology adoption with an analysis of the implications of these two types of AI models is still rare. Addressing this research gap is crucial for realizing inclusive and sustainable AI ecosystems aligned with SDG 9 (Industry, Innovation, and Infrastructure), SDG 11 (Sustainable Cities and Communities), and SDG 17 (Partnerships for the Goals). The surge in generative technology heightens the urgency of this study to leverage rapid innovation in real-time data-driven predictive analytics. It also underscores the increasing government commitments across nations to build inclusive and sustainable smart cities, in line with the United Nations Sustainable Development Goals (SDGs).
This systematic literature review aims to fill the gap. The study reviews empirical findings, theoretical frameworks, and implementation practices to identify key behavioral factors, adoption barriers, and governance considerations that can promote inclusive and sustainable AI integration in urban ecosystems. Based on these objectives, this research focuses on answering three main questions:
How does the application of generative and predictive AI models in the context of smart cities affect user readiness and technology adoption?
What behavioral, technical, and governance factors influence the successful adoption of these two types of AI models in urban environments?
What are the strategic implications of the differences in the application of generative and predictive AI for inclusive and sustainable smart city policies?
2. Method
With an emphasis on generative and predictive AI models, researchers employed a Systematic Literature Review (SLR) to identify, assess, and synthesize scientific research on user readiness and technology adoption in AI-driven smart cities. Employing this approach ensures a rigorous, transparent, and reproducible synthesis of the existing body of knowledge (Purssell & McCrae, 2024). Mapping current trends identifies recurring themes and highlights gaps in the literature.
2.1 Data sources and search strategy
The review draws from reputable academic databases to ensure high-quality, peer-reviewed studies. Scopus and Web of Science are selected for their multidisciplinary coverage, rigorous indexing, and advanced Boolean search capabilities. Publications from 2018–2025 are included to capture the maturity of AI, IoT, edge computing, and 5G in smart city contexts (Mahomed & Saha, 2025). For transparency, Table S2 in the supplementary material provides the complete Boolean search strings. Two independent reviewers conduct study selection, applying arbitration when necessary. Quality appraisal uses CASP and STROBE checklists to ensure consistency, minimize bias, and strengthen methodological rigor.
2.2 Inclusion and exclusion criteria
The study systematically applied established inclusion and exclusion criteria (see Table S1 in supplementary material) to identify and select relevant literature. It incorporated studies that explicitly examined user readiness and technology adoption within the context of artificial intelligence-enabled smart cities, particularly those utilizing generative or predictive models. The review encompassed empirical research employing quantitative, qualitative, or mixed-methods approaches, along with theoretical frameworks that address the behavioral, technical, and governance dimensions of AI adoption in smart city contexts. To maintain rigor and reliability, the review focused on peer-reviewed journal articles, conference proceedings, and book chapters indexed in Scopus between 2018 and 2025. The review also included only studies published in English and available in full text for comprehensive analysis and data extraction. These selection criteria ensured that the final dataset comprised high-quality and methodologically robust studies. Concentrating on recent peer-reviewed publications enhanced the validity of the findings and captured the most current discourse on AI adoption in urban contexts. Furthermore, emphasizing studies that applied generative or predictive models allowed for a more nuanced understanding of how emerging AI technologies shape urban innovation, governance, and citizen engagement.
2.3 Study selection process
This study follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines proposed by Page et al. (2021) to ensure methodological transparency, reproducibility, and accuracy. The process begins by identifying potentially relevant studies retrieved from selected databases and continues with the removal of duplicate records. The study then screens titles and abstracts to assess the initial relevance of each work to the research scope, explicitly focusing on user readiness and technology adoption in AI-based smart cities using generative and/or predictive models. Articles that meet the initial screening criteria proceed to a full-text review to evaluate their alignment with the established inclusion and exclusion criteria. This stepwise screening process confirms 50 studies as eligible and includes them in the final synthesis. Figure 1, the PRISMA flow diagram, illustrates the overall selection workflow.
The flowchart is titled “Identification of studies via databases and registers”. The flowchart shows three vertical text boxes representing three stages, arranged in a vertical series on the left. From top to bottom, these are labeled: “Identification”, “Screening”, and “Included”. In the “Identification” stage, a text box reads “Records identified (n equals 974): Scopus (n equals 580); Web of Science (n equals 394)”. A right-pointing arrow leads to another box labeled “Records removed before screening: Duplicate records removed (n equals 30); Records marked as ineligible (n equals 234); Book chapter (n equals 111)”. A downward-pointing arrow leads from “Records identified (n equals 974)” to a text box labeled “Title and abstract screened (n equals 599)” in the “Screening” stage. A right-pointing arrow from this box leads to a text box labeled “Records excluded: Restricted access (n equals 125); non-accredited journal (n equals 87)” in the same stage. A downward arrow leads from “Title and abstract screened (n equals 599)” to a text box labeled “Reports sought for retrieval (n equals 387)” in the same stage. A right-pointing arrow from this box leads to “Reports not retrieved (n equals 191)” in the same stage. A downward-pointing arrow from “Reports sought for retrieval (n equals 387)” leads to a box labeled “Reports assessed for eligibility (n equals 50)” in the same stage. A right-pointing arrow from this text box leads to a detailed box labeled “Reports excluded”, which contains the following exclusion reasons: Review study (n equals 213); Other language (n equals 1); Thesis (n equals 5); Statements (n equals 40); Conference abstract (n equals 25); Editorial (n equals 9); Conceptual framework (n equals 44). A final downward arrow leads from “Reports assessed for eligibility (n equals 50)” to a text box reading “Studies included in review (n equals 46); Reports included in review (n equals 4); Total (n equals 50)” in the “Included” stage.The PRISMA flow diagram
The flowchart is titled “Identification of studies via databases and registers”. The flowchart shows three vertical text boxes representing three stages, arranged in a vertical series on the left. From top to bottom, these are labeled: “Identification”, “Screening”, and “Included”. In the “Identification” stage, a text box reads “Records identified (n equals 974): Scopus (n equals 580); Web of Science (n equals 394)”. A right-pointing arrow leads to another box labeled “Records removed before screening: Duplicate records removed (n equals 30); Records marked as ineligible (n equals 234); Book chapter (n equals 111)”. A downward-pointing arrow leads from “Records identified (n equals 974)” to a text box labeled “Title and abstract screened (n equals 599)” in the “Screening” stage. A right-pointing arrow from this box leads to a text box labeled “Records excluded: Restricted access (n equals 125); non-accredited journal (n equals 87)” in the same stage. A downward arrow leads from “Title and abstract screened (n equals 599)” to a text box labeled “Reports sought for retrieval (n equals 387)” in the same stage. A right-pointing arrow from this box leads to “Reports not retrieved (n equals 191)” in the same stage. A downward-pointing arrow from “Reports sought for retrieval (n equals 387)” leads to a box labeled “Reports assessed for eligibility (n equals 50)” in the same stage. A right-pointing arrow from this text box leads to a detailed box labeled “Reports excluded”, which contains the following exclusion reasons: Review study (n equals 213); Other language (n equals 1); Thesis (n equals 5); Statements (n equals 40); Conference abstract (n equals 25); Editorial (n equals 9); Conceptual framework (n equals 44). A final downward arrow leads from “Reports assessed for eligibility (n equals 50)” to a text box reading “Studies included in review (n equals 46); Reports included in review (n equals 4); Total (n equals 50)” in the “Included” stage.The PRISMA flow diagram
2.4 Data analysis and synthesis
This study used inductive, coding-based thematic analysis to identify patterns and themes emerging from the reviewed literature. Two independent researchers conducted the initial coding, compared results, and resolved discrepancies through discussion to ensure reliability. Similar codes cluster into behavioral, technical, and governance themes related to user readiness and AI adoption in smart cities. Microsoft Excel supports coding organization, while VOSviewer enables bibliometric mapping of keyword co-occurrences and clusters. Combining thematic and bibliometric analyses provides nuanced insights and a broad overview of dominant topics, emerging trends, and conceptual linkages in the adoption of generative and predictive AI in smart cities within policy and governance contexts.
3. Results
3.1 Bibliometric analysis result
The bibliometric analysis using VOSviewer identifies three primary keyword clusters (Figure 2). The first links AI adoption and machine learning with smart cities and sustainability, underscoring efforts to integrate advanced AI into urban development. The second focuses on predictive AI, digital twins, and analytics that enable data-driven planning and efficiency. The third connects general AI concepts with technology adoption, reflecting debates on readiness and acceptance. Taken collectively, these clusters highlight the convergence of technological innovation, sustainability, and user adoption in AI-driven smart city research.
The network map displays multiple clusters of nodes, each represented by circles with labels, connected by thin lines indicating relationships, with labels adjacent to the nodes. At the center, a large blue node labeled “artificial intelligence”, is directly connected to the nodes “technology adoption”, “a i”, “machine learning”, and “smart cities”. Above this cluster, a red node labeled “machine learning”, is connected to “a i adoption”, “smart city”, and “sustainability”. To the lower right, a green node labeled “smart cities”, is connected to the nodes “digital twins”, and “predictive analytics”.Keyword co-occurrence
The network map displays multiple clusters of nodes, each represented by circles with labels, connected by thin lines indicating relationships, with labels adjacent to the nodes. At the center, a large blue node labeled “artificial intelligence”, is directly connected to the nodes “technology adoption”, “a i”, “machine learning”, and “smart cities”. Above this cluster, a red node labeled “machine learning”, is connected to “a i adoption”, “smart city”, and “sustainability”. To the lower right, a green node labeled “smart cities”, is connected to the nodes “digital twins”, and “predictive analytics”.Keyword co-occurrence
3.2 AI studies in smart city
Table 1 shows that a few countries with strong digital infrastructure and active innovation agendas dominate research on AI in smart cities.
Geographical distribution of AI studies in smart cities
| Region | Number of studies | Key countries (study count) | Dominant AI type |
|---|---|---|---|
| Asia | 20 | China (6), Saudi Arabia (3), India (2) | Predictive (65%) |
| Europe | 15 | Poland (4), Portugal (3), Spain (2) | Hybrid (Generative/Predictive) |
| Australia | 5 | Sydney/Melbourne (3), Regional (2) | Generative (80%) |
| Africa | 4 | Ghana (1), South Africa (1), Tunisia (1) | Predictive (75%) |
| Americas | 3 | USA (2), Brazil (1) | Predictive (100%) |
| Global | 3 | Multi-country comparisons | Governance-focused |
As shown in Table 1, China leads in studies on traffic emission monitoring, AI-based governance such as City Brain, and social well-being assessments in major urban areas. Saudi Arabia contributes through research on participatory planning and environmental management, while in Europe, Poland advances human-centric AI adoption and Portugal develops sustainable logistics using electric vehicles. Serbia and Spain explore AI-enabled transport optimization, and Greece expands digital twin applications. This concentration of evidence in technologically advanced contexts highlights the need to extend research to underrepresented regions for globally inclusive insights into AI-driven smart city development.
3.3 Risk of bias identification
This systematic review revealed critical insights into the methodological strengths and limitations of 50 studies examining AI adoption in smart cities. The risk of bias assessment, conducted using CASP (for qualitative/mixed-methods studies) and STROBE (for quantitative/observational studies), highlighted three dominant issues:
Selection and Reporting Biases
A total of 65% of the studies (35 out of 50) employ non-representative sampling strategies, such as single-city case studies and sector-specific participant groups. Furthermore, the frequent reliance on self-reported data, including corporate sustainability reports (Wang et al., 2022), may introduce bias and overstate the influence of AI.
Limited generalizability
Eighty percent of the studies (40 out of 50) lack cross-regional validation. For instance, Skoropad et al. (2025) demonstrate AI-driven traffic models exclusively in Belgrade, while Huang, Bibri, and Keel (2025) confine the validation of generative AI frameworks to Lausanne. The findings highlight gaps, as research primarily focuses on high-income cities such as Lisbon (Ferreira dos Santos, de Matos, & Groznik, 2025) and overlooks low-resource settings.
Ethical and Transparency Gaps
Only 12% of the reviewed studies (6 out of 50) explicitly address ethical risks. For instance, Aseen and Al-Amarneh (2025) investigate algorithmic bias in the banking sector of the UAE and Qatar. Moreover, only 5% of the studies (3 out of 50) offer open datasets, such as those used by Wang et al. (2021), limiting research reproducibility and transparency.
3.4 Thematic analysis of reviewed study
The comparative analysis in Table 2 highlights fundamental differences between predictive and generative models. Predictive models demonstrate lower bias risk through real-world data validation, provide robust short-term decision-making support, and emphasize technology-centric metrics for user readiness. Generative models, by contrast, exhibit higher bias risk due to limited empirical testing. However, they excel in scenario planning and long-term forecasting while incorporating citizen engagement through participatory feedback loops in smart city governance.
Key advantages of predictive and generative AI models
| Aspect | Predictive models | Generative models |
|---|---|---|
| Bias risk | Low to moderate, supported by validation using real-world data | High, as most rely on conceptual or theoretical frameworks without extensive empirical testing |
| Adoption insights | Stronger relevance for short-term and operational decision-making | More effective for scenario planning and long-term strategic assessments |
| User readiness focus | Limited, emphasizing technology-centric metrics | Greater focus on citizen engagement through feedback loops |
The reviewed study highlights that predictive AI is the most widely applied across smart city domains, particularly in transportation, logistics, energy forecasting, and environmental monitoring. At the same time, generative AI increasingly drives innovative applications such as urban digital twins, construction, and participatory governance. Hybrid approaches appear in integrative contexts such as smart building frameworks and sustainability reporting. The analysis also emphasizes the central role of user and organizational readiness, which includes people, processes, and data in enabling adoption, while connecting AI applications to multiple SDGs, including clean energy (SDG 7), sustainable cities (SDG 11), and climate action (SDG 13). Viewed collectively, these findings demonstrate that predictive AI provides operational reliability, generative AI contributes social and strategic value, and hybrid models integrate the strengths of both. Strengthening governance frameworks and readiness strategies ensures the sustainability and inclusiveness of AI-driven smart cities. To further explore these dynamics, Table S3 in the supplementary material presents a detailed summary of the reviewed studies, highlighting how generative, predictive, and hybrid AI applications align across behavioral, technical, and governance dimensions. Building on this synthesis, Table 3 provides a structured comparison between generative and predictive AI, illustrating how each approach contributes distinct yet complementary value to smart city development. Overall, these insights establish the empirical basis for the following discussion on how AI models foster sustainable and inclusive smart city governance.
Generative vs predictive
| Factor | Generative AI | Predictive AI |
|---|---|---|
| Behavioral |
|
|
| Technical |
|
|
| Governance |
|
|
4. Discussion
Building on the empirical findings, this discussion interprets how generative, predictive, and hybrid AI models advance user readiness, governance, and sustainable smart city transformation in relation to the SGDs. Predictive AI contributes to SDG 9 (Industry, Innovation, and Infrastructure) and SDG 13 (Climate Action) by enhancing operational efficiency and environmental resilience, while generative AI supports SDG 11 (Sustainable Cities and Communities) and SDG 4 (Quality Education) through participatory planning and digital literacy. Hybrid AI aligns with SDG 17 (Partnerships for the Goals) by promoting cross-sector collaboration and inclusive governance. Collectively, these AI types perform complementary roles in advancing sustainable urban transformation.
Predictive AI optimizes energy grids, transport networks, and waste systems through real-time IoT data, achieving energy savings of 15–20% in smart buildings. Generative AI enables co-creation and scenario exploration, but it also poses risks, such as the misuse of deepfakes and the exclusion of low-literacy groups. Policymakers and designers can mitigate these risks through citizen juries and multimodal interfaces. Predictive AI raises privacy and surveillance concerns, particularly in facial recognition, and policymakers can mitigate these risks through transparency-by-design and accountable data governance. Tiered adoption strategies and hybrid governance models integrate the strengths of both approaches to balance inclusivity, efficiency, and equity.
The reviewed studies align strongly with multiple SDGs. SDG 11 (Sustainable Cities and Communities) is most prominent, covering urban planning, digital twins, transportation, and waste management. SDG 13 (Climate Action) focuses on emissions prediction and energy optimization, while SDG 7 (Affordable and Clean Energy) focuses on energy forecasting, electric vehicle integration, and smart buildings. SDG 9 (Industry, Innovation, and Infrastructure) features in IoT and edge computing, SDG 3 (Good Health and Well-being) in AI-enabled healthcare, and SDG 12 (Responsible Consumption and Production) in logistics and waste reduction. Overall, generative and predictive AI enhance technical efficiency while advancing sustainable development across multiple, reinforcing goals.
4.1 Effects of generative and predictive AI on user readiness in smart cities
Generative AI models in smart cities enhance user readiness by fostering participation, visualization, and scenario-based planning. These functions align with the Technology Acceptance Model (TAM) and the Unified Theory of Acceptance and Use of Technology (UTAUT), which emphasize perceived usefulness, ease of use, and social influence as predictors of readiness. Urban digital twin simulations (Ferreira dos Santos et al., 2025), gamified citizen engagement platforms, and co-design tools help residents visualize their contributions to urban development. By making AI outputs accessible, interactive, and visually intuitive, generative systems increase perceived control and trust, which are key drivers of behavioral readiness and sustained adoption (Davis, Bagozzi, & Warshaw, 1989; Van Huy, Nguyen, Vo-Thanh, Thinh, & Dung, 2024). This participatory process advances SDG 11 and SDG 4 by empowering digitally literate citizens to participate in data-driven urban governance.
Predictive AI reinforces readiness through efficiency, reliability, and short-term benefits. Applications in traffic management (Skoropad et al., 2025), environmental monitoring (AlSalehy & Bailey, 2025), energy forecasting, and logistics optimization (Ferreira & Esperança, 2025) demonstrate that accuracy and reliability strengthen user confidence in AI systems. Such performance-oriented outcomes align with Rogers' Diffusion of Innovations theory, where relative advantage and trialability drive adoption among early adopters and public managers. Predictive AI thus supports SDG 9 and SDG 13 by enabling efficient resource use and resilient urban operations.
Hybrid AI combines the analytical rigor of predictive systems with the participatory strengths of generative tools, enabling evidence-based policymaking that is transparent and inclusive (Xu, Cugurullo, Zhang, Gaio, & Zhang, 2024). These integrative models build effective, socially trusted socio-technical systems, supporting SDG 17 through cross-sectoral collaboration and co-governance in smart city ecosystems. In summary, generative AI fosters engagement, predictive AI drives performance, and hybrid approaches unite both to promote equitable and sustainable smart city transformation.
4.2 Behavioral, technical, and governance factors influencing AI adoption
The successful adoption of generative and predictive AI in smart cities is shaped by three interrelated dimensions: behavioral, technical, and governance factors. These dimensions influence how users perceive AI, how effectively systems operate in complex urban environments, and how institutions support integration.
Behavioral factor
AI adoption depends on perceived usefulness, ease of use, trust, and social influence. Citizens and public officials are more likely to adopt AI when they perceive tangible benefits such as reduced congestion, energy savings, or improved service efficiency (Rathnayake, Nguyen, & Ahn, 2025; Wang, Wang, & Liu, 2025). Generative AI fosters acceptance through intuitive and participatory interfaces, while predictive AI builds confidence through measurable outcomes. Transparent algorithmic decision-making enhances trust (Yaseen & Al-Amarneh, 2025), whereas concerns about bias or unequal service delivery undermine readiness (Frimpong, 2025). From the perspective of SDG 4, digital literacy and civic education programs play a crucial role in strengthening readiness and reducing fears of automation.
Technical factor
Adoption is driven by robust data infrastructure, model reliability, and strong cybersecurity. Within this context, predictive AI relies on high-quality, real-time data to generate accurate insights (Gkontzis, Kotsiantis, Feretzakis, & Verykios, 2024; Huang et al., 2025), whereas generative systems depend on diverse, representative datasets to ensure realistic scenario generation. Limitations such as legacy system incompatibility and high computational requirements have accelerated the adoption of edge computing for latency-sensitive applications, such as autonomous vehicles (Chaymae, Mhamed, & Soumia, 2025). Balancing accuracy with explainability remains essential for policy acceptance (AlSalehy & Bailey, 2025). Cybersecurity risks, including adversarial attacks or data manipulation, threaten both model types, prompting the integration of blockchain and secure data-sharing protocols to enhance resilience (Gondhalekar et al., 2025). Strengthening technical robustness contributes to SDG 9 by fostering resilient and innovative digital infrastructure.
Governance factor
Governance factors encompass ethical regulation, institutional capacity, and inter-organizational collaboration. Ethical frameworks and bias audits help reduce discriminatory outcomes and enhance legitimacy (Fatorachian, Kazemi, & Pawar, 2025), while GDPR-like privacy regulations are critical in surveillance, health, and mobility contexts (Wibowo et al., 2025). The lack of standardized protocols for interoperability and digital twin integration remains a barrier to large-scale deployment (Kalfas, Kalogiannidis, Spinthiropoulos, Chatzitheodoridis, & Ziouziou, 2025). Institutional silos, financial limitations, and low technical literacy among urban administrators further impede adoption (Varzeshi, Fien, & Irajifar, 2025). Therefore, governance reforms that promote transparency, cross-sector collaboration, and capacity-building are vital for achieving SDG 16 and SDG 17.
4.3 Strategic implications for inclusive and sustainable smart city policies
Generative and predictive AI each offers unique advantages for inclusive and sustainable smart city governance but also poses distinct risks that require adaptive policies. Generative AI fosters participatory governance by integrating citizen input into urban planning simulations, co-creating public policies, and supporting scenario-based education. However, unequal access to digital resources and literacy can reinforce exclusion. Policies aligned with SDG 4 should promote digital literacy programs, multilingual and voice-based interfaces, and inclusive access through affordable or offline channels. Predictive AI enhances evidence-based policymaking by improving efficiency in congestion pricing, waste management, and predictive maintenance. However, biased historical data may deepen inequality unless addressed through fairness audits and continuous feedback. Incorporating sustainability audits into predictive modeling advances SDG 13 by reducing emissions and optimizing energy use.
Hybrid AI combines the creativity of generative systems with the precision of predictive tools, enabling both inclusion and resilience. These models support SDG 11 and SDG 17 by powering digital twins for disaster preparedness, optimizing renewable energy, and visualizing long-term sustainability scenarios. Strengthening ethical and transparent data governance further contributes to SDG 16 by building accountability and public trust. Policymakers should invest in data governance frameworks, cross-sector partnerships, and AI literacy to ensure that AI-driven smart cities evolve as equitable, sustainable, and human-centered ecosystems.
5. Limitations and future research directions
This review is subject to several limitations that merit careful consideration. The reliance on Scopus and Web of Science, although ensuring scholarly rigor and consistency, may unintentionally exclude regionally relevant or domain-specific studies that are not indexed in these databases. The exclusive focus on English-language publications introduces potential language bias, potentially leading to the underrepresentation of important contributions from non-English-speaking regions, especially in emerging smart city contexts. The selected timeframe from 2018 to 2025 captures recent advancements but may omit earlier foundational studies and the most recent research that has not yet been indexed. Furthermore, variations in research design, sample characteristics, and policy environments among the reviewed studies may affect comparability and limit the generalizability of findings.
Future research should broaden the range of databases, include translated studies from multiple languages, and conduct cross-regional longitudinal analyses to capture contextual and institutional diversity better. Employing mixed-methods designs supported by open and interoperable datasets can also enhance empirical validity and strengthen theoretical integration. Advancing these methodological and contextual dimensions will not only refine subsequent analyses but also inform the formulation of inclusive, evidence-based AI governance frameworks that promote equitable and sustainable smart city development, consistent with the Sustainable Development Goals. Furthermore, future investigations should assess how AI adoption outcomes correspond with specific SDG indicators to produce more tangible evidence of sustainability impacts. In support of this agenda, Table S4 in the supplementary material presents a curated overview of recent studies (2021–2025) that explore emerging directions and methodological innovations in AI adoption and smart city governance, providing a concise reference base for continued scholarly inquiry in this field.
6. Conclusion
This review systematically integrates perspectives on user readiness and technology adoption through a comparative analysis of generative and predictive AI in smart cities. The findings reveal that generative AI enhances participatory engagement and inclusivity by facilitating interactive citizen input, collaborative urban simulations, and scenario-based planning. In contrast, predictive AI strengthens operational efficiency and evidence-based policymaking through real-time forecasting, optimization, and data-driven decision-making. Combined, these approaches address the social, technical, and governance dimensions of smart city transformation.
The results highlight the strategic alignment of AI-driven innovation with the United Nations SDGs. Generative and predictive AI contribute directly to SDG 11 (Sustainable Cities and Communities) through inclusive planning and resilient infrastructure, SDG 7 (Affordable and Clean Energy) through demand forecasting and efficient energy management, and SDG 13 (Climate Action) through emission monitoring and climate-responsive policy modeling. They further advance SDG 9 (Industry, Innovation, and Infrastructure) by reinforcing IoT and edge computing ecosystems; SDG 3 (Good Health and Well-Being) through AI-enabled healthcare; and SDG 12 (Responsible Consumption and Production) through data-enabled logistics and waste management optimization.
Academically, this review contributes to the theoretical understanding of AI adoption by synthesizing user-readiness theory with behavioral, technical, and governance determinants and by positioning hybrid AI models as an integrative framework that links participatory engagement with operational performance. Practically, it contributes by providing a foundation for policymakers and urban stakeholders to design transparent, inclusive, and accountable AI governance systems that foster citizen trust and social legitimacy. Advancing equitable AI-driven cities requires strategic investments in digital literacy, collaboration, and ethical data governance to drive sustainable, inclusive, and resilient urban futures aligned with the SDGs.
The authors would like to thank the BPI (Indonesian Education Scholarship), the PPAPT (Center for Higher Education Funding and Assessment), the LPDP (Indonesian Endowment Fund for Education), and STIE YKPN Yogyakarta for their valuable support and encouragement throughout this research.
The supplementary material for this article can be found online.
