Natural Language Processing (NLP) is increasingly adopted by government for supporting and enhancing citizens' participation and public service co-creation. However, governments that implement NLP technologies like chatbots and sentiment analysis tools for the above purposes should now address critical concerns including ethical implications and sustainability issues. The study investigates NLP uses for supporting and enhancing citizens' participation and public service co-creation and reveals outcomes as well as the ethical, energy, and sustainability challenges posed. In particular, we aim to address two research questions: the first of them concerns the impacts of NLP on public engagement, participation, and co-creation in public organizations, and the second research question addresses the ethical, energy, and sustainability challenges that the use of NLP in public organizations faces. For this purpose, we conduct a systematic literature review based on PRISMA methodology. It is concluded that NLP promises inclusive solutions for public engagement, participation, and co-creation but is hindered by algorithmic bias, data privacy risk, and the environmental impact of large-scale language models.
In the last decade, NLP, which involves artificial intelligence (AI) that enables machines to understand human language, has increasingly been adopted by public services and changes citizens participation and public services’ co-creation [1]. As more governments globally adopt such technologies [2], [3], it is important to understand how they are used as well as their impact. NLP is a technology whereby machines or computers comprehend, manipulate, generate, or respond to human language in relevant contexts [4], [5]. It helps government agencies deliver more personalized and context-dependent services through automatization and enhancement of vast volumes of text-based information, such as citizen inputs [6], [7], policy briefs, regulatory guidance [8], and social media postings [9]. The usage of NLP in public administration refers to the automation of responses to citizens' questions using chatbots, opinion-measuring systems that gauge public opinion, multilingual support systems, and intelligent data platforms that support policymaking [10]. These enhance citizen participation in public decision-making processes, citizen engagement with government 1∗ Corresponding author. † These authors contributed equally. services and information and public service co-creation through the usage of collaborative platforms and other mechanisms such as feedback mechanisms. Such applications include systems with automated responses, sentiment analysis tools for gauging public opinions, multilingual support systems capable of breaking down language barriers and intelligent platforms that support citizens in contributing to design and improvement of services [11].
Citizen participation refers to the involvement of citizens in governmental decisionmaking processes (e.g., policy development and public consultations) [12]. Citizen engagement encompasses the broader interaction between citizens and public organizations, including information seeking, service usage and communication [13], [14]. Finally, service co-creation involves collaborative processes where citizens actively contribute to the design, development and improvement of public services alongside government officials [15], [16].
At the same time, as NLP technologies become more widely used for the above purposes, they raise significant ethical governance concerns, necessitating greater transparency to mitigate algorithmic bias, and sustainability issues due to the ecological cost of these systems[17], [18]. Ethical governance for NLP means eliminating inherent biases in its training algorithms and datasets so that public services could be more inclusive and not reinforce existing social inequalities. Also, large-scale NLP applications, particularly those that use powerful deep learning architectures like BERT, GPT, and other models, consume a considerable amount of energy, raising concerns about their carbon footprint [19]. Sustainable NLP implementation could involve innovations ranging from computational efficiency to anything that optimizes a model and eco-friendly use of data centers.
So, our study aims to address the following two research questions: RQ1: What are the effects of the use of NLP by public organizations on citizens' participation, engagement, and services co-creation? RQ2: What are the key ethical, energy efficiency, and sustainability challenges that the use of NLP in public organizations for these purposes faces? For this purpose, we conduct a systematic literature review based on PRISMA methodology [20-21], as described in the following section.
Our research attempts to provide a systematic assessment of the new NLP applications in these three specific areas, identify challenges in the current approach, and make recommendations regarding future research directions. The final aim is to provide insight for policymakers, technologists, and researchers on how to apply NLP ethically and sustainably while maximizing its potential for transforming public administration into a genuinely responsive, collaborative, inclusive, and green endeavor.
The paper is structured as follows: The next section 2 describes our methodology: the systematic review process we conducted in accordance with the PRISMA guidelines. In sections 3 and 4 the results are presented, while the final section 5 includes our conclusions.
As mentioned in the Introduction we performed a systematic literature review to investigate the role of NLP in citizens’ participation, engagement and public service co-creation, specifically in the light of ethical and sustainability considerations. We adopted this approach as systematic reviews allow for structured and replicable synthesis of existing research on a topic. PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) was used to enforce transparency and methodological rigor; this approach aligns with established guidelines for rigorous literature reviews [20] and recognizes varied review strategies (e.g., 'miner' vs. 'prospector' approaches, [21]). To ensure a comprehensive and inclusive collection of relevant literature, the search was conducted across Scopus and Google Scholar databases. For addressing the RQ1 a Boolean search query was formulated as follows: " ("Natural Language Processing" OR "NLP") AND ("public service" OR "public administration" OR "public sector" OR "public governance") AND ("co-creation" OR "co-production" OR "participation" OR "engagement" OR "consultation")".
For addressing RQ2, we focused on papers within the retrieved set of papers using the above query, which specifically mentioned “ethical” OR “ethics” OR “sustainability” OR “energy” in their content, recognizing that papers addressing ethical and sustainability challenges are a subset of the broader literature retrieved by the main query.
Peer-reviewed journal articles and conference proceedings were searched for recent and related studies for the period from 2017 to date. The only language included in the publications is English to ensure uniformity in analysis. Articles that were purely technical advancements in NLP but had no reference to use and implications in government were excluded. Thus, this directly addresses RQ1 regarding the use of NLP by government in the areas of citizens’ participation, engagement and co-creation, while also setting forth propositions on how ethical and sustainability issues raised in the RQ2 might be approached. Predetermined inclusion and exclusion criteria ensured the selection of valid, relevant studies while maintaining sufficient scope for comprehensive analysis. Excluded were studies unrelated to public sector context, non-peer-reviewed articles, non-English publications, and those lacking full-text access. This screening process ensured both research questions were addressed, focusing on NLP’s impact on governance, inclusiveness, and ethical considerations. Figure 1 outlines the PRISMA-based screening steps.
In the identification phase, our search yielded 112 articles. With no duplicates, all were screened based on titles and abstracts for thematic relevance: 55 papers were excluded for being outside the research scope, leaving 57 for full-text review. During the eligibility stage, these 57 studies were critically evaluated for methodological quality, relevance, and alignment with research questions. Ten were excluded due to methodological flaws, unavailable full texts, or low relevance. In the final inclusion phase, 47 studies met all criteria, focusing on NLP in the areas of citizens’ participation, engagement, co-creation, sustainability, and ethics. Data from each selected study was systematically collected and coded for thematic synthesis. Extraction included bibliographic details (authors, year, publication), NLP application type (e.g., sentiment analysis, chatbots, translation, text mining), and public sector domain (egovernment, participatory governance, policymaking).
For RQ1, we performed a descriptive analysis to map NLP applications and their impact on citizens’ participation, engagement and public service co-creation. Key attributes (year, context, NLP tools, domains, outcomes) were extracted, tabulated, and summarized to identify dominant patterns. For RQ2, we conducted a thematic analysis focusing on the subset of studies within the identified, 25 papers that (as mentioned above) specifically mentioned “ethical” or “ethics” or “sustainability” or “energy” in their content. Study results were coded line-by-line, grouped into descriptive themes, and synthesized into higher-level analytic themes related to ethics, and sustainability. To ensure rigor, multiple researchers coded independently and resolved differences through discussion. This synthesis highlighted overarching concerns such as fairness, transparency, and environmental impact. Together, these analyses offered comprehensive insights linking study features to themes on NLP’s role in governance.
There are many papers describing the use of AI-based NLP applications in government that
aim to enhance citizens’ participation, engagement and co-creation; sentiment analysis is used
to gauge public opinion on the quality of public services. Zhou et al. (2022) [22] applied
pretrained Chinese language models to thousands of policy documents to assess public
support for policy instrumen
In administrative contexts, the DODFMiner NER tool achieved F1 scores of ≈0.78 (text) and 0.85 (entities) on Brazil’s official gazette [31]. NLP was also proposed for extracting definitions and constraints from service-level agreements. Delhi Metro feedback was sentiment-analyzed using an XLNet-BiLSTM model [32], and a recursive neural NLP system improved response accuracy for an electric utility’s customer queries [33]. These studies highlight NLP’s global role in analyzing reviews, social media, and official texts to generate insights and enhance public service delivery. NLP is being used for the development of different methods and technologies such as sentiment analysis, AI-driven chatbots, public feedback analysis, smart cities, policy analysis, legal document parsing, emergency services, and transportation analysis. We have listed the corresponding references for these solutions.
NLP technologies have emerged as disruptive methods concerning citizen participation, involvement, and co-production of services within public administrations. Recent studies demonstrate how NLP instruments can overcome the communication gap between citizens and governments, organize administration, and make democratic participation more inclusive systems.
Wan et al. [38] demonstrated the power of NLP in crisis communication through the analysis of 197,430 messages during the Zhengzhou Rainstorm, building an emotion-behavior framework on ALBERT models to detect citizen emotions (fear, anger, sadness, positive) and behavioral trends (social support provision, seeking help, deviance, avoidance). Their findings demonstrated that NLP-driven analysis converts passive citizens' data into actionable intelligence, such that government communication strategies have varying implications for citizen engagement behaviors, with the most strategies improving social support while responding passively to help-seeking behavior.
Rizun et al. [15], [16] carried out a systematic review of 75 studies on NLP and text analytics in public service co-creation. They identified ten key categories, with machine learning approaches (17.69%), chatbots (16.92%), and sentiment analysis (13.08%) standing out as the most common. The study shows that most NLP applications concentrate on the codesign phase (71.70%), in which they support consultation and the generation of new ideas. Two main benefits are revealed by the analysis: economic value, recorded in just over half of the cases (52.94%) through efficiency improvements, and citizen value (25.88%) achieved through greater transparency and administrative simplification. Two platforms in the real world illustrate the use of NLP for citizen engagement. Dumrewal et al. [39] created CitiCafe that utilizes Latent Dirichlet Allocation (LDA) to categorize complaints with a 90.6% accuracy rate and Conditional Random Fields (CRF) to extract location information with an 84.59% accuracy rate. The system has processed more than 50,000 complaints by citizens and continuously monitors worried social media posts. Similarly, Ingole et al. [40] suggested a chatbot based on BERT, which achieved 85% user satisfaction and 90% response accuracy. It was able to process around 70% of citizen queries independently, reducing the workload of human staff by roughly half.
Son et al. [6] addressed scalability challenges in participatory governance with their BERTbased approach using KR-BERT for online petition categorization within South Korean government platforms. Their approach achieved 76% accuracy in 12 categories of petitions, demonstrating the potential for achieving significantly better performance with the inclusion of historical petition data in models trained on current platform data, pinpointing how domain-specific NLP variations can provide computationally inexpensive solutions for resource-poor government applications.
Language accessibility remains a central challenge for meaningful citizen engagement and to address this, Sangeetha et al. [41] designed a multilingual chatbot framework that integrates dynamic meta-learning with cross-lingual embeddings. In evaluation, the system achieved 92% accuracy for English, 89% for Spanish, and 84% for Luxembourgish and with the capability of real-time adaptation to different linguistic communities, their framework promotes digital inclusivity and illustrates how multilingual conversational AI can function as a practical bridge to more equitable access to government services.
Digital transformation in government services demonstrates significant NLP implementation potential. Alves et al. [42] investigated how digital transformation is unfolding in Brazilian public agencies, focusing in particular on the use of NLP within the judiciary. They highlight applications such as legal document review, case management, and citizen-facing services. Their study shows that Brazilian institutions are adopting a growing number of NLP-driven tools, including platforms for virtual trials, AI-based document processing that enables citizen participation, and natural language interfaces designed to simplify access to government services and according to the authors, these technologies create new opportunities for citizen engagement. At the same time, they point out several challenges that remain, and some of them are to ensure system interoperability, to provide adequate staff training, and to safeguard cybersecurity.
A more recent example is provided by Oksama et al. [24], who developed SI-PEKA, a consultation system built with GPT-4. In testing, the system achieved high acceptance rates, demonstrating that large pre-trained language models can overcome many of the traditional obstacles that have limited the deployment of digital solutions in government contexts.
Morocho et al. [43] conduct a systematic review of 27 studies concerning chatbot development in the health and education domains, examining technologies, frameworks, and evaluation metrics; it demonstrates strong biases towards open-source platforms like Rasa with critical gaps identified in evaluation methods and responsible AI principles.
Spiliotopoulos et al. [44] utilized NLP architectures to engage citizens in legislative processes through semantic search engines and argument mining technologies to process human judgments over open government data. Their human-evaluated results showed that NLP semantic search achieved the greatest extent of citizen engagement, demonstrating NLP's significance in rendering legislative data intelligible for non-expert citizens and achieving effective policy participation.
Rommelfanger [45] examined policymaking in German national parks and found that systematic text analysis can highlight how different stakeholders work together, pointing out, at the same time, that genuine citizen participation depends less on advanced NLP applications and more on fostering open communication and inclusive practices. Additionally, text analysis tools can still add value by helping participation become more meaningful and effective.
The literature reviewed shows that NLP technologies hold considerable promise for reshaping citizen- government interaction through AI-driven content analysis, pattern recognition, and conversational systems. Still, realizing this potential requires addressing challenges such as multilingual support, reliable evaluation methods, institutional resistance, and adherence to Responsible AI principles. Literature shows that NLP-supported platforms of civic engagement ought to balance the technological elements with democratic values so that accessibility remains fair with openness and accountability in public decision-making processes.
It can be concluded that NLP applications in public administrations are moving beyond simple automation to more sophisticated systems that can facilitate productive co-creation of public services, yet institutional and technical barriers remain significant impediments to expansion. Overall, the reviewed studies invariably note that while NLP has the potential to drive participation, engagement, and co-creation, its impact is contingent on institutional openness, thoughtful design, and inclusivity mechanisms. The main uses of NLP methods and technologies extracted from the reviewed literature are summarized in Figure 2.
Despite the potential of AI-based NLP described in the previous section, challenges do
exis
The digital transformation in the public sector is impeded by the digital divide, which restricts the reach of services to citizens and further citizens engagement, participation, and co-creation of the public services. For instance, rural areas continue to face challenges in accessing reliable internet and digital infrastructure [42]. Furthermore, some of the risks come from the algorithmic implementation of the NLP methods as well, as mentioned by the Mellouli et. al. 2024 [5], AI faces risks such as bias in training data, lack of explainability, lack of trust in decisions and policies, prediction mistakes and failures, and Less inclusion. These risks suspect the citizen engagement and co-creation in the public services. In NLP systems, language is considered a major source of developing the system. In some cases, as mentioned by Morocho et. al. [43], languages through which a public service is developed makes the system inclusive or exclusive. Language dependencies have been seen in the chatbot development, and English language-based NLP system, specifically, chatbots for public engagement and participation is dominant. One other sustainability challenge is less availability of open-source tools, most of them are paid, and public and private engagement is difficult in this scenario. There is a lack of global standards for the NLP services assessment is also biggest challenge in sustaining the public confidence[43].
Future research should focus on the exploration of the current strategies to reduce energy costs in training or deploying large NLP systems and establishing multilingual NLP systems for inclusivity and accessibility across different linguistic domains. In addition, any AI governance surrounding public services should be audited for ethical sustainability and, in doing so, should also be careful to emphasize fairness, transparency, and accountability. The wide variety of challenges extracted through the review are shown in Figure 3. These challenges are related to NLP models’ capabilities to understand the language and associated context, data privacy and security, bias and fairness, multilingual support, speech recognition for accessibility of NLP services, meaning extraction from legal documents, sentiment analysis, public feedback processing, and real-time processing. With the advent of transformer architecture in 20172, there has been a noticeable surge in the NLP realm. Transformers are capable of both understanding and generating NLP by their architecture, which is encoder and decoder blocks. The large language models developed using the transformer architecture are capable of solving most of the current problems being faced in the public sector regarding the NLP capabilities. A few examples of these models are ChatGPT, Mistral, DeepSeek, and Perplexity. These models are multilingual and some support multimodal inputs and outputs, enhancing their utility in diverse public service contexts. However, these models require considerable computational resources (memory, processing power, & storage). Most are openly available for adaptation to specific tasks, enabling public agencies to leverage them for improved service applications.
This study reviews a broad range of NLP applications in government for supporting citizens’ participation, engagement and co-creation, such as sentiment analysis, named entity recognition, machine translation, and chatbots, demonstrating their deep integration in policy design, citizen feedback analysis, and administrative efficiency. It has been concluded that NLP plays a vital role in enhancing public service by supporting and enhancing evidencebased decision-making, citizens’ participation, engagement as well as co-creation, leading 2 https://arxiv.org/abs/1706.03762 finally to an increase of service responsiveness. However, large-scale models can perpetuate biases, requiring mitigation. Their high energy demands require greener strategies (model pruning, efficient hardware, cloud optimization). Effective NLP requires strong policies (accountability, transparency, controlled data sharing) and explainable systems to build trust. Interoperability across agencies, interdisciplinary collaboration (policymakers, legal experts, data scientists), and citizen involvement (priority setting and feedback) further enhance NLP’s impact. Since most pretrained models are English-centric, future research should develop inclusive models for local dialects and scripts. Low-cost, energy-efficient NLP solutions are also needed for resource-constrained governments.
The study has some limitations: only English-language literature was reviewed, based on two databases (Scopus, Google Scholar), excluding non-English studies, gray literature, and industry reports. Also, it focuses on the use of NLP in ‘extrovert’ functions of government agencies concerning their interaction with their external environment (citizens’ participation, engagement and co-creation), so it has not examined the use of NLP in internal functions of government agencies for increasing their efficiency and effectiveness.
This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 955569. The opinions expressed in this document reflect only the author’s view and in no way reflect the European Commission’s opinions. The European Commission is not responsible for any use that may be made of the information it contains.
The authors have not employed any Generative AI tools.
[8] V. Ribeiro et al., “Natural Language Processing Applied in the Context of Economic Defense: A
Case Study in a Brazilian Federal Public Administration Agency:,” in Proceedings of the 25th
International Conference on Enterprise Information Systems, Prague, Czech Republic: SCITEPRESS
Science and
Platforms, Chatbots, and Publi
Case Study of Water Consump
Relevance for better digital public services,” 24th Annual International Conference on Digital Government Research (DGO’23). Association for Computing Machinery, New York, NY, USA, pp. 224–232, 2023. doi: 10.1145/3598469.3598494. [47] J. Ignacio Criado, J. Villodre, and A. Meijer, “From experimentation to public service delivery in social media. An analysis of institutionalization dynamics in Dutch local governments,” in Proceedings of the 20th Annual International Conference on Digital Government Research, Dubai United Arab Emirates: ACM, Jun. 2019, pp. 342–352. doi: 10.1145/3325112.3325250. [48] European Commission (High-Level Expert Group on AI), “Ethics Guidelines for Trust-worthy Artificial Intelligence,” European Commission, 2019.