Introduction

In the light of growing awareness, sustainability concerns are more topical than ever for all actors in society [1]. More often part of management decision-making choices [2,3,4] accounting practices [5], as well as those of Reporting [6], both within companies and public institutions [7].

Consequently, responsible commitment is manifested through sustainable actions following a strategic approach of performance management and results measurement, with an approach of adapting to the requirements in the field [8], thus making it possible to respond to demands for legitimacy [9, 10].

In recent years, many organisations have voluntarily committed to increasing the disclosure of ESG information in response to the growing expectations of internal and external stakeholders, including shareholders [11]. Thus, meeting stakeholders’ expectations in terms of transparency, and especially their specific expectations, requires their involvement in the process of drawing up these reports [12].

Currently, the field of non-financial disclosure is characterised by the diversity of global initiatives, which has made the subject more complex to deal with [13]. In the same context, we can also mention the implementation of different requirements in this area, through laws, regulations, guidelines and standards.

Whether voluntary or mandatory, the publication of this information has become a widespread practice within companies [14].

However, the international scope of the subject, its role in the evolution of corporate operations and its ability to generate long-term value while respecting economic objectives as well as social and environmental resources, mean that we need to continue exploring the subject in depth.

Scientific research regarding ESG disclosure and its determinants has acquired an increasingly considerable importance, in this regard, several studies have been conducted at this level. A few have looked at the relationships of different practices with ESG reporting [15], or focus on a specific context or country [16,17,18,19].

Given the diversity and complexity of the determinants of ESG communication, this study uses artificial intelligence tools to complement the traditional qualitative systematic approach. The use of natural language processing (NLP) and clustering analysis techniques makes it possible to objectively identify the conceptual structures of the determinants and motivations of ESG communication.

We propose an exhaustive mapping of the determinants of ESG communication using a systematic literature review; we have carried out a thematic and contextual analysis of the articles collected using NLP; and we have developed, using artificial intelligence algorithms that create clusters of the variables identified, a conceptual framework that links these groups of determinants to the underlying organisational motivations.

Our study offers a unique methodological contribution to the literature by incorporating artificial intelligence methods into the systematic review of ESG disclosure determinants. While previous systematic reviews rely primarily on manual qualitative analysis techniques, this research implements Natural Language Processing (NLP) algorithms to conduct systematic thematic and semantic analysis. This reduces the risk of subjective interpretation and improves the external validity of conclusions. Our major innovation lies in the use of TF-IDF vectorization combined with unsupervised clustering methods to identify, classify and organize factors into homogeneous clusters of motivations.

This computational approach makes it possible to reveal latent patterns in the literature and to propose an empirically grounded taxonomy of the factors that explain ESG disclosure, thus overcoming the limitations of the theoretical classifications commonly used in previous studies. Our study offers a unique methodological contribution to the literature by incorporating artificial intelligence methods into the systematic review of ESG disclosure determinants. While previous systematic reviews rely primarily on manual qualitative analysis techniques, this research implements Natural Language Processing (NLP) algorithms to conduct systematic thematic and semantic analysis. This reduces the risk of subjective interpretation and improves the external validity of conclusions. Our major innovation lies in the use of TF-IDF vectorization combined with unsupervised clustering methods to identify, classify and organize factors into homogeneous clusters of motivations.

This computational approach makes it possible to reveal latent patterns in the literature and to propose an empirically grounded taxonomy of the factors that explain ESG disclosure, thus overcoming the limitations of the theoretical classifications commonly used in previous studies.

The study is therefore structured as follows: the next section presents the conceptual and theoretical background to the research, before moving on to a detailed description of the methodology adopted in the second section. This is followed by a presentation and discussion of the results.

Theoretical background

CSR: from commitment through to disclosure

The absence of a definition of CSR, shared by scientists and professionals, tells us about the degree of transversality and complexity of this concept, which is both spongy and vague [20, 21]. However, it is clear that CSR aims to consolidate the economic, social and environmental objectives of organizations [22]. It challenges companies to optimize their positive externalities and limit their negative ones, while adopting ethical behavior.

Even if there is no consensus on a definition of CSR, the one adopted in November 2010, as part of the ISO 26 000 guidelines, currently constitutes one of the most accomplished references in terms of CSR/RSO principles [23,24,25,26,27,28].

This is the conception of CSR we have chosen for this article. According to iso 26,000, CSR reflects an organization’s commitment to the impact of its decisions and activities on society and the environment. Such a commitment should:

  • Contribute to the sustainable development, health and well-being of society,

  • Take into account stakeholders’ expectations,

  • Comply with current legislation and international standards of behavior,

  • Be reflected throughout the organization and implemented in its relationships.

Socially responsible companies should earn the trust of investors, customers, employees and other stakeholders through transparent non-financial disclosure. This includes the publication of ESG indicatorsFootnote 1 in ESG reports, social balance sheets, carbon footprints and other relevant documents to ensure ongoing transparency and compliance with applicable standards and regulations. Companies are assessed on their overall performance, which includes both financial and environmental, social and governance (ESG) aspects.

Sustainability is a protean concept that highlights the interdependence between society and the environment [29, 30]. Its complexity stems from the difficulty of balancing its four dimensions: economic, social, sustainable and environmental [31, 32].

Non-financial disclosure refers to the regular publication of a company’s ESG performance [33]. The term ESG encompasses environmental, social and governance considerations [34].

ESG disclosures therefore represent all dimensions of corporate social responsibility and sustainable development [35,36,37]. Reporting through transparent ESG disclosure is not an ethical choice, but a strategic imperative that creates unique competitive advantages [38].

Determinants of environmental social and governance (ESG) disclosure: theoretical justifications

ESG disclosure is justified by a variety of theories that underline the importance of socially responsible companies reporting, on the social and environmental impacts of their activities.

  • Contingency Theory

Contingency theory is an extension of neoclassical theory. It is based on the idea that there is no single approach to organisation. Rather, the internal and external characteristics of the company have a strong influence on its management approach. Organisational and management approaches therefore affect the CSR strategy and its communication. As a result, a company’s sustainable behaviour in this area can be influenced by contingency factors [39].

  • Legitimacy theory

Legitimacy can be defined as a generalized principle that makes a company’s activities suitable, desirable or adequate against a set of social norms, measures, values and considerations [40].

This theory asserts that a company’s survival is conditioned by the conformity of its operations to society’s expectations [36]. It must meet social norms, values and expectations to maintain its social license to operate.

  • Institutional theory

Institutional theory suggests that organizations adopt a range of practices and structures to suit the institutional and regulatory expectations of their environment [41,42,43].

Institutional pressures (laws, standards, market constraints), drive companies to disclose non-financial information to demonstrate their compliance and alignment with societal expectations [44]. As a result, non-financial disclosure is perceived as a tool enabling companies to legitimize their activities in the light of social, economic or political pressures.

  • Stakeholder theory

Stakeholder theory supports the need for the company to take into account the expectations of all its stakeholders, not just those of shareholders [45,46,47]. Indeed, different stakeholder groups are likely to impact, favorably or unfavorably, the company’s performance and reputation [48]. Non-financial disclosure, therefore, makes it possible to demonstrate how the company meets the expectations and concerns of its various stakeholders. Making public the social and environmental impacts of its activity would enable the company to build stronger relationships with all its stakeholders and, consequently, gain their trust and support [49].

  • Agency theory

Proposed by [50], this theory examines the relationship between the owner (principal) and the manager (agent) of companies. It places the possession of information and its sharing between principal and agent at the heart of its thinking. The existence of a situation of informational asymmetry between the two sides fuels the extent of conflicts of interest. By extension, agency theory could be applied to the relationships a company might have with all its stakeholders (shareholders, suppliers, customers, the community, ….).

  • Signal theory

Signal theory [51] represents a central theoretical model for understanding the determinants of ESG disclosure. According to this theory, companies use ESG disclosure as a signalling tool to reduce information asymmetry with stakeholders and to distinguish themselves from underperforming competitors [52]. In this context, ESG disclosure acts as a reliable indicator that enables environmentally and socially conscious companies to publicize their non-visible commitments. This creates a competitive advantage by lowering financing costs and enhancing organizational reputation [53, 54].

  • Resource-Based View (RBV) Theory

The Resource-Based View (RBV) theory, initiated by [55] and developed by [56], offers a strategic framework for companies seeking to understand how to obtain, maintain and strengthen a sustainable competitive advantage. It suggests the need for unique resources and their deployment to develop unique competencies.

By focusing on internal assets rather than external factors, the Resource-Based View helps companies to exploit their distinctive strengths to create value and outperform competitors [57].

The literature review and the associated determinants can be summarised as follows, in Table 1.

Table 1 Summary of theoretical justifications and associated determinants. Source Researchers constructs
Full size table

Research methodology

Therefore, we conducted a systematic literature review to identify, analyze and synthesize knowledge on the determinants of ESG disclosure [54]. This systematic review is conducted in accordance with the PRISMAFootnote 2 guidelines, which is generally regarded as a useful compass in this area [58].

Systematic review steps

The assessment of various previously proposed methods [59,60,61], has enabled us to structure our study around the following stages, illustrated in Fig. 1.

  • a. Study preparation:

  • Formulation of the research question

Fig. 1
Fig. 1
Full size image

Systematic review steps. Source Researchers constructs

In general, the first step in a systematic review is to clearly and precisely formulate the research question. In our case, as noted above, we seek to explore previous studies in order to identify, analyze and synthesize the determinants of ESG disclosure.

  • Identification, choice and justification of databases and keywords:

Databases we have tried to cover a wide range of researches in order to guarantee the reliability of our results. To this end, we have carried out in-depth searches in various databases. In this respect, we decided to explore five databases containing journals related to management sciences (Web of science, ScienceDirect, Cairn, Scopus, Google Scholar).

Keywords To ensure the most exhaustive study possible and minimize selection bias, our search was based on a wide range of keywords and their synonyms, chosen on the basis of the main concepts of our research question. Given that the basic idea is to find studies looking into the determinants of non-financial disclosure, the most logical search statement, formulated in the simplest and most direct way, is as follows:

“Determinants” AND ‘‘ESG disclosure’’.

Nevertheless, things can always be said differently, which may limit our search if we use this single equation. For this reason, and as shown in Table 2, it would be wise to opt for several statements including synonyms for the two key words in the search.

  • Determining selection criteria:

Table 2 Research keywords. Source Researchers constructs
Full size table

At this stage, we will draw up a list of eligibility criteria (Inclusion Criteria and Exclusion Criteria) that will act as filters applied to the results found to select the studies that will be considered for the synthesis [60, 62]. Table 3 shows the inclusion and exclusion criteria for our sample:

Table 3 Choice and justification of inclusion and exclusion criteria. Source Researchers constructs
Full size table

The accessibility of full-text CI1 studies is a prerequisite for the completion of all phases of study selection and evaluation. As a result, all documents for which we were unable to gain access were excluded.

For CI2, the language of writing, only documents written in French or English have been included, the adoption of this criteria being justified by the language skills of the researchers. To this end, all studies published in other languages were excluded.

For CI3, we have defined a search period of 20 years. As recommended by J. Paul, et al. [63], it is preferable not to specify a start date for the research period for emerging topics. On the other hand, for well-established or more mature topics, it is recommended to define a more precise research period of up to 30 years. For the question of the determinants of non-financial communication, this is a more or less emerging subject with a reasonable number of publications, and this explains the period chosen.

We also took into account the nature of the study, adopting an exclusion criteria (EC1). Thanks to this criteria, we were able to limit the studies collected to those that had conducted an empirical study. Theoretical studies, such as literature reviews and systematic reviews, will not be examined.

  • b. Collection and selection:

At the end of the identification phase, 866 references were collected from various databases. We began the process of filtering the studies, by excluding duplicates and articles for which we did not have access to the full text. After reviewing the titles, abstracts and conclusions of the articles and applying our predefined criteria to determine whether the articles met our research objective, we reduced our sample to 76 articles. To reinforce the credibility of our systematic review, a complete reading of the full texts enabled us to exclude articles not relevant to our central research question. In the end, our sample comprises only 71 articles.

The selection process is shown in the PRISMA flow chart (Fig. 2.).

  • c. Data extraction and processing:

Fig. 2
Fig. 2
Full size image

Prisma flow diagram. Source Researchers constructs

To extract and organise the data from each study, we first used Nvivo software to collect the elements characterising each of the included studies. We then carried out a thematic study using natural language processing (NLP). In a second, more in-depth extraction phase, we manually collected the results of each study and organised them around a framework. Once all the elements of the framework had been identified, we carried out clustering to identify clusters of ESG disclosure drivers.

NLP pre-processing and analysis results

In addition to content analysis, we have implemented a computational approach. This is based on the application of algorithmic methods and advanced computing techniques to the analysis of complex data. It is an algorithmic analysis that makes it possible to discover relationships and structures linked to unstructured data [64, 65].

The methodology for processing the determinants of ESG communication is based on two main phases. Firstly, the description and preparation of textual data using language processing (NLP). Subsequently, a clustering analysis enabled us to identify coherent groups of determinants.

Natural language processing (NLP) analysis

NLP analysis aims to transform textual determinants into usable numerical representations. This approach involves three sequential stages:

  • Data preparation and pre-processing The theoretical corpus identified by the systematic review comprised 71 articles. These textual data were pre-processed in several stages; Textual normalization is the first step in the standardization of our theoretical corpus, involving the removal of special characters. This was followed by the reduction of linguistic noise, enabling the processing of compound terms and the deletion of empty words. The standardization of variables and acronym processing was thus ensured through terminological harmonization.

These NLP pre-processing operations were implemented in Python using the re (regular expressions) and string libraries.

  • TF-IDF vectorisation The NLP processing had a secondary stage in which our pre-processed theoretical corpus was transformed into digital vectors using the TF-IDF (Term Frequency-Inverse Document Frequency) technique.

TF (Term Frequency) is used to evaluate the frequency of a term in the processed texts. For a term t in a document d, TF(t,d) = (number of occurrences of t in d) / (total number of terms in d). IDF (Inverse Document Frequency), on the other hand, takes into account the importance of the term in the corpus as a whole by penalising common terms. Formally, IDF(t) = log(N/df_t), where N is the number of documents processed and df_t is the number of documents containing the search term [66].

The TF-IDF score is the result of the two indicators: TF-IDF(t,d) = TF(t,d) × IDF(t).

The TF-IDF text processing technique is widely used in management science for processing annual reports [67] and ESG reports and communications [68]. TF-IDF text analysis thus serves as a preparatory step for thematic clustering of academic literature (Durach et al., 2017).

Exploiting ESG information involves converting unorganized textual content into quantitative data that can be manipulated using computerized methods. We use the TF-IDF (Term Frequency-Inverse Document Frequency) vectorization technique, a recognized method in the field of information retrieval that evaluates the importance of terms by merging their local occurrence in each document with their overall rarity in the corpus as a whole (Salton & Buckley, 1988; Ramos, 2003). This method facilitates the automatic detection of specific terms characterizing company-specific ESG strategies, while eliminating generic jargon shared in company reports [67].

The resulting TF-IDF vectors are then used by a K-means clustering algorithm designed to reduce within-group variance. This algorithm is used to organize elements into homogeneous groups. This undirected approach highlights hidden patterns in ESG communication strategies without advocating a predefined theoretical framework, which is particularly suitable for the diversity of reporting methods being developed [69, 70]. Optimal determination of the number of clusters is achieved by using both the elbow method and the silhouette coefficient, in order to guarantee the statistical rigor of the segmentation [71].

To achieve this, we applied the following parameters to the scikit-learn library:

  • vectorizer = TfidfVectorizer (min_df = 1, ngram_range = (1, 2));

  • min_df = 1: Include terms that occur at least once;

  • ngram_range = (1, 2): include simple terms and expressions consisting of two words.

Vectorisation has allowed us to transform our corpus into a numerical matrix of dimensions [X × Y]. where each term can be represented by a vector in a multidimensional space. This step is crucial for our research and represents a first preparatory step for clustering.

  • Semantic similarity calculation The NLP analysis was completed by calculating the cosine similarity matrix between the TF-IDF vectors of the main search terms. In order to evaluate the semantic relationships between the basic concepts of the study, the similarity matrix, with dimensions [N × N], provides a quantitative representation of the theoretical corpus.

Clustering analysis

Based on the results of the NLP analysis, we implemented clustering techniques to identify conceptually coherent subgroups among the selected determinants. This second stage also involves several methodological steps:

  • Determining the optimal number of clusters The optimal number of clusters was determined using the silhouette score method [71]. This tool is used to assess the quality of the distances between clusters.

We tested hypotheses relating to between 2 and 7 clusters. For each, we calculated an average silhouette score. The results are shown below:

  • For k = 2, silhouette score: 0.0124

  • For k = 3, silhouette score: 0.295

  • For k = 4, silhouette score: 0.295

  • For k = 5, silhouette score: 0.396

  • For k = 6, silhouette score: 0.600

  • For k = 7, silhouette score: 0.600

The analysis showed that the optimal solution is associated with a silhouette score of 0.60, with a number of clusters k = 6 clusters.

Identification of the optimum number of clusters was carried out by systematically examining the silhouette coefficient for various values of k (ranging from 2 to 7 clusters). The data show a gradual improvement in cluster quality: k = 2 (score = 0.012), k = 3 (score = 0.295), k = 4 (score = 0.295), k = 5 (score = 0.396) to reach the highest value at k = 6 and k = 7 (score = 0.600). The silhouette score of 0.60 obtained for k = 6 and k = 7 reveals an appreciable clustering performance [72], demonstrating intense cohesion within the same cluster and pronounced inter-cluster distinction.

We opted for k = 6 because it provides the best balance between statistical clustering quality and simplicity of interpretation, preventing possible over-segmentation at k = 7. This ideal configuration confirms that ESG factors naturally organize themselves into six conceptually distinct and statistically sound clusters of disclosure motivations.

  • Application of the K-means algorithm Using the optimal number of clusters identified, we applied the K-means algorithm [73] with the following parameters.

  • kmeans = KMeans (n_clusters = 6, random_state = 42, n_init = 10).

  • random_state = 42: to ensure reproducibility.

  • n_init = 10: ten different initialisations to choose the best solution.

  • The K-means algorithm was used to divide the determinants into 4 different groups, minimising the intra-cluster variance.

  • Analysis of associated motivations The identification of clusters of drivers was combined with an analysis of organisational motivations. This stage included.

  • Analysing co-occurrences between drivers and motivations.

  • Establishing conceptual relationships between clusters of drivers and categories of motivations.

This methodological process enabled us to develop a two-dimensional framework that integrates both drivers and motivations, thus providing a more complete understanding of the phenomenon of ESG disclosure.

Findings and discussion

The survey identified 71 research papers dealing with the determinants of ESG disclosure. The following sections summaries the main findings of the review.

Description of results

  • The trend in ESG disclosure publications: The studies examined are published over a ten-year research period, from 2005 to mid-2024. As shown in Fig. 3, after a period of stagnation, with an average of one publication per year, the field of research on the determinants of extra-financial disclosure has been on an upward trend since 2011. In 2014, the number of publications increased significantly to ten, followed by nine publications in 2020 and 2021.

    This upward trend can be explained by the importance of ESG criteria in creating value for different stakeholders [74].

  • Most influential authors and countries: Since the end of the twentieth century, the scientific output of authors from different countries in the field of ESG disclosure has evolved considerably. Figure 4, shows the global map of publications by country and region. The intensity of the color reflects the number of publications; the darker the color, the greater the number.

Fig. 3
Fig. 3
Full size image

Sum of publications by years. Source Generated by power BI

Fig. 4
Fig. 4
Full size image

Total publications mapping. Source Generated by power BI

The geographical distribution of the articles is highly diversified, covering more than 45 countries and 3 regions. China led the way with seven articles, followed by Indonesia with six. Among the Arab countries, Saudi Arabia produced the largest number of publications, with a total of 5 articles.

Giannarakis, Grigoris is one of the most productive authors with four publications. Amran, Azlan, Simon S. Gao, Lavin, Jaime F and Orazalin, Nurlan share second place with two publications each on the same topic.

In terms of research methodology, the studies included in the review share the same research approach. The quantitative approach is used extensively to identify and analyse the determinants of ESG disclosure.

NLP pre-processing and analysis results

  • Systematisation and categorisation of the variables:

First, we performed NLP processing, using python code, to create a World cloud showing the frequency of words in the files studied, Fig. 5.

Fig. 5
Fig. 5
Full size image

Word cloud representation. Source Generated via python script

The first observation we can make at this level is related to the appearance in the center of the figure of the terms ‘disclosure, CSR, reporting, corporate…, which represent our dependent variable. This reflects the relevance of the collected studies to our research question. Secondly, the appearance of a few determinants in the cloud and their positioning represents a high frequency of these terms. This represents the importance of these variables and the significance of their impact on ESG disclosure.

The Fig. 6, highlights the 20 terms with the highest average TF-IDF scores in the analysed corpus, providing valuable insight into the relative importance of terms in the ESG communication literature.

Fig. 6
Fig. 6
Full size image

Analysis of average TF-IDF scores for the main search terms. Source Generated via python script

The term ‘disclosure’ clearly stands out with the highest TF-IDF score (above 0.20), underlining the relevance of the collected articles to our research question. This is followed by a group of heavily weighted terms (scores between 0.14 and 0.17), including ‘company’, ‘csr’, ‘corporate’ and ‘social’.

This TF-IDF analysis complements the word cloud by precisely quantifying the relative importance of the terms, confirming the preponderance of the notion of disclosure and the importance of concepts related to the company and its social responsibility in the corpus studied.

The categorization of the variables identified and the calculation of their frequency of occurrence resulted in the Fig. 7.

Fig. 7
Fig. 7
Full size image

Determinants of ESG disclosure identified. Source Generated by power BI

In numerical terms, out of a total of seventy-one mentions, which is the total number of studies examined, the governance structure was examined the most, with 50 mentions, followed by the size of the company, which was assessed in the majority of studies. And with a lower frequency of appearance, profitability, the company’s sector of activity, its CSR commitment, age, leverage, ownership structure, legal system and the characteristics of ESG reports are the main determinants of ESG disclosure.

The other determinants refer to factors that are rarely evaluated. These include liquidity, media exposure, corruption, national culture, origin of control, tangibility, volatility, etc.

  • Analysis of the significance of the results:

These determinants were the subject of a manual collection of the significance of their impact in each study. The observations collected were centralized and processed in an Excel file in order to visualize the trend of the results at this level. Observations on this issue can be presented as follows; in Table 4.

Table 4 Significance and nature of the impact of variables. Source Researchers constructs
Full size table

In terms of the significance of the impact of these variables, the results generally tend towards a significant positive impact. This reflects the relevance of these determinants in the different studies. Company size is the determinant with the best combination of frequency and significance, with unanimous agreement on the positive nature of the impact. CSR commitment and ESG report characteristics are similar to company size, with high levels of significance. And with such logical results, the legal system and the sector of activity have a significant impact in the majority of the studies in which they were processed.

The significance of the effect is reduced for governance structure, financial leverage and ownership. These factors are among the most important determinants of ESG disclosure, but their significance has not been confirmed in all studies.

The results of the significance of the impact of each determinant were correlated with the context of each determinant. In order to visualise the impact of this on the results, in Fig. 8. We have assessed the dispersion of the observations from one context to another by highlighting the country or region of the study in the presentation of the results.

Fig. 8
Fig. 8
Full size image

Significance and nature of the impact of variables. Source Generated by power BI

Each band refers to a determinant and represents the results observed in the different contexts evaluated. Consequently, the percentages are calculated on the basis of the significance of the impact of the drivers in the regions.

For most of the drivers, we see a strong dispersion of results from one region to another. Company size and ESG engagement alone produce the same results in almost all contexts. On the other hand, governance structure, together with the other variables, shows different results from country to country.

This can be explained by the context of each study, except that this contextual factor cannot affect the importance of these determinants as variables likely to have an impact in other contexts.

Clustering analysis results

The systematic assessment of the determinants of ESG communication was complemented by a computational approach based mainly on artificial intelligence tools, natural language processing and clustering.

The identification of clusters of determinants and underlying motivations led to the development of a two-dimensional conceptual framework, shown in Fig. 9. This framework illustrates the different mechanisms that influence ESG disclosure.

Fig. 9
Fig. 9
Full size image

Two-dimensional ESG disclosure framework. Source Generated via python script

The first cluster focuses on the structural characteristics of the company, including factors such as size, age, industry, ownership and governance. This first set provides the basic framework that defines the identity of the organisation.

The determinants related to performance and financial position are the subject of the second cluster. These are profitability, financial leverage and financing needs, and CSR performance.

Another set of determinants related to contextual and institutional factors has been identified. This cluster includes the impact of the external environment, such as the country’s culture and legal system, media exposure and stakeholder pressure.

Finally, there is the cluster of motivations, which are divided into three main categories: economic, strategic and legitimating.

The lines between the different clusters illustrate the dynamic interactions that exist between the different types of ESG communication drivers and motivations. This illustrates the complexity of the factors that influence ESG communication.

This diagram, Fig. 10, illustrates the systematic relationships between the four clusters of drivers identified in our analysis and the types of motivations underlying ESG communication. The thick lines represent the main relationships, while the thinner lines indicate secondary but significant influences.

Fig. 10
Fig. 10
Full size image

Relationship between clusters of drivers and motivations. Source Generated via python script

The diagram of relationships between the different clusters illustrates the specific relationships between the different clusters. This provides a deeper understanding of the two-dimensional framework of determinants and motivations by integrating the intensity and directionality of influences.

Generally speaking, there are two types of relationship, primary and secondary. This reveals not only the existence of direct and privileged relationships, but also the determinants and motivations of the same family. In particular, the structural characteristics that are associated with strategic motivations, the economic motivations that derive from financial and CSR performance, as well as the contextual and institutional factors that develop direct links with motivations of the same nature.

But also the multidimensional nature of these links, through the secondary relationships between determinants and motivations of different kinds. For example, the structural characteristics of the company can lead to legitimating motivations.

Clustering quality evaluation

  • Statistical validity:

Main metric: Average silhouette score of 0.60

Scores per cluster:

  • Structural features: 0.62

  • Performance and financial position: 0.59

  • Contextual and institutional factors: 0.54

  • Economic motivations: 0.63

  • Strategic motivations: 0.64

  • Legitimating motivations: 0.60

Inter-category separation (determinants vs. motivations) 0.72

A silhouette score above 0.5 indicates substantial clustering, while a score above 0.7 is associated with a strong clustering architecture [71]. As a result, our clustering indicates strong intra-cluster cohesion with a full model silhouette score of 0.6. This represents a good distance between clusters. Consequently, our analysis falls within the high validity interval.

  • Methodological robustness

The combination of the two techniques, TF-IDF analysis and K-means clustering, forms a robust and relevant method for working out the underlying structures of qualitative data. It is an approach that is increasingly used in the management sciences and is widely accepted in the literature [65].

Theoretical justification of the developed model

Each cluster of determinants and motivations identified was compared against the theoretical framework used in order to validate the theoretical relevance of each dimension identified, as presented in Table 5.

Table 5 Theoretical framework of ESG disclosure determinants and motivations. Source Researchers constructs
Full size table

The examination of the determinants and motivations of ESG disclosure highlights a complex theoretical structure, presented in. The structural characteristics of the company, its size, age, business sector and governance structure are crucial strategic elements that determine both the capacity and the imperative to communicate on environmental, social and governance issues [56]. The institutional approach offers an additional perspective by showing that isomorphic, coercive, normative and mimetic constraints encourage companies to standardise their ESG disclosure practices within the same organisational domain [75]. This demonstrates the convergent adoption of ESG disclosure strategies in response to institutional expectations [84].

Economic, strategic and legitimacy motivations come together to create a coherent system in which signalling theory [51] clarifies the use of ESG disclosure as a tool for reducing capital costs and accessing resources. Furthermore, strategic motivations find their theoretical justification in the Resource-Based View theory [56], which supports the pursuit of sustainable competitive advantages through ESGdisclosure. As for legitimacy motivations, according to legitimacy theory [40] and institutional theory, they represent a mechanism for maintaining social acceptability and managing image.

Discussion

The systematic review of the determinants of ESG disclosure identified more than 20 determinants. The analysis focused mainly on the drivers that were addressed in the majority of the studies and were considered to have a significant impact in this area.

The applied NLP analysis and clustering methodology enabled us to identify the underlying structures of the ESG disclosure drivers. The resulting conceptual framework consists of three clusters of determinants (structural characteristics, financial performance and contextual factors) and three clusters of motivations (economic, strategic and legitimacy), as shown in Fig. 11.

Fig. 11
Fig. 11
Full size image

Hierarchical framework of ESG disclosure determinants and motivations

The first cluster concerns the main characteristics of the company (size, age, ownership structure and governance). These findings are aligned with [75] contributions on institutional isomorphism, in that structural characteristics form the typology of pressures that the company may face. Thus, firm characteristics are likely to guide the firm’s ESG communication practices [85].

Our analysis of the determinants thus highlights the importance of company size as a key determinant of ESG disclosure, echoing the contributions of [86, 87], who confirmed the importance of the company size as a significant predictor of the level of environmental disclosure.

The second cluster, mainly related to the financial aspect of the company, shows strong internal consistency (silhouette score 0.59). This clustering is consistent with the instrumental perspective put forward by [82, 88], according to which the company thus has tangible economic objectives behind its ESG disclosure.

Financial profitability and financial leverage are the most representative determinants of this group. However, this aspect qualifies the slack resources hypothesis of [89] by suggesting that ESG disclosure is not only influenced by the company’s financial performance, but is part of a broader financial strategy.

Institutional factors are grouped in the third cluster. This cluster shows greater variability (silhouette score 0.54), reflecting the diversity of external influences. The analysis is based on the strong representation of the country’s culture and system in this cluster. These results confirm the findings of [90] on the impact of legal traditions on organisational transparency.

The motivations emerge in three clusters, economic motivations is the first cluster at this level, which is particularly consistent (silhouette score 0.68). reducing the cost of capital is the central economic motivation. Since ESG communication reduces the cost of capital [54].

The second cluster of motivations is the strategic one (silhouette score 0.64), of which seeking competitive advantage and market differentiation are the most representative motivations in this cluster. This clustering is in line with [38] vision, according to which responsible practices are no longer seen as a cost but rather as a strategic opportunity.

The third group of motivations focuses on legitimacy (silhouette score 0.60), based on legitimacy theory [40], which helps to understand ESG communication actions. Brand image and respect for societal expectations stand out as the most representative factors. This finding is consistent with Bansal & Clelland’s (2004) research on environmental legitimacy.

The clusters identified have principal and secondary relationships. First, the strong association represented by a principal relationship between institutionnal factors and legitimating motivations is explained by the neo-institutional theoretical framework, according to which the fact that the firm’s attributes and normative expectations drive the firm to further maintain its legitimacy [75]. Similarly, the significant relationship between financial performance and economic motivation is based on the instrumental perspective developed by [88].

Secondary links also emerge, such as the one between financial performance and strategic motivations, which is consistent with the vision of shared value creation proposed by Porter & Kramer [38]. These interlinkages deepen our understanding of the processes underlying ESG communication and reveal its multidimensional nature.

The examination of the determinants and motivations of ESG disclosure highlights a complex theoretical structure, presented in. The structural characteristics of the company, its size, age, business sector and governance structure are crucial strategic elements that determine both the capacity and the imperative to communicate on environmental, social and governance issues [56]. The institutional approach offers an additional perspective by showing that isomorphic, coercive, normative and mimetic constraints encourage companies to standardise their ESG disclosure practices within the same organisational domain [75]. This demonstrates the convergent adoption of ESG disclosure strategies in response to institutional expectations (Seow et al., 2024).

Economic, strategic and legitimacy motivations come together to create a coherent system in which signalling theory [51] clarifies the use of ESG disclosure as a tool for reducing capital costs and accessing resources. Furthermore, strategic motivations find their theoretical justification in the Resource-Based View theory [56], which supports the pursuit of sustainable competitive advantages through ESGdisclosure. As for legitimacy motivations, according to legitimacy theory [40] and institutional theory, they represent a mechanism for maintaining social acceptability and managing image.

Conclusion

This study contributes to the literature on ESG disclosure by proposing a two-dimensional taxonomy of the determinants that influence this practice. The proposed framework structures the determinants into three conceptually coherent clusters and associates them with three types of organizational motivations. Using a primarily artificial intelligence-based methodology, combining a systematic literature review with NLP analysis and assisted clustering, we identified the underlying determinant structures with satisfactory statistical robustness (overall silhouette score of 0.60).

Our systematic review, supported by AI, highlighted six clusters of determinants that provide a two-dimensional framework for analysis, with significant practical consequences for various players in the ESG ecosystem. This empirical classification of the elements influencing ESG communication offers company managers the possibility of pinpointing the contextual, organizational and strategic factors best suited to their specific situation, thus optimizing their ESG reporting investment choices by taking into account their size, field of activity, financial performance and governance structure. This segmentation enables investors and financial analysts to adjust their ESG evaluation models by assigning different weights to criteria according to the predominant group of determinants for each company examined, thus improving the accuracy of their forecasts on the evolution of disclosure practices.

Theoretically, this research enriches unidimensional approaches by establishing a framework that integrates not only the determinants of ESG communication, but also its motivational mechanisms. This perspective enriches our understanding of this practice by revealing the privileged links between the different clusters identified.

Like all research, our study has certain limitations that leave the door open for future research. First and foremost, our taxonomy is theoretical and needs to be empirically verified through quantitative research that examines the impact of different combinations of factors and motivations on current disclosure practices.

In an environment where the demand for ESG transparency continues to grow, it is essential for researchers and practitioners alike to understand the factors and motivations that influence such disclosure. Our two-dimensional classification is a step towards a more detailed and holistic understanding of this complex phenomenon, and we hope it will encourage further study in this ever-changing field.