Matrix Analysis of Structural Convergence of Energy-Relevant and Policy-Relevant AI Research: Implications for Energy Policy

Abstract

The rapid expansion of artificial intelligence (AI) research does not automatically imply its structural integration into industry governance systems. In the energy sector, this raises the question of whether a policy-relevant AI regime has already emerged or whether a structural gap persists between technological development and institutional integration. This study is based on a dataset of 792,417 publications indexed in Scopus (1981–2025). Using the AI-Assisted Research Methodology, a piecewise linear phase segmentation of the AI corpus publications was applied. A matrix model was developed to analyze the distribution of energy relevance (Y) and policy relevance (X) in X–Y coordinates. The results indicate that AI research entered a phase of unstable growth after 2017 and a phase of methodological acceleration after 2021. Despite the growth of both indicators (X, Y), the structural concentration of research related to energy and policy remains moderate (zone 2). The adoption of AI in policy is significantly faster than its integration into energy, suggesting an institutional lag. This study introduces the concept of a “synchronization zone” as an indicator of structural convergence and proposes a framework for assessing the degree of AI integration in energy governance. The findings shift the analytical focus from the growth of publications to the structural configuration and contribute to the development of more coordinated strategies in digital and energy policy.

1. Introduction

Over the past two decades, the growth of publication activity and the expansion of Artificial Intelligence (AI) applications have encompassed virtually all sectors [1,2,3,4,5], including energy [3]. The energy sector is characterized by high technological complexity, economic significance, and political sensitivity [3,6]. With the emergence of new energy challenges and the need to make decisions under uncertainty, the role of AI tools supporting analysis, forecasting, and optimization is increasing [4]. However, the quantitative expansion of AI research does not automatically mean its integration into energy-specific and policy-oriented governance mechanisms. In this regard, the question of not only the scale of AI application in energy but also the nature of the structural convergence between energy-relevant and policy-relevant research within the field of AI research becomes fundamental.

1.1. Research Gap

Despite the growth of AI research, the structural relationship between energy relevance and policy relevance remains understudied in a single body of publications. A formalized approach that allows simultaneous assessment of the dynamics of these two dimensions and their mutual configuration remains lacking. Existing research primarily focuses on the technical aspects of AI applications in energy [3,7,8]. This research includes demand forecasting, network optimization, generation modeling, and distributed systems. In parallel, the field of policy-relevant AI, which aims to support managerial and economic decisions across sectors, is developing [4,9,10]. However, these areas are often analyzed separately. Review and bibliometric studies are typically limited to assessing publication volume, citation rates, or thematic clusters [1,3,10]. They do not consider the structural relationship between sectoral (energy-relevant, E) and political (policy-relevant, P) orientations within a single AI corpus. There is a lack of tools to quantitatively assess the degree of their synchronization and determine whether a stable area of overlap has formed, corresponding to the policy-relevant regime in the energy sector.

1.2. Objective of the Study

This study aims to develop and test a matrix approach to analyzing the structural dynamics of the AI corpus, enabling us to assess the relationship between energy relevance and policy relevance over time and identify the nature of their interaction.

The significance of this task extends beyond academic interest. The energy sector is at the center of global transformations driven by decarbonization, digitalization, and greater infrastructure resilience. In this context, the quality of analytical and managerial decisions directly impacts economic stability and social well-being. Understanding whether a structurally integrated AI regime is emerging in the energy sector or whether an institutional lag persists between the development of AI as an infrastructure and its sectoral integration is essential for the development of long-term energy policy. Based on the identified research gap, this paper aims to answer the following research questions:

RQ1. How do energy-relevant and policy-relevant research sectors evolve during different phases of AI corpus growth?

RQ2. Do the combined dynamics of energy-relevant and policy-relevant sectors indicate a structural shift toward a policy-relevant configuration?

RQ3. Is there an institutional lag between the expansion of policy-relevant AI and its sectoral integration in the energy sector?

Answers to these questions allow us to move beyond describing the volume of publication activity to analyzing structural convergence and assessing the degree of integration of AI research into energy governance mechanisms.

1.3. Contribution of the Study

The contribution of this study lies not in the two-dimensional representation itself, but in its targeted operationalization for analyzing the structural dynamics of AI in energy policy. A matrix approach is proposed that combines energy relevance and policy relevance dimensions with phase segmentation of the AI corpus, enabling a transition from describing publication growth to analyzing development regimes and their mutual configuration.

The theoretical contribution consists of developing and operationalizing a matrix model of structural convergence. The concept of a “synchronization zone” is introduced as an indicator of the formation of a policy-relevant regime in the energy sector. This allows for the identification of transitional states between the quantitative expansion of AI and its institutionalization in the industry.

The methodological contribution lies in integrating phase segmentation of the time series into the analysis of structural shares. It is shown that changes in the slope of the publication activity trajectory reflect not only accelerated growth but also a redistribution of research attention within the AI corpus. This fact allows for interpreting AI development not as a linear accumulation but as a shift in regimes. An additional contribution is the development of the AI-Assisted Research Methodology approach as a tool for identifying structural gaps and transition points in the scientific field. Unlike traditional reviews, the focus shifts from thematic classification to the analysis of configurations and development modes.

A separate result is the empirical verification of the institutional lag concept. It is shown that policy-relevant AI develops faster than its sectoral integration in the energy sector. This effect manifests as asymmetry in the dynamics of the X and Y indicators and the absence of a stable transition to a “synchronization zone.” Thus, institutional lag is quantified.

Overall, a conceptually and empirically sound framework for analyzing the structural convergence of AI in the energy sector is proposed. It combines phase dynamics, matrix configuration, and institutional interpretation. This combination allows AI to be viewed not only as a technology but also as an element of the energy sector’s management infrastructure. The practical significance of this study lies in providing a tool for assessing the level of AI integration in the energy sector, not based on the volume of publications, but on the structural synchronization between energy relevance and policy relevance. The proposed matrix model allows for the identification of transitional states between the technological development of AI and its institutionalization in energy policy. This model reduces the risk of strategic misinterpretation of digital maturity in energy systems.

The study’s results can be applied at several levels. At the international and European levels, the findings contribute to aligning digital and energy transformations and to reducing the potential for conflicts between regulatory and sectoral developments. At the public administration level, the model enables more accurate energy policy development based on the actual configuration of AI integration. At the level of research organizations, the model provides a basis for developing interdisciplinary programs that balance energy and policy relevance. At the individual researcher level, the model enables positioning one’s research within the coordinates of energy and policy relevance, identifying areas of shortage, and adjusting the thematic focus. The combined use of results at multiple levels reduces the risk of fragmented research and improves its relevance to the emerging AI-driven energy governance regime.

The identified institutional gap has practical implications. The study demonstrates that the expanded use of AI as an infrastructure does not automatically lead to its effective integration into energy governance. Therefore, the key practical outcome of the study is the development of an analytical framework for a more synchronized and systemic implementation of AI in energy policy.

The article is organized as follows: Section 2 discusses the general areas of AI use in research and the theoretical foundations of this study. Section 3 describes the data used, the procedures for operationalizing the energy relevance and policy relevance indicators, and the phase segmentation method and matrix model construction. Section 4 presents the results of the AI-assisted source analysis from 1981 to 2025. It also interprets the matrix configuration and assesses the degree of structural convergence. Section 5 contains a discussion of the results. Section 6 describes some multi-level recommendations.

2. Literature Review

2.1. Empirical Context of AI Corpus Development

This section serves a supporting role within the study. It is used not for independent analysis but to provide an empirical context for further examination of the structural dynamics of the AI corpus using a matrix approach.

To analyze the dynamics of changes in the number of publications for the keyword “Artificial Intelligence,” the Scopus database was used. Data was processed for the period 1981–2025. Article titles, abstracts, and keywords were used to perform the search. The total number of publications searched was 792,417 (Figure 1).

Figure 1. Total number of publications for the keyword “Artificial Intelligence” in the Scopus database, 1981–2025 (total 792,417 publications).

Figure 1 serves two functions. First, it substantiates the topic’s relevance: its scale over the past five years precludes its treatment as a niche area. Second, it provides an empirical basis for periodization and subsequent matrix analysis. Of particular interest is the analysis of AI publications by field of study. Such an analysis for 2025 is presented in Figure 2.

Figure 2. Distribution of publications by field of knowledge.

In 2025, the Scopus database recorded 144,390 publications for the keyword “Artificial Intelligence” (Figure 2). This section shows the relative share of publications across fields such as “Business, Management, and Accounting,” “Decision Sciences,” and others. Fields such as “Energy” and “Economics, Econometrics, and Finance” are grouped under the heading “Other.”

These observations provide the empirical basis for the subsequent theoretical analysis, which is developed in Section 2.2 and formalized as a matrix model in Section 2.3.

2.2. Typical Examples of AI Use in Research in 2021–2025

AI research has progressively shifted from a technology-centered inquiry to an ecosystem-oriented paradigm in which AI constitutes a structural layer of the knowledge economy. Early debates focused on AI adoption, perceptions, and sectoral efficiency, examining behavioral responses, organizational culture, and productivity effects across marketing, SMEs, fintech, and labor markets [11,12,13,14,15]. At the same time, bibliometric and trend analyses documented the rapid expansion and diversification of AI-driven enterprise management and innovation systems, revealing an acceleration in both patent activity and cross-country AI dynamism after 2021 [16,17]. The structural underpinnings of this shift are captured by the concept of AI ecosystem pillars and vibrancy subindices, which link the institutional, infrastructural, and human-capital dimensions of AI development to macroeconomic performance and labor-market restructuring [18,19]. In parallel, analyses of total factor productivity and the “AI paradox” suggest that AI’s transformative role lies less in isolated productivity gains and more in systemic reconfiguration of innovation architectures [20]. Together, these strands reveal a transition from AI as an object of investigation toward AI as an infrastructural condition shaping research, governance, and economic coordination.

Within energy-relevant domains, AI has moved beyond optimization tasks and is now embedded in sustainable industrial and municipal systems. Applications in waste management, smart classification, and industrial automation illustrate how AI integrates sensor networks, CNN-based recognition, and Industry 4.0 frameworks to support green transition objectives [21,22]. AI also contributes to energy justice and equity by enabling data-driven allocation, monitoring, and predictive assessment in energy systems [23]. Regulatory and entrepreneurial analyses in renewable energy and clean-digital sectors demonstrate that AI interacts with governance regimes, start-up ecosystems, and public funding procedures, thereby influencing structural conditions for sustainable innovation [24,25]. At the intersection of digital development and university–industry collaboration, digital maturity has been identified as a bidirectional driver of R&D integration, reinforcing AI’s infrastructural function in knowledge transfer networks [26]. Patent-based evaluations of EU innovation performance further indicate that AI-driven metrics reshape how technological leadership is assessed, particularly in green and energy-intensive sectors [27]. These findings collectively position AI not merely as a tool applied to energy research, but as a connective matrix linking sustainability, governance, and innovation infrastructures.

Simultaneously, policy-oriented AI has penetrated organizational, public, and research environments, transforming analytical and managerial architectures. Evidence from project management, public health modeling, and pharmaceutical R&D shows that AI enhances efficiency, nonlinear forecasting, and investor–researcher coordination, thereby embedding algorithmic reasoning within core research workflows [28,29,30]. In law enforcement, healthcare leadership, and occupational safety, AI systems are increasingly framed as dual-use infrastructures requiring ethical governance and adaptive leadership rather than as experimental tools [31,32,33]. Generative AI has triggered a paradigm shift in knowledge management, immersive industrial metaverses, and digital twin modeling, reinforcing AI’s role as a simulation and decision backbone of complex socio-technical systems [34,35]. The transformation extends to education, digital innovation transfer, and debates over AI proficiency, where AI tools reconfigure pedagogical processes and institutional leadership patterns [36,37,38]. Cultural, creative, and publishing industries similarly demonstrate that AI reshapes value chains and intellectual production processes [39,40]. Across these domains, AI evolves from an analytical instrument into a structural decision layer embedded within organizational and research infrastructures.

The infrastructural turn of AI between 2021 and 2025 is further accompanied by ethical, institutional, and socio-technical reframing. Ethical tensions, anomic trends in leadership, and sociotechnical risks emphasize the need to reconceptualize AI governance as a systemic component of institutional architecture rather than a compliance add-on [41,42]. Digital inclusion, economic equity, and enterprise transformation under market disruption illustrate that AI-enabled analytics mediate structural stratification and resilience [43]. Innovation dynamics within merged research organizations reveal that exploratory and exploitative skills are increasingly mediated by AI-supported processes, reinforcing AI’s infrastructural role in innovation cycles [44]. Employee engagement, productivity, and smart feedback management systems also demonstrate AI’s integration into continuous organizational sensing mechanisms [45,46]. Consequently, the emerging scientific landscape depicts AI as a matrix that interconnects energy-relevant applications and policy-relevant architectures, signaling a structural shift from AI as an object of research to AI as a constitutive element of research infrastructure. This transformation underscores the need for matrix-based analytical frameworks capable of mapping convergence zones and structural reconfigurations in contemporary AI research.

2.3. Theoretical Framework

This study is based on the assumption that AI should be viewed not only as a technology but also as a methodological infrastructure for knowledge production. In this context, of particular interest is the moment when AI ceases to be solely an analytical tool and begins to perform a decision-support function in specific industry systems [3,4,16,17,20,26,34,35,39,40,44]. Energy is one such industry system. It combines technological complexity, economic significance, and political sensitivity [3,6,14,26,39]. Therefore, it is not simply the presence of AI in energy research that is important, but its integration with managerial, economic, and political decision-making logics. Based on this, a matrix analysis model is proposed and presented in Figure 3.

Figure 3. Energy relevance (Y-axis) and policy relevance (X-axis) distribution matrix published within the AI corpus.

Figure 3 shows the distribution of publications within the AI corpus along two independent dimensions:

-

Policy-relevance (x-axis, X)—the share of publications focused on management, economic, and political decisions (this study uses the key term “policy relevance.” This refers to the institutionally oriented focus of research on managerial, economic, and political decisions);

-

Energy-relevance (y-axis, Y)—the share of publications related to energy research.

The proposed matrix (Figure 3) aims to capture the structural relationships among the independent dimensions (X and Y). Each dimension reflects the share of publications in the AI corpus, rather than mutually exclusive categories. As a result, the matrix represents a space of two coexisting trends. The analytical value of the matrix lies in its ability to identify not only the presence of these trends but also their relative balance. The four zones reflect qualitatively different configurations of AI development. Thus, the matrix is a diagnostic framework rather than a descriptive tool. It allows us to assess whether AI evolution is following a path of structural synchronization or remains characterized by asymmetry between sectoral and political development.

Zone 1: Low energy relevance and high policy relevance.

AI is actively used to support management, economic, and political decisions outside the energy sector. This is where policy-making tools are developed for various industries (medicine, education, finance, and others). Overlap with energy research may be absent or partial. This zone reflects the policy orientation of AI without any industry tie to the energy sector.

Zone 2: Low energy relevance and low policy relevance.

AI is applied in empirical, fundamental, or other research without a clear energy or political focus. From this study’s perspective, this is a peripheral area.

Zone 3. High energy relevance with low policy relevance.

AI is applied in the energy sector primarily within the engineering and technological logic. This logic relates to modeling, technical optimization, and technical analysis. However, political aspects remain secondary. This is a technological, but not policy-relevant, mode of integrating AI into the energy sector.

Zone 4. Simultaneously high energy relevance and high policy relevance.

This is the area where the two logics intersect. Here, AI is used to formulate, evaluate, and support both technical and political decisions in the energy sector. In this zone, artificial intelligence acquires direct significance for energy policy, strategic management, and institutional design. This zone is considered a target for this study. It is not necessarily the largest in volume, but it is the most significant in terms of consequences.

AI-Assisted Research Methodology [47,48] is a research methodology that uses artificial intelligence (AI) to support the research process. This approach enables automating specific tasks, identifying patterns in data, and formulating hypotheses. This research methodology is considered in this study as an approach that combines algorithmic data processing with analytical interpretation of structural changes in the scientific field [49,50,51,52,53,54,55]. AI-Assisted Research Methodology has already proven itself in energy research [9,56]. Unlike traditional scientometrics and bibliometrics, our study aims to identify developmental modes and structural turning points.

The proposed matrix is not intended to describe the entire AI field. Its purpose is to capture the moment when artificial intelligence becomes both energy-relevant and policy-relevant in energy research. Thus, the matrix allows for a transition from descriptive scientometrics to an analysis of structural shifts reflecting the emergence of policy-relevant AI in the energy sector.

3. Materials and Methods

3.1. Data and Operationalization

The empirical basis of the study is a set of publications indexed in the Scopus database. The earch query was formed in the TITLE-ABS-KEY field using the keyword “Artificial intelligence.” The observation period covers 1981–2025. The total sample size was 792,417 documents. Aggregated annual data were used to analyze phase dynamics and structural configuration. In this study, the AI corpus is defined as the set of publications (N) that satisfy the specified search query for the corresponding year. The unit of analysis is the publication, aggregated at the annual level. To analyze the structural orientation of the AI corpus, two dimensions were operationalized:

-

Policy relevance (X);

-

Energy relevance (Y).

The policy relevance indicator (X) reflects the degree of political, managerial, and economic focus of AI research. The P set is formed based on the publications’ affiliation with the subject areas of (a) Business, Management, and Accounting; (b) Decision Sciences; and (c) Economics, Econometrics, and Finance. The annual value is calculated as follows:

X = 100 × |P|/N, %

(1)

where |P| is the number of policy-relevant publications within the AI corpus.

The policy relevance (X) indicator is operationalized by classifying publications into Scopus subject areas: Business, Management, and Accounting; Decision Sciences; Economics, Econometrics, and Finance. These areas reflect the formalized contours of policy-oriented research and serve as proxies for assessing political significance. The terms “decision-oriented,” “managerial,” “political orientation,” and similar terms are used in the text solely as interpretative synonyms and are not considered separate operational categories.

The selection of subject areas for forming the P set is based on their institutional focus on managerial, economic, and political decision-making. The included areas (Business, Management, and Accounting; Decision Sciences; Economics, Econometrics, and Finance) reflect the formalized contours of policy-oriented research within the Scopus framework. The exclusion of the Social Sciences area is methodological. This category is highly heterogeneous and includes normative-political, sociocultural, and theoretical research. This outcome creates the risk of artificially inflating the policy relevance indicator (X) by including publications that are thematically incomparable. Therefore, the choice of the P set is aimed at increasing the indicator’s operational certainty rather than maximizing coverage. This decision ensures comparability of results over time, but simultaneously creates the risk of partially underreporting individual policy-relevant studies, which is taken into account when interpreting the results.

The energy relevance (Y) metric is defined as the proportion of publications within the AI corpus that are related to energy topics. Empirically, the E set is formed based on the presence of energy topics in the TITLE-ABS-KEY. The annual value is calculated as follows:

Y = 100 × |E|/N, %

(2)

where |E| is the number of energetically relevant publications within the AI corpus in the corresponding year.

The sets E and P are not mutually exclusive. A distinctive feature of the Scopus data structure is its multi-classification: the same document can be assigned to multiple fields of knowledge. Consequently, the total number of publications by field of knowledge exceeds the actual volume of documents. This circumstance is taken into account when interpreting the indicators and is not considered a statistical anomaly. Therefore, the sets E and P are not mutually exclusive. Publications can simultaneously belong to both energy- and policy-relevant logic. Their joint distribution is interpreted using the matrix model presented in Figure 3. This definition allows us to move from descriptive scientometrics to an analysis of the structural synchronization of energy and policy orientation within the AI corpus.

3.2. Phase Segmentation for Time Window Selection

Within the AI-Assisted Research Methodology, the publication activity time series for the period 1981–2025 (Figure 1) was segmented using piecewise linear approximation [47,48,57,58]. The purpose of the segmentation was to identify statistically distinct growth modes characterized by changes in the trajectory’s slope. That is, the time series of the total number of publications was approximated by linear segments to minimize total error. The segmentation validity criterion was the coefficient of determination (R²) for each interval.

The ChatGPT 5.2 neural network was used for processing. ChatGPT was used in a supportive and strictly formalized manner. The model was applied to implement a time-series segmentation procedure using the specified algorithm (Appendix A), including breakpoint selection and linear segment parameter estimation. To ensure reproducibility:

–

A fixed prompt was used (presented in Appendix A);

–

Input data (YEAR, Y) were provided unchanged;

–

The number of segments (K) was specified in advance, with alternatives tested (K = 3, 4, 5);

–

The results were compared with calculations performed using standard statistical tools (Windows 10).

The consistency of the coefficients of determination (R²) and segment slopes confirmed the robustness of the obtained results. Thus, ChatGPT was used as a tool for implementing an algorithmic procedure that is reproducible with the specified input parameters, rather than as a source of interpretation.

Four stages of growth of the AI research field from 1981 to 2025 were identified:

-

Pre-institutional stage—low and unstable growth;

-

The institutional expansion stage is a stable linear regime;

-

The transitional stage of a structural breakthrough is a sharp change in slope;

-

The methodological acceleration stage is intense growth.

The differences between the stages are determined not by the absolute level of publication activity, but by the change in the growth rate in the sections of Figure 1. Thus, the segmentation captures the change in the AI field’s operating mode.

Based on the resulting phase structure, the period 2012–2025 was selected for a detailed analysis. The chosen period of 2012–2025 represents the minimum sufficient interval for analyzing the synchronization of energy relevance and policy relevance in research in the context of the transition from the instrumental use of AI [1,4,11,12,13,14,15] to its infrastructural role [3,4,16,17,20,26,34,35,39,40,44].

3.3. Matrix Construction and Bibliometric Workflow

Based on annual values of the policy relevance (X) and energy relevance (Y) indicators for the period 2012–2025, a matrix model of the AI corpus distribution is constructed in X–Y coordinates. The matrix is interpreted as the space of the structural relationship between two independent variables.

To clarify the thematic structure of the E and P segments, bibliometric analysis is additionally applied. Its purpose is to identify content clusters and research areas that form energy-relevant and policy-relevant publications.

The bibliometric analysis was employed as a complementary methodological layer to explicate the thematic and semantic structure underlying the quantitative matrix. While the matrix model captures the relative positioning of energy relevance and policy relevance within the AI corpus, bibliometrics is used to uncover how these dimensions are internally articulated through recurring research topics, conceptual linkages, and clustering patterns. The analysis follows the PRISMA framework to ensure transparency and reproducibility in dataset construction, including identification, screening, eligibility assessment, and final inclusion of publications. The data source was the Scopus database, with the search strategy restricted to the period 2012–2025 and applied to titles, abstracts, and keywords. Only peer-reviewed journal articles and conference papers written in English were retained, ensuring both disciplinary relevance and methodological consistency with the matrix-based analysis.

For thematic mapping and cluster detection, the VOSviewer software (version 1.6.19) was used to construct keyword co-occurrence networks based on author keywords and indexed terms. Co-occurrence analysis was selected because it allows identification of stable semantic associations rather than isolated topic frequencies, thereby revealing the internal logic through which energy-relevant and policy-relevant AI research evolves. The resulting network visualizations were interpreted not as standalone outputs but as structural complements to the matrix model: clusters were analyzed in terms of their proximity to the AI core, their degree of interconnection, and their position between technological, economic, regulatory, and governance-oriented themes. This integration of bibliometric mapping with matrix analysis enables a shift from measuring publication volumes to interpreting modes of convergence, thereby allowing the assessment of whether observed quantitative dynamics correspond to meaningful structural integration between energy- and policy-oriented AI research. Thus, the matrix model captures the quantitative distribution of indicators, while bibliometric analysis reveals their thematic content. The combined use of these methods allows us to move from assessing share dynamics to analyzing the structure of convergence within the AI field.

4. Results

To ensure logical coherence of the results with the posed research questions (RQ1–RQ3), the presentation of the results is organized as follows. Section 4.1 aims to identify the phases of the AI corpus’s development (RQ1). Section 4.2 and Section 4.3 reveal the dynamics and structural configuration of the energy relevance and policy relevance indicators (RQ1, RQ2, RQ3). Section 4.4 complements the quantitative analysis with thematic interpretation necessary to explain the identified structural effects (RQ2, RQ3).

4.1. Selecting Time Windows

Figure 4 shows the result of dividing the curve of the number of publications in the AI corpus (Figure 1) into straight-line sections.

Figure 4. Four stages of AI research development, 1981–2025, Scopus.

Figure 4 shows the temporal dynamics of publication activity in the AI corpus for the period 1981–2025. Four segments approximate the original curve of total publications, each reflecting a distinct growth mode and as close as possible to a straight line. This is a structural segmentation based on minimizing approximation error. Each segment is described by a linear function equation and is characterized by its own slope and R² coefficient of determination.

Graphically, the difference between the stages is determined not simply by changes in the level of publication activity, but by changes in the trajectory’s slope. It is the change in slope, not the absolute number of publications, that serves as the criterion for identifying stages.

The gap between the second and third stages is particularly revealing: beginning in 2017, there is a noticeable acceleration, which will transform into a new regime of rapid growth in 2021–2025. Thus, Figure 4 empirically confirms the transition of artificial intelligence from a developing field to a research infrastructure. It is important to emphasize that this periodization serves as the basis for selecting time stages for subsequent analysis of the energy and policy relevance of research. Thus, Figure 4 links the empirical dynamics of the AI field with the theoretical framework presented in Section 2.3.

The graphical segmentation presented in Figure 4 requires both quantitative and qualitative descriptions. The visual differences in slope reflect changes in the growth regime, but a rigorous comparison of stages is only possible by accounting for the numerical parameters of the linear approximations. For this purpose, the R² coefficients of determination were calculated for each time segment. The resulting numerical data were verified using the built-in statistical tool in Windows 10. The verification showed consistency between the data obtained using the standard method and that obtained using the specially developed prompt. The resulting R² values are presented in Table 1.

Table 1. Piecewise-linear growth phases of AI-related publications (1981–2025).

Stage	Period	Growth Pattern (Slope)	Meaningful Interpretation
I	1981–1999	Weak and unstable, R2 = 0.62	Pre-institutional period. Publication activity is low and fragmented, and growth is unstable. AI does not perform a systemic methodological function and is not integrated into policy-relevant research.
II	2000–2016	Moderate, persistently positive, R2 = 0.97	A period of institutional expansion. AI is becoming established as a recognized research object and tool. Publication growth is linear and reflects the normalization of AI methods across various subject areas, including energy, with no signs of structural acceleration.
III	2017–2020	Abrupt change in slope, unstable growth, R2 = 0.93	Transitional (turning point) period. A structural shift in publication dynamics is observed, associated with the spread of data-driven and deep learning approaches [52,57,59,60]. This period marks the AI field’s emergence from its previous growth phase and its preparation for a qualitatively new stage.
IV	2021–2025	High, accelerated growth (new regime), R2 = 0.93	A period of methodological acceleration. Publication dynamics indicate that AI has transformed from an auxiliary tool into an element of research infrastructure [17,19]. Growth is expansive and is accompanied by an increase in policy-relevant [21,22,28,30,39,40,61,62].

Table 1 presents both a qualitative description of the stages and their quantitative characteristics. While Figure 4 captures the change in trajectory shape, Table 1 demonstrates the scale of differences between the growth regimes. In other words, Table 1 presents the quantitative parameters of the piecewise linear approximation corresponding to the segments highlighted in Figure 4. The resulting values demonstrate not only an increase in publication activity but also a qualitative change in the growth rate between phases.

During the pre-institutional stage (1981–1999), the slope remains low, and the R² value reflects the limited predictability of the dynamics. The publication field is emerging but has not yet demonstrated a stable development regime. This period is excluded from further study.

During the institutional expansion stage (2000–2016), the slope increases by an order of magnitude, and the R² value reaches high levels (0.97). This fact indicates stabilization of the growth trajectory and the formation of a predictable structure of publication activity. A sharp increase in the slope characterized the transition phase (2017–2020), while the linear model maintains high explanatory power (R² = 0.93). This observation indicates a structural break rather than a random fluctuation. The methodological acceleration phase (2021–2025) exhibits the highest slope. The rate of publication growth is several times higher than in previous phases. Moreover, the high R² value (0.93) confirms that the observed expansion is systemic. The increased publication rate during the acceleration phase reflects not only the quantitative expansion of the AI field but also the greater likelihood of structural alignment between industry and managerial logic. This phase is directly related to the formation of Zone 4 as the target area of analysis.

Thus, Table 1 quantitatively confirms the phased structure of the AI field’s development. The difference between the segments is determined not by the absolute volume of publications, but by changes in the growth pattern. This circumstance is crucial for the present study, as the subsequent matrix analysis relies precisely on the differences between the regimes. Consequently, Figure 4 and Table 1 form a single analytical construct: the former reveals the phase structure, while the latter explains and describes it.

The choice of the 2012–2025 time window is not arbitrary and is not based on a visual impression of publication dynamics. It follows from the structural segmentation presented in Figure 4. The 2000–2016 period is characterized by steady linear growth, reflecting the institutionalization of artificial intelligence as a research object and tool. Within this period, a relative stabilization of the slope is observed, making the early years less sensitive to structural changes in the distribution of thematic areas. The period beginning in 2017 marks a shift in the growth regime. Nevertheless, for an accurate analysis of the convergence dynamics between the energy relevance and policy relevance variables, a representative “baseline” point before the turning point is necessary. The 2012–2016 subinterval was chosen as such a point. This is the final section of the institutional period, consisting of five points, characterized by stable linear growth, no signs of acceleration, and publication activity comparable to subsequent years. So, the period 2012–2025 includes three logically distinguishable states of the field:

-

A late institutional norm (2012–2016);

-

A structural breakthrough stage (2017–2020);

-

A methodological acceleration stage (2021–2025).

This time window (2012–2025) allows us to analyze not the overall growth of publications, but rather the changing configuration of the AI corpus in the context of the transition from the instrumental use of artificial intelligence to its infrastructural role. It is in this context that it becomes possible to assess the dynamics of the interaction and overlap between energy relevance and policy relevance. The choice of the period 2012–2025 ensures the matrix’s methodological sensitivity to structural changes. It is precisely in this interval that it is possible to record not just an increase in publication activity, but also a redistribution of attention within the AI corpus and the dynamics of synchronization between energy relevance and policy relevance.

These results allow us to answer RQ1: the development of the AI corpus occurs through a series of qualitatively distinct phases, each with a different growth pattern. These phases form the basis for a subsequent analysis of the structural dynamics of energy relevance and policy relevance.

4.2. Analysis of AI Penetration into Policy Relevance and Energy Relevance Research

The results of the quantitative analysis presented in Figure 5 serve as an empirical basis for the subsequent interpretation of the dynamics of energy and policy relevance within the AI corpus.

Figure 5. Dynamics of AI penetration into policy relevance and energy relevance research.

Figure 5 demonstrates the dynamics of artificial intelligence’s penetration into two analytical dimensions (policy relevance and energy relevance) from 2012 to 2025. Unlike Figure 4, which reflects the overall growth pattern of the AI field, this figure shows a redistribution of publications within the AI corpus.

Over the period under review, both the sum of energy and policy relevance publications demonstrated steady growth. In 2012, the corresponding indicator was approximately 111–339 publications per year. In 2025, it exceeded 6645 publications (for the “Economics, Economics, and Finance” field of knowledge). Overall, the sum of policy relevance publications exceeds the sum of energy relevance publications. This outcome indicates a gradual shift toward a stronger managerial, economic, and political focus within AI research. Despite lower absolute values, a steady trend toward greater energy issues in the AI corpus is observed. The key here is not the absolute number of publications, but their simultaneous positive dynamics. Figure 5 shows that after 2017, a period of synchronous increases in indicators for all four scientific fields begins. And during the methodological acceleration phase (2021–2025), this trend intensifies. This fact means that the AI field is not simply expanding quantitatively, but is also structurally redistributing attention toward policy-relevant and energy issues. Thus, Figure 5 establishes the empirical basis for the subsequent structural analysis of the matrix configuration (Figure 3). The dynamics of the indicators demonstrate not only quantitative growth but also a targeted increase in their policy and energy relevance. This observation provides an empirical basis for answering RQ2, which assesses the structural shift in the AI corps’ configuration.

4.3. Temporal Dynamics of the X–Y Configuration, 2012–2025

The observed dynamics (Figure 5) indicate an increasing likelihood of intersection between energy relevance and policy relevance research within a single AI corpus. The quantitative values of the corresponding indicators by year are presented in Table 2. This table serves as the empirical basis for subsequent analysis, as it is the data used to construct the matrix configuration of the energy and policy relevance distributions. Table 2 thus provides a transition from the description of dynamics (Figure 4 and Figure 5) to a structural interpretation presented in matrix form. Based on the annual values of policy- and energy relevance for the period 2012–2025, an X–Y matrix model is constructed, enabling visualization of the structural configuration of the AI corpus. Table 2 compiles preliminary data to create the energy and policy relevance matrices for publications.

Table 2. The share of energy and policy relevance publications in the Scopus database for the period 2012–2025.

Year	Total Number of AI Publications, N	Number of Policy-Relevant Publications, P	Share of Policy Relevance Publications, %	Number of Energy Relevance Publications, E	Share of Energy Relevance Publications, %
2012	17,390	733	4215	198	1139
2013	16,368	887	5419	196	1197
2014	17,069	923	5407	250	1465
2015	18,484	1171	6335	387	2094
2016	22,469	1521	6769	779	3467
2017	23,772	2070	8708	689	2898
2018	27,044	3002	11,100	1045	3864
2019	30,053	3997	13,300	1525	5074
2020	38,365	5805	15,131	1644	4285
2021	46,520	8167	17,556	2184	4695
2022	54,763	8537	15,589	3152	5756
2023	72,131	12,061	16,721	4216	5845
2024	104,758	20,158	19,242	6156	5876
2025	144,390	26,249	18,179	7160	4959
2012–2025	633,576	95,341	15,048	29,581	4669

Table 2 contains annual values of the policy relevance and energy relevance indicators for the period 2012–2025, including those calculated as the share of relevant publications within the AI corpus. These data form the quantitative basis for constructing the matrix model presented below. Analysis of the values reveals several consistent patterns.

First, policy relevance demonstrates consistent growth throughout the period. Starting at approximately 4–5% at the beginning of the period, the indicator gradually increases. By 2024–2025, it reaches values exceeding 19% (Table 2). The dynamics are predominantly monotonic, indicating a structural strengthening of the managerial, economic, and political components within the AI field.

Second, energy relevance also shows positive dynamics, but at a significantly lower scale. The growth is more gradual, starting at approximately 1% at the beginning of the period and increasing to 4–6% in recent years (Table 2). Despite the difference in scale, the direction of change coincides with the policy relevance dynamics.

Thirdly, after 2017, both indicators began to grow synchronously. In 2021–2025, not only absolute growth but also a higher rate of change is observed. This pattern indicates a stronger joint movement of X and Y within the matrix space.

Therefore, Table 2 records not only quantitative growth but also a redistribution of structural emphases within the AI corpus. These changes serve as the empirical prerequisite for interpreting the matrix configuration and assessing the degree of synchronization between energy relevance and policy relevance variables.

Using the data in Table 2, an X–Y matrix was constructed for the period 2012–2025, reflecting the relative positions of the policy relevance and energy relevance indicators within the AI corpus (Figure 6).

Figure 6. Matrix of AI penetration into energy and policy relevance research, 2012–2025 (Zone 2).

Figure 6 visualizes the matrix configuration of the policy relevance (X) and energy relevance (Y) indicators for the period 2012–2025. Each point corresponds to a separate year, and its position is determined by the values presented in Table 2. The matrix enables quantitative interpretation of structural changes within the AI corpus.

During 2012–2016, the points remained concentrated in the lower range of both indicators, reflecting a stable but limited level of structural orientation. In the 2017–2020 period, a directional shift becomes observable, with simultaneous increases in both dimensions. In 2021–2025, this movement intensifies, indicating a consistent trajectory toward higher levels of combined relevance. Thus, the matrix captures not only the growth of indicators, but a directional transformation of their joint configuration over time.

The resulting configuration demonstrates directional co-movement between energy and policy relevance, with a persistent difference in their relative levels. These results directly address RQ2 and serve as the empirical basis for the subsequent analysis of RQ3.

4.4. Bibliometric Analysis

The matrix model presented above enabled us to capture the quantitative structure of the AI corpus and assess the degree of synchronization between energy relevance and policy relevance. However, the matrix reflects the structural relationship of shares without revealing the substantive content of these areas. To clarify the nature of the observed asymmetry and identify the thematic nodes that form the energy- and policy-relevant segments, bibliometric analysis is used. Its goal is not to re-evaluate the volume of publications, but to identify the semantic clusters, research foci, and directions that define the internal structure of the AI corpus. Thus, bibliometric analysis complements the matrix model. While the matrix captures the direction of structural movement, bibliometrics enables us to understand the thematic and institutional configurations that drive it.

The bibliometric analysis was conducted according to the PRISMA diagram (Figure 7). Analysis tool—VOSviewer, version 1.6.19, ©2009–2023 Nees Jan van Eck and Ludo Waltman—was employed.

Figure 7. PRISMA diagram for bibliometric analysis (a) and explicit search string in text ((b), screenshot from Scopus database).

The keyword co-occurrence map (Figure 8) provides a semantic projection of the AI corpus’s internal structure at the intersection of energy relevance and policy relevance, complementing the matrix-based analysis by revealing how thematic clusters are organized, interconnected, and hierarchically positioned.

Figure 8. Query “artificial intelligence” AND “energ*”: keyword map, 2012–2025 (https://www.scopus.com/ accessed on 21 February 2026, analysis tool—VOSviewer).

At the center of the map (Figure 8), artificial intelligence appears as the dominant hub, confirming its role not merely as a technical label but as a structural connector linking heterogeneous research trajectories. Its centrality and density indicate that AI functions as a transversal methodological layer rather than a bounded technological niche. Closely adjacent to this core, energy efficiency and energy form a tightly connected semantic nucleus, signaling that energy-related AI research is increasingly framed not in isolated engineering terms but in relation to system-level performance, optimization, and sustainability outcomes.

The centrality of energy efficiency in Figure 8 does not contradict the conclusion that policy-relevant AI matures faster than its energy-specific institutionalization; rather, it helps explain the mechanism of this transition. In the energy domain, efficiency is not merely an engineering concept. It also functions as a managerial and policy-relevant category because it is directly linked to cost reduction, resource allocation, decarbonization targets, infrastructure optimization, and system resilience. For this reason, energy efficiency appears as an intermediate semantic bridge between technically oriented AI applications and broader policy-relevant energy governance. The map, therefore, suggests that the movement from Zone 3 toward Zone 4 does not begin with abstract governance terminology alone, but with a technically grounded efficiency discourse that is easier to operationalize, measure, and institutionalize. In this sense, efficiency is the point at which technical AI becomes progressively policy-relevant, even though full institutional synchronization has not yet been achieved.

Around this central core, several partially overlapping clusters can be identified. One cluster, oriented toward energy management, energy management systems, smart energies, and clean energy, reflects the technological-operational logic of AI integration in energy systems. This cluster emphasizes optimization, control, and forecasting within complex infrastructures and is strongly connected to algorithmic and model-centric terms such as algorithms, AI models, and technological forecasting. Its position close to the core but slightly offset toward the engineering side of the map corresponds to the Zone 3 configuration in the matrix model (Figure 3), where energy relevance is high. At the same time, decision and policy orientation remain secondary.

Partially overlapping cluster is structured around innovation, research and development, technological change, and business technology adoption. This cluster is more explicitly decision-oriented and economically framed, linking AI to investment dynamics, adoption processes, and institutional transformation. The strong co-occurrence between investments, economic analysis, and economic growth indicates that AI research increasingly addresses questions of allocation, efficiency, and strategic impact rather than purely technical feasibility. This semantic configuration aligns with the policy-relevant dimension of the matrix (X-axis) and corresponds to Zones 1 and 2 (Figure 3), where decision orientation grows faster than sector-specific embedding.

The map reveals a bridging zone where energy efficiency, clean energy, environmental regulations, and AI technologies intersect. This area marks the main intersection between technological AI research and regulatory, environmental, and governance-oriented themes. The presence of environmental regulations as a mediating keyword suggests that AI-driven energy research increasingly operates within normative and institutional constraints, rather than purely technical optimization frameworks. However, despite this emerging overlap, the density of links in this bridging zone remains lower than in the core AI or innovation clusters, suggesting that the semantic integration of energy and policy relevance remains incomplete (Figure 3, Zone 2).

To make the complementarity between the bibliometric and matrix analyses more explicit, the main semantic clusters identified in Figure 8 can be mapped onto the matrix zones (Table 3). The Zone 3 configuration (high energy relevance, low policy relevance) is characterized by keywords such as smart energy, clean energy, algorithms, AI models, and technological forecasting. These terms indicate strong sectoral embedding of AI in energy systems, while decision-oriented and regulatory dimensions remain secondary. By contrast, the more explicitly policy-relevant cluster structured around innovation, research and development, technological change, business technology adoption, energy management, energy management systems, investments, economic analysis, and economic growth corresponds mainly to Zones 1–2, where policy-relevant AI expands faster than its energy-specific institutionalization. The overlap around energy efficiency, clean energy, environmental regulations, and AI technologies most clearly represents the bridging trajectory toward Zone 4. This intersection does not yet form a fully consolidated Zone 4 core, but it marks the semantic pathway through which technological AI in energy begins to acquire policy relevance.

Table 3. Thematic clusters of Figure 8 and their correspondence to the matrix zones.

Matrix Zone/Transition	Dominant Keywords	Interpretation
Zones 1–2 (policy-relevant but weakly energy-embedded)	innovation; research and development; technological change; business technology adoption; energy management; energy management systems; investments; economic analysis; economic growth	Policy-relevant AI develops faster than sector-specific embedding in energy
Zone 3 (High energy relevance, low policy relevance)	smart energy; clean energy; algorithms; AI models; technological forecasting	AI is strongly embedded in energy systems, but mainly in technical-operational terms rather than regulatory or strategic ones
Bridging zone toward Zone 4	energy efficiency; clean energy; environmental regulations; AI technologies	Transitional overlap where technical AI begins to connect with regulation, governance, and sustainability-oriented decision making
Emerging Zone 4 logic	partial overlap of energy efficiency with regulatory and decision-oriented terms	Early semantic convergence is visible, but a dense and stable synchronization zone has not yet formed

Peripheral but conceptually significant clusters include forecasting, prediction, Industry 4.0, and Industry 5.0. These terms reflect AI’s infrastructural role in anticipatory governance, system modeling, and industrial transformation. Their position on the margins of the energy-policy intersection suggests that, while forecasting and industrial paradigms are widely adopted, their explicit linkage to energy policy and governance remains fragmented. Similarly, the presence of augmented reality and other advanced digital technologies at the periphery underscores the expansion of the AI ecosystem, but also highlights the dilution of thematic focus that accompanies infrastructural growth (Figure 3, Zone 4).

The keyword map shows that the overlap between technological and policy-relevant themes remains limited and uneven (Figure 8). Instead, it visualizes a field characterized by strong methodological centralization around AI, moderate thematic clustering around energy efficiency and management, and a still-emerging semantic overlap with policy, regulatory, and governance domains. This configuration is consistent with the matrix results (Figure 6) and shows that the semantic overlap remains uneven.

The bibliometric analysis demonstrates that the structural configuration of AI research at the intersection of energy relevance and policy relevance is characterized by asymmetrical convergence rather than full synchronization. While artificial intelligence clearly functions as a unifying methodological core, its thematic integration into energy-policy-relevant research remains partial and uneven. The dominance of clusters related to energy efficiency, optimization, and technological innovation indicates that AI has been successfully embedded as an operational and analytical tool within energy systems. However, the comparatively weaker density of links connecting these clusters to regulatory, governance, and decision-making themes reveals that the transition from technological application to policy-relevant integration is still ongoing (Figure 6).

The observed semantic structure shows a stronger expansion of policy-relevant themes than of sector-specific policy integration within energy-relevant AI research. This imbalance mirrors the matrix-based finding that empirical observations remain confined to Zone 2, despite a precise directional movement toward Zone 4 (Figure 6). Thus, the results of the bibliometric analysis are consistent with the quantitative findings of the matrix model and clarify the thematic structure related to RQ3.

5. Discussion

5.1. RQ1: How Do Energy-Relevant and Policy-Relevant Sectors Evolve During Different Phases of AI Corpus Growth?

The answer to RQ1 should not be considered in isolation from the overall dynamics of the AI field, but rather within the logic of changing growth modes identified in Figure 4 and Table 1. It is the change in the slope of the publication activity trajectory that sets the context for interpreting the structural shares.

Late phase of the institutional expansion period (2012–2016).

During this period, both indicators (policy relevance and energy relevance) are at a relatively low level (Figure 5, Table 2). AI is expanding quantitatively, but its sectoral and policy focus remains unclear. Policy relevance gradually increases, but remains within the single digits. Energy relevance demonstrates an even more limited share. This state can be characterized as infrastructural preparation without sectoral concentration. AI is gaining a foothold as a method, but not as a policy tool, in the energy sector.

Transitional phase of a structural breakthrough (2017–2020).

Since 2017, both indicators have grown synchronously. This fact coincides with a change in the AI corpus’s growth pattern (Figure 4). Policy relevance demonstrates an accelerated increase in share (Table 2). Energy relevance is also growing, but more moderately. What’s important here is not the difference in scale, but the convergence of trends. It is here that the structural correlation between the technological acceleration of AI and the strengthening of managerial and industry-specific focus first becomes apparent. AI is no longer merely an analytical tool. It is becoming part of policy-making (Figure 5).

The phase of methodological acceleration (2021–2025).

Amid the expansive growth of the AI field, both sectors continue to expand (Table 2; Figure 6). Policy relevance reaches double-digit values and stabilizes at around 16–19%. Energy relevance grows more slowly and remains in the 4–6% range. This range means the following:

-

The policy orientation of AI is becoming structurally entrenched;

-

Energy integration is expanding, but lags behind managerial dynamics.

Thus, the evolution of the sectors is asymmetrical yet synchronous. The simultaneous direction of change in X and Y confirms synchronicity. The asymmetry is manifested in the scale of shares.

Thus, energy-relevant and policy-relevant segments go through three stages:

-

Low concentration with quantitative growth of AI (institutional norm);

-

Simultaneous acceleration with a structural breakthrough;

-

Entrenchment of managerial orientation while maintaining a gap with industry dynamics.

AI is evolving from a neutral research infrastructure to a managerially oriented knowledge system. At the same time, the energy component is developing more slowly. It is this difference in pace that underlies the analysis of RQ2 and RQ3: the movement toward a policy-relevant configuration and the presence of an institutional lag.

The evolution of energy-relevant and policy-relevant segments across different growth phases of the AI corpus reveals a shift from parallel expansion to partial coupling (Figure 8). In the late institutional phase, AI research primarily expands as a methodological infrastructure with limited sectoral or policy specificity. Energy-related applications exist but are predominantly framed within engineering and optimization paradigms, whereas policy-oriented AI primarily develops outside the energy domain. During the transitional and acceleration phases, both segments grow simultaneously. Still, their growth reflects different logics: energy relevance increases through efficiency-driven and system-management applications, whereas policy relevance expands through innovation, investment, and governance-focused research. This outcome indicates convergence in direction, but not yet convergence in structure.

5.2. RQ2: Does the Joint Dynamics of Their Energy-Relevant and Policy-Relevant Sectors Indicate a Structural Movement Toward a Policy-Relevant Configuration?

RQ2 requires a more rigorous answer than simply observing simultaneous growth in the X- and Y-axis indicators. Co-movement alone does not necessarily indicate structural convergence. The question is: is the field configuration changing, or are we merely observing parallel expansion of segments?

Quantitative growth vs. structural configuration.

Over the period 2012–2025, both indicators increase (Table 2, Figure 6). However, the X–Y matrix shows that all empirical points remain within Zone 2. This fact indicates growth, but concentration remains low to moderate. If a transition to a policy-relevant configuration were occurring in the strict sense, we would expect a steady shift in points toward the zone of simultaneous high energy and policy relevance (Zone 4) in Figure 3. This does not occur.

Direction of movement.

However, the trajectory of the point shift is not chaotic (Figure 6). Since 2017, we have seen a consistent movement to the right (increase in X) and upward (increase in Y). This is not a one-time fluctuation. It is a systemic shift, coinciding with the phase of methodological acceleration of the AI field (Figure 4). Consequently, the combined dynamics indicate not complete convergence, but rather a directed structural movement.

Asymmetry as an indicator of incompleteness.

Policy relevance is growing faster and reaching significantly higher shares. Energy relevance is increasing, but its magnitude remains 3–4 times lower. If a fully fledged policy-relevant configuration were to emerge in the energy sector, the share of energy relevance should grow in proportion to the policy dynamics. In fact, we observe the following structure:

-

AI is becoming policy-oriented;

-

The AI-energy component is developing with a lag. This lag may indicate partial, but not complete, synchronization.

AI research is structurally shifting toward a policy orientation. Energy research is gradually joining this process. However, the transition to a stable policy-relevant configuration in the energy sector by 2025 remains incomplete. It is this intermediate state between acceleration and synchronization that determines the current mode of development of the AI field.

The joint dynamics of energy- and policy-relevant research point toward a gradual shift toward a policy-relevant configuration without crossing the threshold of structural synchronization. The matrix trajectory (Figure 6) and the keyword map (Figure 8) both show a consistent upward and rightward shift, signaling increasing overlap. However, the absence of empirical points in Zone 4 (Figure 6) and the limited density of semantic links between regulatory and energy-technical clusters (Figure 8) suggest that this movement remains incomplete. The evidence, therefore, supports the interpretation of an emerging, but not yet consolidated, policy-relevant AI regime in the energy sector.

5.3. RQ3: Is There an Institutional Lag Between the Expansion of Policy-Relevant AI and Its Industry Integration in the Energy Sector?

RQ3 takes the discussion to the interpretive level. While RQ1 captures dynamics and RQ2 shows the direction of structural movement, the key result here is the disproportionate pace of evolution.

Empirical basis for the lag hypothesis.

Between 2012 and 2025, policy relevance increases from 4–5% to 18–19%, while energy relevance grows from ~1% to 4–6% (Table 2). That is, growth is present in both segments. However, the scale of growth differs. In the methodological acceleration phase (2021–2025), the gap does not narrow; instead, it stabilizes. Policy relevance grows faster than energy relevance (Figure 6). This is a quantitative signal of a possible institutional lag.

However, it should be noted that the observed asymmetry may partly be due to the different scales on which the indicators are operationalized. Policy relevance (X) is defined at the subject area level. Energy relevance (Y) is defined at the keyword level. This fact may create differences in measurement sensitivity. This effect is considered a limitation and is taken into account when interpreting the results (see the Limitations section).

Lag as a structural phenomenon, not a time delay.

This is not about the energy sector being “a few years late.” It is about different implementation modes. Policy-relevant AI is spreading across three subject areas: Business, Management and Accounting; Decision Sciences; Economics, Econometrics, and Finance. These fields have a relatively low institutional barrier: the publishing infrastructure is already prepared for AI integration [18,19,26,37,61]. Energy is an entirely different type of system. It is an industry with high technological complexity, capital intensity, and regulatory rigidity [3,6,9,23,27,56]. Integrating AI into energy policy requires not only algorithms, but also:

-

Alignment with regulatory regimes;

-

Integration into infrastructure cycles;

-

Adaptation to long-term investment horizons.

Consequently, we observe an institutional lag in sectoral integration (in energy) with a faster growth in political orientation.

The matrix as a tool for capturing the lag.

Figure 6 shows a shift in empirical points to the right and upward. However, all points remain in Zone 2. This fact means that:

-

AI has already become politically relevant;

-

AI has not become simultaneously politically and energetically focused.

This discrepancy is precisely the manifestation of the lag. If the sectoral component (energy) were catching up with the political one, the group of green points would approach Zone 4. However, this does not occur in the 2021–2025 interval under consideration.

Interpretation within the logic of development modes.

The methodological acceleration phase (2021–2025) is characterized by the expansion of AI as a research infrastructure (Figure 4, Table 1). However, the acceleration of infrastructure growth is not identical to the acceleration of sectoral growth. Thus, a gap is emerging between:

-

The growth rate of AI research in the politically oriented sector;

-

The growth rate of AI research in the energy sector.

This is the structural institutional lag. Thus, empirical data point to the presence of an institutional lag. AI becomes policy-relevant in the political domain faster than it is integrated into the energy sector as a systemic tool for managerial, economic, and political decision-making. This lag does not negate the movement toward convergence. It explains why Zone 4 in Figure 6 has not yet formed. It is this discrepancy in pace that determines the current state of AI research’s structural evolution in the energy sector.

The observed asymmetry, however, requires a more explicit interpretation in terms of the mechanisms driving this effect. First, differences in funding structures influence the rate of AI diffusion across different segments. Policy-relevant AI is more often developed within flexible research and interdisciplinary programs, while energy-relevant AI is tied to more capital-intensive and technologically specialized areas. Second, disciplinary silos persist between engineering, economics, and management research. This outcome slows the formation of sustainable intersections between energy relevance and policy relevance within a single research body. Third, there is slow policy uptake. Even with the availability of analytical tools, their integration into real-world management processes takes time. At the same time, this asymmetry is not solely a lag effect. It reflects differences in development regimes. Policy-relevant AI functions as a supra-sectoral analytical layer and scales more quickly. Energy-relevant AI is developing within the context of sectoral challenges and requires deeper contextual integration. Thus, the “institutional lag” captures not only a temporal gap but also a structural asynchronousness between the “politicization” of AI and its sectoral specification.

A concrete example of this institutional lag is evident in the current interaction between the EU AI Act and energy-sector regulation. The AI Act sets requirements for high-risk AI systems, including transparency, risk management, and accountability. However, in the energy sector, AI is already integrated into critical infrastructures such as smart grids, demand forecasting systems, and grid balancing [63,64].

In practice, this creates a situation in which energy operators must simultaneously comply with sector-specific technical standards (e.g., grid codes, reliability requirements, cybersecurity protocols) and cross-sector AI regulations. These regulatory layers are not fully synchronized. As a result, AI-based solutions that have been technically validated in energy systems may face additional uncertainty regarding compliance with the AI Act, particularly regarding classification as high-risk systems and the requirements for documentation and human oversight. This uncertainty slows the adoption of AI in energy systems. Instead of simply integrating, organizations must adapt AI-based solutions to overlapping regulatory regimes. This adaptation leads to implementation delays, increased compliance costs, and a tendency to prioritize technically sound but less policy-integrated solutions. In this sense, the observed institutional lag is not abstract: it manifests itself in the misalignment between the EU AI Act and sectoral energy regulation.

The bibliometric map provides qualitative evidence of an institutional lag: AI expands rapidly as an infrastructural and decision-support technology, but its consolidation as a stable component of energy governance and policy design has not yet been achieved. Consequently, the bibliometric analysis reinforces the study’s central conclusion: the current AI-energy nexus is defined not by equilibrium but by a transitional configuration in which semantic convergence precedes structural synchronization. The combined matrix (Figure 6) and bibliometric evidence (Figure 8) clearly indicate an institutional lag between the expansion of policy-relevant AI and its integration into energy systems. At the same time, the prominence of energy efficiency in the keyword map suggests that this lag is mediated through an intermediate stage in which AI first acquires policy relevance via technocratic optimization and performance-oriented discourse before becoming fully embedded in broader regulatory and governance frameworks.

Policy-relevant AI research matures more quickly, developing sophisticated decision-support frameworks and governance narratives that are not yet fully incorporated into energy-specific research agendas. Energy-relevant AI, in turn, remains anchored mainly in technological and efficiency-oriented applications, with slower incorporation of policy, regulatory, and strategic decision dimensions (Figure 8). This lag is not accidental but structural, reflecting differences in institutional incentives, disciplinary boundaries, and temporal rhythms of technological versus policy integration. As a result, the current configuration should be interpreted as a transitional regime in which semantic convergence precedes institutional alignment, rather than as a failure of AI integration in energy governance.

The results are consistent with existing studies examining AI as a factor in transforming management and decision-making systems. Several studies [4,9,10,65,66] have shown that AI advances faster in analytical and management models than in their industry integration [3,7,8]. This fact corresponds to the identified asymmetry between policy relevance and energy relevance. At the same time, in contrast to studies focusing primarily on the technological aspects of AI in the energy sector [1,3,10,63,64], the results presented here indicate a structural gap between the development of analytical tools and their institutionalization in industry. Thus, this study complements the existing literature by demonstrating that a key limitation is the asynchronous integration of AI tools into management and industry frameworks.

6. Conclusions

This study aimed to quantify the degree of structural convergence between energy- and policy-relevant research sectors within the AI corpus and to determine whether AI has reached the threshold of sustainable integration in the context of energy policy. Unlike traditional scientometrics and bibliometrics, the focus was not on publication volume but on the configuration of their sectoral and policy orientations over time.

The main conclusion of this study is that the development of artificial intelligence in the energy sector is characterized not by a lack of growth, but by a structural asymmetry. Policy-relevant AI has already established a stable trajectory, while its integration into the energy sector remains incomplete. This observation confirms that the current stage should be viewed more as a transitional state than as a complete configuration with policy implications. Given the identified asymmetry between policy relevance and energy relevance variables, it is appropriate to formulate recommendations at multiple levels of decision-making.

Global energy policy.

At the level of international organizations and global energy platforms (IEA, UN Energy, etc.), the following should be prioritized:

-

Institutional recognition of AI as an element of energy infrastructure, not just an analytical tool;

-

Development of an international framework for assessing the AI readiness of energy systems;

-

Inclusion of AI integration indicators in energy transition monitoring mechanisms.

The key challenge is to synchronize the digital and energy transformations. Without this, AI will enhance analytical potential, but not the sector’s strategic governance.

European Union level.

For EU energy policy, the study’s findings have practical implications for the Green Deal, digital sovereignty, and AI regulation. Here, the following recommendations can be made:

-

Integrating AI metrics into strategic documents on energy security and decarbonization;

-

Creating interdisciplinary programs that combine AI, energy, and public policy, rather than considering them in parallel tracks;

-

Align AI regulations (AI Act) with industry energy standards.

The EU has a well-developed regulatory infrastructure. Therefore, the problem is not a lack of regulation, but its fragmentation. The strategic objective is to reduce the institutional lag by coordinating digital and energy regulatory instruments.

National energy policy level.

At the national level, it is recommended to

-

Include AI components in national energy strategies not as technological applications, but as elements of the political decision-making system;

-

Establish specialized AI centers for energy policy;

-

Revise long-term forecasting models taking into account AI tools.

Particular attention should be paid to the “regulator—infrastructure operator—analytical center” link. Without institutional “coupling,” AI will remain at the pilot project level.

Research institute level.

Universities and research centers play a key role in reducing the identified lag. At this level, it is recommended to

-

Develop research programs at the intersection of energy relevance and policy relevance, rather than within isolated disciplinary segments;

-

Stimulate joint interdisciplinary research, projects, and publications between the faculties of energy, management, economics, and public policy;

-

Develop educational modules on AI-driven energy governance.

The institutional structure of science largely determines the configuration of the X–Y matrix. If the academic environment remains fragmented, synchronization slows.

Individual researcher level.

For researchers, the strategic conclusion is the need to go beyond narrow engineering or purely political logic. Promising areas include

-

Analysis of the intersection of sets E ∩ P (energy relevance and policy relevance);

-

Development of indicators of AI effectiveness in energy policy;

-

Study of institutional barriers hindering implementation. A publication strategy focused exclusively on technical optimizations strengthens Zone 3. Working at the intersection of energy and policy analytics facilitates progress toward Zone 4.

Taken together, the recommendations boil down to one principle: AI should be integrated into energy policy as a systemic governance component, not as an auxiliary optimization tool.

The structural lag identified in this study is not a sign of weakness in the energy sector. It is an indicator of incomplete institutional adaptation. Reducing this lag requires coordinated action at the global, European, national, institutional, and individual levels. Only with such multi-level synchronization is it possible to develop a sustainable, policy-relevant AI configuration in the energy sector.

This study has limitations. The analysis is based on aggregated Scopus data and operationalizes energy and policy relevance through subject areas and keywords. A significant limitation of the study is the operationalization of the policy relevance (X) indicator across Scopus subject areas (Business, Management, and Accounting; Decision Sciences; Economics, Econometrics, and Finance). This approach ensures the reproducibility of the analysis but does not eliminate the risk of classification bias. On the one hand, publications classified in these areas do not always possess genuine managerial or policy significance, as they may be highly technical or applied in nature. On the other hand, studies with clear policy relevance may be classified in other subject areas (e.g., Social Sciences or Environmental Science) that were not included in the calculation of the X indicator. This fact ensures replicability but does not reveal the depth of specific integration cases.

Therefore, the values of X and Y should be interpreted as structural indicators of the distribution of research attention, rather than as precise measurements of substantive policy or sectoral significance. To improve the robustness of the results, it seems advisable to use alternative approaches in further research, including keyword analysis, topic modeling, and expanding the set of subject areas. These tools will allow for sensitivity testing of the results and refinement of the boundaries of the identified structural configurations. The proposed matrix model can be further expanded by incorporating additional dimensions, such as regulatory intensity, technological maturity, and the pace of implementation in specific sectors. This model will allow for a transition from a two-dimensional representation to a multi-level analytical framework capable of accounting for more complex AI integration configurations.

A promising direction for further research is an in-depth analysis of the overlap of the E ∩ P sets, as well as a comparative study of individual countries and institutional contexts. Future research should also test the proposed model across sectors beyond energy, such as healthcare, transportation, and public administration. A cross-sector comparison will allow us to test the robustness of the institutional lag concept and determine whether similar patterns of asymmetry emerge in other fields using artificial intelligence.

Overall, the results obtained allow us to draw a key conclusion: artificial intelligence has already become a global research infrastructure, but its structural integration into energy policy is still in its infancy. The observed movement toward synchronization is directional, but incomplete. It is this intermediate state that defines the current stage of AI research in the energy sector.