Persuasive Conversational Agents for Environmental Sustainability: A Survey

Persuasive Conversational Agents for Environmental Sustainability: A Survey MATHYAS GIUDICI,Department of Electronics, Information, and Bioengineering, Politecnico di Milano, Milano, Italy PIETRO CROVARI,Department of Electronics, Information, and Bioengineering, Politecnico di Milano, Milano, Italy FRANCA GARZOTTO,Department of Psychology, University of Milan-Bicocca, Milan, Italy and Department of Electronics, Information, and Bioengineering, Politecnico di Milano, Milan, Italy In the next few years, people are called upon to collectively contribute to environmental sustainability, such as mitigating climate change, reducing waste, conserving biodiversity, or promoting sustainable resource management. With this literature review, we are interested in investigating how conversational agents have been used to persuade people toward environmental sustainability behavioral change, and which design features and methods are used in the persuasion process. This field sits at the crossroads of multiple disciplines, including Computer Science, Human-Computer Interaction (HCI), Environmental Science, and Psychology, each contributing unique insights into the design and effectiveness of persuasive conversational agents. The survey proposes a structured report analyzing the current state of the art in persuasive conversational agents for environmental sustainability, considering the multidisciplinary nature of the issue. We explored multiple perspectives, including the conversational agents’ design features, the persuasion strategy adopted, the environmental sustainability issue considered, and the empirical evaluation method (if an empirical study was performed). From the lessons learned, we propose a research agenda to fill the gaps in the field and a checklist to guide future research in persuasive conversational agents applied to environmental sustainability. CCS Concepts: • Human-centered computing → Natural language interfaces;Personal digital assistants;Ambient intelligence; Additional Key Words and Phrases: Survey, conversational agent, environmental sustainability, persuasive technology, human-computer interaction ACM Reference Format: Mathyas Giudici, Pietro Crovari, and Franca Garzotto. 2025. Persuasive Conversational Agents for Environmental Sustainability: A Survey. ACM Comput. Surv. 58, 6, Article 139 (December 2025), 37 pages. https://doi.org/10.1145/3774751 This project has been funded by the European Union’s Horizon Europe Research and Innovation Framework under grant agreement No 101160720 and by the Italian Ministry of University and Research (MUR) and the European Union (EU) under the PON/REACT project. Authors’ Contact Information: Mathyas Giudici (corresponding author), Department of Electronics, Information, and Bioengineering, Politecnico di Milano, Milano, Lombardy, Italy; e-mail: [email protected]; Pietro Crovari, Department of Electronics, Information, and Bioengineering, Politecnico di Milano, Milano, Lombardy, Italy; e-mail: [email protected]; Franca Garzotto, Department of Psychology, University of Milan-Bicocca, Milan, Lombardy, Italy and Department of Electronics, Information, and Bioengineering, Politecnico di Milano, Milan, Lombardy, Italy; e-mail: [email protected]. This work is licensed under a Creative Commons Attribution 4.0 International License. © 2025 Copyright held by the owner/author(s). ACM 0360-0300/2025/12-ART139 https://doi.org/10.1145/3774751 ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. 139:2 M. Giudici et al. 1 Introduction Conversational Technologies, which enable users to interact with digital tools through natural language [ 108 , 140 ], are getting more and more pervasive. They are embedded in smart physical devices (such as Alexa and Google Home) and integrated into various software applications. Conversational Technologies are currently used in many contexts of everyday life [ 150 ], such as they empower people to buy something on their preferred e-commerce website [ 119 ], listen to music [ 123 ], follow a cooking recipe [ 7 ], and take notes just using voice [ 97 ]. Conversational Agents are also widely used in phones’ operating systems or applications (like Siri, the Apple virtual assistant) [ 70 ], cars [ 97 ], and customer service [ 66 ]. Additionally, the emergence of Large Language Models (LLMs) in AI has transformed several conversational technology applications, and also unlocked chatbots for code development, story and creative generation, and text completition and summarization [138]. Our research focuses on conversational technology used to persuade people to change behaviors, and we can refer to this category of digital tools as Persuasive Conversational Agents (PCA). PCAs are effectively used to lead people to adopt healthier habits, such as committing to regular exercise [148], reducing snacking [88], or limiting alcohol consumption [100]. PCAs are also a promising interaction paradigm toward improving people’s awareness and effectiveness in adopting environmentally sustainable behaviors [ 73 ]. However, their persuasive use for environmental sustainability is quite a novel research area, and a clear literature landscape on PCA is still missing. The Intergovernmental Panel on Climate Change argues with about 95% certainty that climate change is anthropogenic [ 109 ]. The urgent need for a green turnaround for all nations on the planet has been highlighted in many contexts, and world leaders at the Sustainable Development Goals (SDG) summit in September 2019 defined the 2020s as a Decade of Action, in which nations have to collaborate in the development of scientific knowledge and cutting-edge tools to achieve sustainability goals [ 48 ], pinpointing that everyone must implement sustainable behaviors and contribute as individuals to the “green” collective effort. Our survey focuses on the human-centered design and evaluation of this class of systems and provides a review of the current state of the art contributing to the Human-Computer Interaction (HCI) field. The survey has been carried out by investigating five research questions posed to the articles involved in the analysis: R1 Which High-Level Design Features are addressed (e.g., sustainable topics covered, persuasion strategies adopted, and underlying theories)? R2 What UX Design Features are adopted in Persuasive Conversational Agents for environmental sustainability (e.g., interaction modes, agent embodiment, personalization capability, ambient awareness)? R3 Which is the Participants Profile in the Empirical Evaluation of Persuasive Conversational Agents for environmental sustainability? R4 Which is the Study Design in the Empirical Evaluation of Persuasive Conversational Agents for environmental sustainability? R5 What are the Lessons Learned on Persuasive Conversational Agents for environmental sustainability? This article presents the literature review process and results performed to answer the above questions. We followed a systematic method to gather and extract relevant data, reducing any possible biases during information collection. We quantitatively and qualitatively analyzed and reported the results, exploring and discussing the results on the different research variables. ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. Persuasive Conversational Agents for Environmental Sustainability: A Survey 139:3 The major contributions presented in this work are the following: — A systematic review of Persuasive Conversational Agents for environmental sustainability clustered into different dimensions (i.e., persuasion strategies, sustainable topics addressed, UX and conversational design features, and approaches to empirical evaluation) — A discussion of the results and research gaps on the different variables identified from the research questions — From the lessons learned by the studies surveyed, we derived – A set of actionable recommendations that identify open research gaps to be filled by future development and investigation in such category of Persuasive Conversational Agents. – A reporting checklist for future works on Persuasive Conversational Agents for environmental sustainability that wants to guide researchers in reporting the variables we consider particularly necessary and useful to design, develop, and report such agents. The article is organized as follows: Section 2presents a background overview of persuasive technology, conversational agents, and environmental sustainability, drafting the conceptual framework and research questions used to evaluate the under-review articles. Section 3presents the systematic procedure followed. Section 4is the core of the article, reporting the result from the analysis based on the 52 surveyed articles (from 2009 to January 2025). Section 5delves into the discussion on the results, while Section 6presenting a research agenda and a reporting checklist. Section 7outlines the current limitations of this survey. Finally, Section 8draws conclusions. 2 Research Context Our inquiry lies at the intersection of three primary areas: Environmental Sustainability, Persuasive Technology, and Conversational Agents (see Figure 1). For each area, we have identified key aspects of interest, focusing on those that are particularly relevant both within each specific domain and across the broader field of Human-Computer Interaction. This approach also reflects themes emerging from related literature surveys [1,2,25,36,129,151]. The conceptual framework has guided the formulation and structure of our research questions, as well as the analysis of the reviewed articles. Before delving into these aspects, we first define the main terms used in this review to prevent potential misunderstandings arising from differences in terminology across disciplines. 2.1 Background and Terminology We define persuasion as modifying another subject’s attitude or behavior while exchanging messages or chatting [ 31 ], inducing a person to recognize the reality of a fact [ 78 ]. In addition, the persuasion process should not deceptively impose a change in human beings but push people to reflect on improving their behaviors for a specific value (i.e., persuasion goal). In the particular case of this survey, the value of enhancing is identified in environmental sustainability. According to Crano and Prislin [33] , the persuasion goal could be achieved with spontaneous or controlled processes. The former is a process directly triggered by self-awareness by people for change and strong motivation, while the latter is a procedure that influences people’s actions to achieve a final intent. In the context of persuasion with controlled processes, in 1991, Ajzen [3] proposed the Theory of Planned Behavior (TPB). TPB provides a conceptual framework for dealing with the complex nature of human social behavior, presenting the basis for understanding and anticipating actions performed by people in specific settings. In addition, TPB provides as its main outcome that controlled persuasive models working for people’s behavioral change jointly depend on motivation (i.e., people’s intention) and ability (i.e., behavioral control). ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. 139:4 M. Giudici et al. Fig. 1. Conceptual framework and related research questions. Still, Prochaska and DiClemente [137] suggested a behavioral change model (called Transtheoretical Model or Stages of Change Model) that implies different phases in a circular process. Motivation is a continuum looping starting from Pre-contemplation, in which the user is unaware of the change, and motivation must be built. Contemplation is the phase related to the users’ awareness of the change, followed by Preparation, where strategies for the behavioral change are set. Change occurs in the Action phase, in which a new lifestyle is embraced. This new behavior is sustained in the Maintenance, which could be interrupted by the Relapse, in which regression into the old behavior occurs. According to the self-blame, the user could restart the cycle or leave it completely. Finally, Michie et al . [114] proposed a hierarchically structured taxonomy of Behavioral Change Techniques (BCTs), setting a systematic and reliable specification of persuasive strategies involved in behavioral change interventions. They identified 93 possible BCTs grouped into 16 macro areas, which are Goal and planning, Feedback and monitoring, Social support, Shaping knowledge, Natural consequences, Comparison of behavior, Associations, Repetition and substitution, Comparison of outcomes, Reward and threat, Regulation, Antecedents, Identity, Scheduled consequences, Selfbelief, and Covert learning. Among the different persuasive strategies clustered by Michie et al . [114] , we report a short description of the ones relevant to the current manuscript. The shaping knowledge strategy in environmental sustainability aims at raising awareness and enhance understanding of environmental issues by addressing knowledge gaps and – in general – educating on topics like climate change, food waste, water usage, and others. Feedback involves providing individuals with information about their environmental impact in a way that is accessible and actionable (e.g., providing information on user energy usage). This strategy helps make abstract information into more meaningful ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. Persuasive Conversational Agents for Environmental Sustainability: A Survey 139:5 metrics, motivating behavior change through real-time or periodic updates. Similarly, and usually coupled with feedback, goal setting encourages positive environmental behaviors by prompting individuals to commit to specific, measurable targets. This strategy relies on the psychological principle that clearly defined goals improve focus and persistence. Finally, the social comparison strategy leverages individuals’ sensitivity to social norms by showing how their behavior aligns with that of others (e.g., the neighborhood) and motivates behavioral adjustment. The milestones – described above – in psychological literature informed the Persuasive Technology (PT), defined by Fogg [55] in 2002, as technology designed to change the attitudes or behaviors of the users through social influence and persuasion techniques provided by digital tools. In the interaction with PT, effective persuasion, as in human-human interaction [ 167 ], is usually achieved by using customized messages and specific strategies to target better the person to convince. However, thanks to advanced machine learning and data science techniques, persuasion can also be achieved using algorithms that automatically fit messages according to the final user’s motivation and capabilities [23,37,88]. In the health field [ 92 ], several persuasive digital systems are particularly relevant in the literature. For example, there are systems influencing positive eating behaviors by avoiding snacking [ 88 ], smoking cessation [ 9 ], and physical activities [ 6 ]. Another intensely explored area is educational persuasive technology to improve motivation in reading and writing in children [ 103 ] or to encourage students to achieve milestones and study deadlines in a course [67]. Environmental sustainability is the practice of utilizing natural resources in a way that protects the environment for future generations [ 96 ]. This includes preserving biodiversity, lowering pollution, and using less energy and water. Individuals and organizations can limit their environmental effects to achieve environmental sustainability [ 46 ]. Examples are substituting fossil fuels with renewable energy sources (e.g., solar and wind power), reducing waste and water usage, using electric or public transportation, and supporting energy conservation programs. In 2015, United Nations Member States signed the 2030 Agenda for Sustainable Development [ 42 ]: a unifying framework for peace and prosperity for people and the planet, both for now and the future, clustering world issues into 17 goals. Among these goals, topics such as combating climate change and fighting to protect oceans and forests represent some of the environmental sustainability issues that must be addressed in the next decade. Different strategies to induce more environmentally sustainable behaviors have been explored in recent years. For example, several studies [ 32 , 74 , 173 ] showed that the real-time visualization of domestic energy consumption leads to more responsible consumption, especially over long-term exposure [ 84 ]. Other works [ 8 , 117 , 173 ], instead, investigated the ability of persuasion techniques in HCI, such as self-monitoring, advice, rewards, and gamification, to lead to more sustainable behaviors. In addition, according to [ 47 , 152 ], in the Human-Computer Interaction field, there is a growing interest in sustainability; several challenges must be addressed, and numerous issues must be well-considered and tackled [18,94]. Conversational Agents (CA)are a promising persuasive technology for environmental sustainability [ 61 , 144 ]. We define Conversational Agents as the set of technologies that interact with users through natural language [ 108 , 140 ]. In the current literature, terms such as CA, conversational technologies, chatbots, agents, and dialog systems are often used as synonyms [ 56 ]. Conversational agents can be classified according to different criteria. Sometimes, they are categorized according to the channel they use to interact with people: voice (speech-based CA) or textual messages (text-based CA). Another taxonomic method involves distinguishing between different implementation approaches [ 80 ]. Conversational technologies have evolved from rule-based systems – exemplified by one of the first conversational agents, ELIZA [ 168 ] – to data-driven approaches that leverage ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. 139:6 M. Giudici et al. traditional machine learning models and, more recently, transformer-based architectures and LLMs [ 164 ]. The advent of transformer-based agents has empowered the conversational interaction capabilities of CAs – overcoming the lack of flexibility and scalability of traditional rulebased systems [ 158 ] and enabling more sophisticated features such as translation, summarization, problem-solving, or creativity [ 138 ]. However, the usage of LLM-based conversational agents raises unpredictable practical and ethical concerns, including bias, misinformation, hallucinations, and privacy issues [172]. Finally, an additional classification attribute for CAs is their embodiment, i.e., the form in which conversational agents are presented. Embodiment ranges from physically actuated robots [ 41 ] to purely digital virtual assistants [ 57 ], as well as IoT devices (e.g., Alexa, Google Home) that can be integrated into home automation ecosystems [112,150]. CA has been proven to influence human behavior in various ways, including increasing adherence to medical treatment, as reviewed by Gentner et al . [58] and influencing consumer behavior [ 83 ]. For instance, van Baal et al . [163] explored an AI chatbot’s ability to deliver COVID-19 protective behavioral interventions, demonstrating an increased likelihood of participants getting tested for COVID-19 symptoms. Similar results were previously found by Miner et al . [118] , which tested a chatbot encouraging the adoption of public health guidelines. Still, Tsai et al . [160] investigated how chatbots and their persuasive communication strategies impact consumer engagement and brand awareness while navigating an online website of a company. 2.2 Research Questions The different themes emerging from the Conceptual Framework depicted in Figure 1provide the classification dimensions and sub-dimensions to organize our space of analysis into four main research questions and to structure a final orthogonal research question focused on “lessons learned”. R1 Which High-Level Design Features are addressed (e.g., sustainable topics covered, persuasion strategies adopted, and underlying theories)? This research question examines three key pillar design dimensions of our survey. First, it investigates the sustainability topics addressed, using SDGs as a framework to understand environmental priorities and critically assess the sustainable topics covered by the works, as SDGs are currently the global standard for tracking sustainable development progress [ 43 ]. Second, it explores the persuasive theories that informed PCA development, recognizing that the application of behavioral theories enhances conceptual rigor and enables hypothesisdriven evaluation of PCAs. Third, it analyzes the specific persuasion strategies adopted, since their selection plays a critical role in how users process persuasive messages [ 28 ], with different strategies being systematically categorized using hierarchically structured taxonomies like the one proposed by Michie et al . [114] to map the different strategies adopted by the analyzed articles. R2 What UX Design Features are adopted in Persuasive Conversational Agents for environmental sustainability (e.g., interaction modes, agent embodiment, personalization capability, ambient awareness)? This research question examines the user experience design characteristics of PCAs across multiple dimensions. It investigates the input and output interaction modes (speech or text), recognizing that the communication channel influences cognitive load, accessibility, and affective response [ 104 ]. The question also explores initiation methods, distinguishing between passive agents that wait for wake-up actions and proactive agents that anticipate user needs, noting that while proactive systems can increase engagement, they may risk perceived ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. Persuasive Conversational Agents for Environmental Sustainability: A Survey 139:7 intrusiveness or reduce user agency [ 13 ]. It analyzes agent embodiment, as the representation of conversational agents–whether visual, physical, or virtual–affects anthropomorphism and can foster engagement or provoke uncanny perception [ 38 ], with different shapes carrying symbolic meanings that affect user reactions. The question addresses gender representation, since digital agents often reproduce social stereotypes and influence user preference and trust [ 49 , 125 ], making it important to critically assess these design decisions. It examines response generation methods, comparing rule-based systems that offer predictability with generative models that provide flexibility and naturalism [ 80 , 113 ]. The question investigates personalization features extracted from users (such as sentiment analysis or emotion recognition), as adaptive agents incorporating user emotion or personality data can significantly enhance persuasion [ 89 ]. Finally, it explores ambient awareness capabilities, examining whether PCAs can perceive external environmental phenomena relevant to sustainability, such as energy consumption and user interactions with home appliances, leveraging the growing integration of conversational agents with IoT and energy data systems [91,150]. R3 Which is the Participants Profile in the Empirical Evaluation of Persuasive Conversational Agents for environmental sustainability? This research question examines the demographic characteristics of participants in empirical evaluations of PCAs for environmental sustainability. It considers sample size as a critical factor, since it directly impacts statistical validity and inferential strength, influencing the generalization of results [ 21 , 98 ]. The question also investigates participants’ age distribution, recognizing that age-related differences in technology adoption, cognitive style, and ecological concern are documented in the scientific literature [ 26 , 130 ]. It analyzes gender composition, as it may affect receptivity to persuasive cues of the technology [ 130 ]. Additionally, it examines participants’ educational level, acknowledging that education influences sustainability literacy and critical thinking, potentially moderating the cognitive processing of persuasive messages [86]. R4 Which is the Study Design in the Empirical Evaluation of Persuasive Conversational Agents for environmental sustainability? This research question examines the methodological characteristics of empirical evaluations of PCAs for environmental sustainability. It investigates the nature of the prototype in PCA behavior, distinguishing between fully autonomous, partially autonomous, and fully humancontrolled systems, as clarifying the level of automation ensures transparency regarding system capabilities and experimental validity [ 12 , 98 ]. The question analyzes the research design methods employed, recognizing that experiments allow researchers to determine causal relationships that can yield significant scientific insights [ 21 , 98 ]. It examines study duration as a critical factor that influences various aspects of user experience and usability [ 159 ]. Additionally, it explores the data collection techniques utilized, acknowledging that within HCI, different methodologies exist to collect data (such as self-reports and interaction logs) which provide a richer understanding of participant engagement [21,98]. R5 What are the Lessons Learned on Persuasive Conversational Agents for environmental sustainability? Synthesizing lessons learned on PCA for environmental sustainability with a more qualitative approach helps to consolidate and ground the field’s progress in order to identify underexplored avenues for future works. Such analysis could be performed on the three main dimensions of the current survey (i.e., PCA design, persuasion, and technological approach). ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. 139:8 M. Giudici et al. 3 Method We performed a complete analysis of the scientific literature to address the research questions of how conversational agents are used in persuading people for environmental sustainability behavioral change and which design features and methods are used in the persuasion process. Inspired by the work of [ 25 , 151 ] and the guidelines by Nightingale [128] to perform a systematic literature review, we conducted this survey following the PRISMA workflow [ 132 ]. In the following sections, eligibility criteria, selection process, and query performed are detailed. 3.1 PRISMA Method Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [ 132 ] is a well-known approach in the literature for systematic reviews and meta-analyses. In fact, it allows researchers to demonstrate the review’s quality, enabling experimental replications and organizing the manuscript with headings and contests in a structured way that is easy for the reader to find. In addition, readers are enabled to evaluate the study’s advantages and disadvantages easily [ 132 ]. As depicted in Figure 2, the PRISMA method consists of five steps. The first phase is the identification , where the articles are gathered from different sources. In the next phase, the screening, duplicates are removed, and articles are filtered according to their eligibility based on titles and abstracts. The eligibility phase analyzed the previously extracted articles, looking at their entire full text and the eligibility criteria. Finally, in the inclusion phase, all the articles selected are deeply analyzed, reporting their results in Section 4. 3.2 Eligibility Following the guidelines pointed out by Kitchenham et al . [93] and the research questions, we surveyed the articles to include the ones reporting on work that intersects conversational agents interaction, persuasive technology, and environmental sustainability. Thus, articles were included in the present survey according to the following eligibility criteria: — They addressed the query based on title, abstract, and keywords described in the Section 3.3; — they explicitly declared to apply persuasion with one or multiple persuasion strategies; — they addressed environmental sustainability goals; — the interaction occurs with a conversational agent; Articles were excluded in the case: — They were published before 2002; — they were published after the day of the query running (i.e., after February 1𝑠𝑡, 2025); — they are inaccessible to the authors; — they do not report quantifiable details on the conversational agent features, persuasion, and sustainability topic to assess the research question. 3.3 Query The method used to collect the set of articles included in the initial dataset of the survey grounds on the terms used in their title, abstract, and keywords. The collection of keywords used in the search aims at targeting conversational agents, persuasion, and environmental sustainability. In order to address the technological component of the survey, we selected terms such as conversational technology, dialog system, chatbot, conversational agent, embodied conversational agent, virtual agent, social robot, intelligent personal assistant, and large language models. Inspired by the taxonomy provided by Michie et al . [114] , to target the possible persuasion strategies, we used terms such as goal setting, planning, feedback, monitor, social support, shaping knowledge, natural consequences, comparison of behavior, associations, repetition, substitution, comparison of outcomes, ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. Persuasive Conversational Agents for Environmental Sustainability: A Survey 139:9 reward, threat, regulation, identity, self-belief, covert learning, social comparison, and awareness; in contrast, persuasion, behavioral change, and nudge as broad terms. Finally, the terms observed as familiar in literature for the topic issue have been identified as environmental sustainability, sustainable development, food management, water management, sustainable mobility, eco-system, food sustainability, water sustainability, energy consumption, and energy management. We carried out the search on Scopus, IEEE Explore, ACM Library, and Web of Science since they are relevant databases in the Computer Science field. Scopus and Web of Science are among the largest datasets containing research items from different fields and venues, while ACM and IEEE organizations have published relevant works in the human-computer interaction field. The different queries have been adjusted to fit the search engine syntax, and the terms listed above have been adjusted to include alternatives and pluralizations. For example, we report the query run on Scopus: TITLE-ABS-KEY ( ("conversational technolog*" OR "dialog* system*" OR "chatbot*" OR "conversational agent*" OR "embodied conversational agent*" OR "virtual agent*" OR "social* robot*" OR "intelligent personal assistant*" OR "LLM*" OR "large language model*" OR "generative AI") AND ("persuasi*" OR "behavior* chang*" OR "nudg*" OR "goal* setting*" OR "planning" OR "feedback*" OR "monitor*" OR "social support" OR "shaping knowledge" OR "natural consequence*" OR "comparison of behavior" OR "associations" OR "repetition*" OR "substitution*" OR "comparison of outcomes" OR "reward*" OR "threat*" OR "regulation*" OR "identit*" OR "self-belief" OR "covert learning" OR "social compar*" OR "awar*") AND ("environment* sustainab*" OR "green*" OR "ecologic*" OR "eco" OR "climat*" OR "sustainable development" OR "food manag*" OR "water manag*" OR "sustainab* mobility" OR "eco-system*" OR "food sustainab*" OR "water sustainab*" OR "energy consumpt*" OR "energy manag*") ) We exported the query results in comma-separated value (CSV) files and merged them into a spreadsheet. We manually included a set of possible topic-related articles that were not in the outcome of the query results but considered relevant based on the authors’ prior domain expertise and literature scraping. After this set expansion, we removed duplicated documents. 3.4 Selection After the initial screening (that removed articles not written in English, published before 2002, and not classified as scientific articles), the remaining 544 articles were analyzed twice on the eligibility criteria (previously defined in Section 3.2), from two different perspectives. The former was high-level access to the criteria based on the title, abstract, and keywords of the works, which led to the exclusion of 356 articles. The latter was an in-depth analysis performed by reading the full text of works considered eligible in the former analysis and performing the backward and forward reference chaining investigation; in this phase, 37 more articles were excluded, leaving 52 studies included in the systematic literature review. Based on the aforementioned process, in both phases, two independent reviewers performed the analysis. To determine if the two reviewers performed consistently, the inter-rater reliability was computed using Cohen’s Kappa statistic [ 29 , 111 ] on the entire set of articles, resulting in a final value of 0.94. In cases of disagreement between the two reviewers, a third reviewer was consulted to make the final decision about the inclusion or exclusion of the article. ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. 139:16 M. Giudici et al. Fig. 10. Empirical Validation Methodology (R4.2) distribution. Fig. 11. Methods of data collection (R4.4) distribution. the scenes. This method is used to test and refine prototypes, particularly those involving AI and complex interactions like Conversational Agents, allowing researchers to gather feedback on design and usability before developing the actual technology. To assess the methodologies used in the design of the studies, we framed what was reported in articles into well-known methods in the literature [ 21 , 98 ]: comparative evaluations (i.e., between or within designs), case reports, and focus groups, graphically reporting results in Figure 10. 17 articles (that included a study) performed a randomized data collection, using a between or within design with fixed factors. Of the remaining articles, 22 studies reported data about a specific case study. In these cases, the sample size of users who tried the conversational agent is small and usually involves experts from a specific sector. Finally, in two cases, a focus group was performed. As regards the data collection methods, grouping was performed, referring to the structures used in the works: interaction and questionnaires, quantitative, and observations. The results are reported in Figure 11. 27 studies were conducted with a data collection method summarizable in interaction and questionnaire: during the experimentation session, first, users interacted with the agent, and then a post-questionnaire to evaluate attitude (persuasion effectiveness) and user experience. Four articles, instead, collected quantitative data about users’ energy consumption. Three articles used qualitative approaches to analyze user interaction with the agents. In addition, four more articles used mixed approaches integrating qualitative insights with quantitative data from energy consumption and logged data generated from user interaction with the application. Finally, two articles organized data collection by taking notes during the observation of the experience. Most of the works used self-developed questionnaires (reporting the items in the material of the study). Sometimes such custom questionnaires were supported or completely based on well-known literature works, such as the Unified Theory of Acceptance and Use of Technology (UTAUT) [ 165 ], Theory of Planned Behavior (TPB) Questionnaire [ 4 ], Parasocial Interaction (PSI) [ 77 ], System Usability Scale (SUS) [ 16 ], perceived humanness [ 76 ], and perceived persuasiveness [ 99 ]. 22 studies report the time duration of the experimentation. 14 works (64%) are short-term, and the experiments last from 30 to 95 minutes. The remaining three articles have longer-term studies conducted in more ecological settings (i.e., testing of their digital tools in the field and not in a laboratory setting). Six studies [ 5 , 54 , 75 , 82 , 106 , 124 ] have a duration in the order of weeks or a month, Suanpang and Pothipassa [154] with a duration of six months, while Bourgeois et al . [14] performed an extended study with a duration of eight months. To sum up, the analysis of the Study Design of the surveyed articles pointed out that conversational agents persuading people to have more sustainable behaviors were mainly tested using single sessions and interaction with the agents to contribute with a case report. The data collection method preferred is questionnaires after interaction with the agent, and sessions usually last ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. Persuasive Conversational Agents for Environmental Sustainability: A Survey 139:17 from 30 to 95 minutes. Finally, researchers preferred to make users interact with a working prototype rather than using a Wizard of Oz technique. 4.5 R5: Lessons Learned In this section, the results extracted by analyzing the R5 research variable with a more qualitative approach that directly maps into the variables are reported. The study’s findings are reported in the subsequent sub-sections, which are divided according to the conceptual framework outlined in Section 2.2. Thus, the sub-sections are Environmental Sustainability,Persuasion, and Conversation and Technology. 4.5.1 Environmental Sustainability. Aligned with the descriptive findings discussed in Section 4.1, a recurring sustainable topic addressed is energy conservation. Multiple studies have found that persuasive feedback, particularly when social or evaluative in nature, results in more significant energy savings than factual feedback alone. Negative social feedback from robotic agents can significantly reduce household electricity usage during appliance operation, emphasizing the importance of affective and socially contextualized cues [ 71 , 72 , 115 , 166 ]. Similarly, PCAs that provide personalized, real-time feedback have demonstrated potential in household energy management by increasing awareness and encouraging more efficient use of appliances [ 27 , 40 , 75 , 90 ]. More specifically, with energy saving directly linked to the building environment, PCAs are recognized to induce energy savings by promoting practices such as turning off unused devices or moderating heating [ 87 , 139 , 170 ]. However, according to de Vos et al . [40] , automating actions (e.g., via smart thermostats) without addressing users’ intrinsic motivations or environmental awareness is insufficient for lasting behavioral change. In the domain of mobility, PCAs have also shown potential to shift user preferences toward environmentally friendly transport. Notably, chatbots that promote shared mobility options such as public transit or e-bikes can positively shape normative beliefs and intentions, contributing to sustainable urban transport systems [ 45 , 95 , 107 ], highlighting the PCAs’ capability to bridge the gap between infrastructural offerings and user behavior, particularly in the recent context of smart cities. Examining food sustainability, chatbots and recommendation engines informed by sustainability metrics, such as carbon or water footprint, have been used to guide consumers toward lower-impact dietary choices [ 69 , 135 , 174 ]. Encouraging plant-based diets or decreasing food waste is viewed not only as a resource conservation issue but also as a behavioral challenge that may be addressed persuasively through digital interventions. In waste management and recycling, educational agents and gamified systems have effectively promoted sustainable behaviors, particularly among children and younger demographics. Social robots and chatbots used in serious games or classroom settings have led to increased recycling knowledge and positive behavioral intentions [24,54,105]. A broader implication across studies is that emotional engagement, personalization, and social contextualization significantly enhance the impact of sustainability-oriented PCAs. Systems that integrate care-based design [ 11 ], community values [ 10 ], or shared cultural narratives [ 127 ] tend to promote more durable and meaningful behavior change. However, studies warn that short-term interactions lack real-world context and can undermine the long-term effectiveness of PCAs in promoting sustainability [ 60 , 61 ]. Also, challenges of interpretability, trust, and factual dependability in AI-generated material for sustainability underscore the need for expert validation and ongoing refinement [22,59,161]. 4.5.2 Persuasion. Among the different findings, it seems there is a consensus on the effectiveness of social feedback over factual feedback across various domains. Studies such as [ 71 , 72 , 166 ] consistently show that PCAs providing social evaluative feedback – particularly negative social ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. 139:18 M. Giudici et al. feedback – tend to be more effective in altering behavior than those delivering neutral or factual information. This pattern holds in both robotic and virtual agents, where evaluative cues (e.g., facial expressions, tone of voice, verbal reactions) amplify the perceived relevance and urgency of the message, thereby increasing compliance with environmentally desirable behaviors. Embodiment and anthropomorphism also emerge as critical factors that enhance persuasive potential. Multiple studies [ 24 , 147 , 166 ] suggest that the physical presence of a robot or human-like conversational features increase perceived social presence, trustworthiness, and credibility, which in turn bolster the PCA’s persuasive impact. In particular, embodied agents were found to be more effective in exerting social influence, especially in contexts involving children or where emotional bonding was relevant [11,105]. Personalization is widely recognized as a beneficial design strategy in PCA-mediated persuasion. [ 10 , 19 , 65 ] show that tailoring feedback and recommendations to the user’s knowledge, preferences, and behavioral stage enhances both receptivity and perceived relevance, thereby supporting behavior change. Latest technologies based on LLMs even pushed the boundaries for personalized interaction, creating agents that could automatically adapt to users’ preferences and behavior. However, as highlighted by Majid et al . [107] , the perceived intrusiveness of personalization can sometimes undermine trust and credibility, indicating the need for careful contextualization and user-centered design. The persuasive effect of PCAs is further supported by integration with motivational theories and behavior change models. Applications grounded in Motivational Interviewing (MI) and the Trans-Theoretical Model (TTM) [ 19 , 174 ] demonstrate enhanced efficacy, particularly when agents express empathy, elicit “change talk”, and support self-efficacy. This design approach was especially effective for male users when MI techniques were applied in dietary behavior change scenarios [174]. Conversational style and multimodality also contribute significantly to persuasion. For instance, [ 11 , 75 , 90 ] indicate that multimedia presence (e.g., voice, avatar presence) and emotionally resonant dialogues foster greater emotional attachment and responsiveness on the final users. Notably, emotionally positive strategies in conversational interaction were found to be more persuasive than purely rational messaging, especially among users with lower pre-existing environmental awareness, as argued by Berney et al. [11]. Still, strategic timing and contextual integration of PCA feedback, as highlighted in Bourgeois et al . [14] , influence its effectiveness. Feedback that is timely, actionable, and situated at the point of decision-making yields greater behavioral adherence than delayed or retrospective feedback. Relatedly, studies underscore the importance of perceived autonomy and user control [ 5 , 40 ], showing that less intrusive, user-initiated interactions are often preferred and more persuasive. Interactive features and gamified elements, while enhancing engagement, do not always directly correlate with behavioral outcomes. Yamawaki et al . [171] found increased user vanity and perceived interaction with an agent, yet no significant behavioral changes in terms of donation or water use. Conversely, in studies such as [ 62 , 105 ], the combination of gamification, feedback, and physical or visual interaction did support sustained engagement and attitude shifts, particularly in younger demographics. Finally, even if it was not one of our primary points of focus for the survey, the role of trust, credibility, and source reliability is repeatedly emphasized. Users are more likely to be persuaded by PCAs perceived as trustworthy, accurate, and aligned with user values [10,95,107]. 4.5.3 Conversation and Technology. The reviewed articles present an expansive set of insights on the design and development of Persuasive Conversational Agents, specifically emphasizing their dialogic design, system architectures, modality choices, and implementation contexts. A central ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. Persuasive Conversational Agents for Environmental Sustainability: A Survey 139:19 theme across these contributions underscores the importance of Natural Language Processing and Large Language Models as technological enablers of adaptive, personalized, and responsive communication. For instance, systems built upon GPT-based architectures demonstrate the potential for real-time, context-aware dialogue, whether in domestic energy management [ 170 ], sustainability education [175], or innovative tourism platforms [154]. Multimodal interaction emerges as a particularly salient feature, enhancing the persuasive efficacy of PCAs by integrating text, voice, and visual modalities. Social robots like iCat and Pepper, as shown in [ 10 , 71 , 72 ], exploit embodied communication, combining verbal cues with facial expressions and gestures. These multimodal capabilities contribute significantly to believability and user engagement, particularly in settings involving children or users with low technical literacy. Similarly, integrating graphical dashboards and avatars in systems (e.g., like Leafy proposed by Giudici et al . [62] ) or Smart Mirror-based agents demonstrates how visual components can augment user understanding and behavioral motivation, echoing previous work in the literature [ 32 , 136 , 149 ]. Several implementations, such as those documented by [ 40 , 61 ], highlight the tension between initiating conversations (proactivity) and responding to user input (reactivity), with implications for user acceptance and perceived intrusiveness. Timing of interventions, as emphasized by Majid et al . [107] , plays a pivotal role in influencing behavioral outcomes, suggesting the need for fine-tuned conversation triggers aligned with user routines and contextual cues. In terms of system architecture, modularity and integration with existing platforms are recurrently identified as key design principles. Studies such as [ 27 , 63 , 87 , 131 ] present architectures combining NLP engines, IoT middleware, and cloud services to facilitate seamless, scalable interactions. RAG strategies are gaining traction, enabling PCAs to connect LLM outputs with real-time external data sources dynamically [ 17 , 170 ], thereby improving response relevance and trustworthiness. The embodiment and anthropomorphization of PCAs further modulate user perception. Humanlike features – ranging from avatars and voice modulation to personal greetings – contribute to social presence and trust [ 45 , 147 ]. However, other findings [ 71 , 72 ] suggest that the effectiveness of social feedback might not necessarily depend on the robot’s perceived agency but rather on the consistency and relevance of the feedback provided. Finally, context-aware agents, informed by sensor data or an appropriate user modeling and behavioral analysis, deliver more persuasive and relevant feedback [ 75 , 124 , 161 ]. These systems tailor their outputs based on individual behavior, preferences, or environmental conditions, enhancing user engagement and behavioral alignment with environmental sustainability. 5 Discussion The surveyed literature underscores several prominent research gaps at the intersection of persuasive technology, HCI conversational interaction, and sustainability, particularly in relation to the three main dimensions under analysis (i.e., sustainability topics addressed, persuasive effectiveness, and conversational interaction). These gaps will be critically discussed in this section, and they are functional to shape the research agenda and guidelines for future work in the field, both summarized in the following section (i.e., Section 6). A recurrent gap in the articles lies in the evaluation of persuasion effectiveness and longterm behavioral and attitudinal changes in users’ positive environmental behavior. behaviour. Many studies develop prototypes or demonstrate short-term effects, but few examine long-term behavioral impact or conduct longitudinal evaluations to assess the sustained influence of persuasive technologies on pro-environmental attitudes [ 10 , 24 , 45 , 60 , 69 , 131 ], taking also into consideration that some interventions focus on self-reported data, without direct measurement of actual behavior change in real-world settings [19,174]. ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. 139:20 M. Giudici et al. The persuasion strategies adopted in the current PCAs for environmental sustainability reveal both valuable insights and limitations. Feedback and monitoring approaches dominate, consistent with established principles in Persuasive Technology [ 55 , 114 ], also for environmental sustainability [ 1 ]. In addition, while many systems implicitly address environmental sustainability, the range of SDGs engaged remains narrow. There is limited integration of SDG frameworks in system design or evaluation, mainly focused on Responsible Consumption and Production (SDG 12) and Affordable and Clean Energy (SDG 7), reflecting a focus on behaviors with relatively immediate, tangible outcomes for final users [ 35 , 169 ]. For example, many examined articles addressed household energy conservation or encouraged more aware consumption, highlighting an effort on smallscale individual behaviors rather than linking them to a broader and global ecological impact (cf. [ 28 ]). While this emphasis affirms the potential for PCAs to prompt home eco-related choices, it also signals an underexplored dimension of additional Sustainable Development Goals, such as equitable resource allocation and enhanced community resilience [ 30 , 53 ]. Still, few studies address interconnected goals comprehensively or synergistically [ 17 , 82 ]. Moreover, equity-related goals are rarely engaged in environmental sustainability, for example, addressing issues of accessibility, inclusivity, and intersectionality [ 87 , 121 , 127 ]. Future investigations could expand the scope of PCAs to lower-addressed SDGs, integrating and combining different goals to establish a multidimensional sustainability framework of action. A salient gap pertains to personalization and user modeling. Although personalization is often mentioned as a design goal, there is limited understanding of how individual differences – such as personality traits, demographics, cultural contexts, or prior environmental knowledge – influence the effectiveness of persuasive strategies on the environmental behaviour of the users [ 72 , 90 , 107 ], similar to previous findings from the literature [ 89 ]. In particular, the role of individual traits in modulating responses to feedback, nudges, or storytelling remains underexplored [ 52 , 116 ]. Furthermore, while some studies emphasize tailoring persuasive strategies, actual implementations of adaptive and context-aware dialogue systems remain scarce [ 139 , 170 ]. The conceptualization and operationalization of persuasion itself also reveal gaps. Many articles implement basic persuasive elements (e.g., reminders, tips), yet fail to rigorously test or compare different persuasive techniques (e.g., social norms, emotional appeals, authority-based messaging) or consider the psychological mechanisms underlying their effectiveness [ 45 , 71 , 101 ]. The limited investigation into multi-strategy and combined persuasion methods also constrains our understanding of how best to structure persuasive dialogues for sustainability [60]. Another aspect to consider is the technological landscape reported in the set of analyzed articles on PCA for environmental sustainability. Many implementations (around 48%) still rely on rule-based architectures (as shown in Figure 12) or lack advanced capabilities in natural language understanding, dialogue management, or multimodal interaction [ 27 , 65 , 131 ]. Such rule-based design implementation often ensures consistency in the messages delivered but can limit the sophistication of the user experience, whereas LLM-based agents could more dynamically adapt to user needs and contextual factors while also carrying risks of factual inaccuracy (i.e., hallucination) [ 79 ]. Although emerging studies experiment with LLMs, their use in environmental or sustainability domains remains nascent and raises concerns regarding bias, ethical, and sustainability deployment [ 59 , 101 , 126 ]. Additionally, integration with physical systems – such as robots, smart devices, or IoT infrastructures – often lacks scalability, robustness, or clear user value [5,62,147]. Moreover, the UX design of PCAs for environmental sustainability was addressed with different modalities and interaction flows, with half of the reviewed articles relying on text-based interfaces and the other set adopting voice-based channels, social robots, or multimodal platforms [ 51 ]. Speech-based systems could also enhance the anthropomorphic qualities of the agent, facilitating rapport, whereas text-based ones can offer reflective, asynchronous interaction (cf. [ 64 ]). Yet direct ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. Persuasive Conversational Agents for Environmental Sustainability: A Survey 139:21 Fig. 12. PCA’s response generation methodology over years. comparative studies of the two modalities remain unexplored; addressing such friction could help developers refine PCAs for real-world user populations whose preferences, technological access, persuasive effectiveness, and environmentally conscious actions vary significantly. The proactivity of PCAs often proves decisive in capturing attention at critical moments [ 155 ]. Such “push-driven” approaches align with broader evidence on environmental-related behavior change interventions, suggesting that timely reminders can effectively nudge individuals toward more sustainable actions. However, an emphasis on agent-driven interactions can lead to perceived intrusiveness or diminished user autonomy [ 146 ], pointing toward finding a balance between beneficial prompting and respect for personal agency. A significant research gap involves the limited diversity in application contexts and populations. First of all, the majority of existing studies are conducted in controlled environments with small, homogeneous participant groups (mainly university-student samples), restricting generalizability [ 10 , 63 , 75 ], posing challenges for external validity [ 21 , 98 ]. Similarly, most of the published findings are based on “case studies” that often lead to incomparable (even contradictory) results, particularly those concerning effectiveness. For instance, Isaza-Giraldo et al . [81] advocated for positive feedback as a common persuasive technique with satisfactory outcomes, while – in the opposite way – [ 71 , 72 ] reported on the effectiveness of negative feedback for effective persuasion. In addition, very few studies are deployed in ecologically valid (in-the-wild[ 142 , 143 ]) studies or include underrepresented geographical and cultural contexts [ 95 , 106 ] as well as evaluation for a prolonged period of time [ 12 ] – since the article under review rarely extending beyond a few weeks. This still raises concerns about the persistence of behavioral changes, as well as the potential for rebound effects [ 68 ] or “green fatigue” [153] caused by the PCA that pushes to more sustainable behavior. At the political and societal levels, new sustainable interventions emphasize collective efforts [ 30 , 162 ], whereas most of the reviewed studies focus on influencing individual actions. The design, development, and evaluation of networked multi-user PCAs that operate at communitylevel, e.g., in contexts of shared energy systems [ 133 , 134 ] or communal waste initiatives [ 176 ], represent a promising but largely unexplored area of research. Expanding PCAs to encompass group negotiations and multi-user interactions could amplify impact, but it also demands more complex, context-specific user modeling and advanced multi-user dialogue management systems. In addition, it opens new challenges in integrating ethical and privacy guidelines, given the potential manipulation and disclosure of sensitive data when multiple users are involved in the dialogue. ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. 139:22 M. Giudici et al. Additionally, in terms of ethical considerations, a growing number of works highlight the need for normative frameworks to guide the development and deployment of environmental-based persuasive systems. Topics such as user autonomy, consent, data privacy, and the ethics of emotional manipulation are underexplored despite their crucial role in persuasive interventions on sensitive domains as environmental sustainability [ 10 , 124 , 154 ]. Finally, it must be considered that LLMbased PCA – throughout their lifecycle – have a significant impact on the environment and energy consumption [ 50 , 85 , 102 ], especially because of their high energy consumption during training and deployment [ 141 ]. Rule-based chatbots, on the other hand, are more environmentally friendly, requiring less energy and leaving a smaller carbon footprint [ 85 ]. According to McTear et al . [113] , both strategies (LLMs and rule-based chatbots) have benefits and drawbacks that future work must consider and comparatively assess. 6 Guidelines In this section, we present actionable recommendations distilled from our survey, which can support researchers and practitioners in the design, development, and assessment of PCAs for environmental sustainability. Lastly, we offer a checklist designed to guide the reporting of future evaluation studies in the field. 6.1 Recommendations Starting from the in-depth analysis performed on the surveyed articles and their discussion, we produced – and reported in this section – a set of recommendations to guide both practitioners (e.g., interaction and conversation designers and technology developers) and researchers in the fields of Persuasive Conversational Agents for environmental sustainability. (1) Transparent and rigorous study designs. To assess different persuasive techniques in conversational agents for sustainability, future studies should include more randomized trials, control groups, or mixed-method designs that are able to create polls of different data (even from different sources). Reporting on sample demographics and data analysis procedures is also critical for reproducibility and guiding future research. (2) Longitudinal and real-world evaluations. To evaluate the impact and persistence of PCAs for environmental sustainability, future studies should use longer time frames and real-world scenarios that extend beyond the short-term laboratory environment. Incorporating actual, logged behavioral measures along with qualitative sustainability insights can help determine whether the persuasive impacts duration over time with a system deployed in the wild. (3) Embed cultural and contextual tailoring. Directly connected to the previous point, analyzed articles and existing systems assume a homogeneous user base, limiting their applicability. Future green PCAs could incorporate cultural norms, local environmental challenges, and region-specific behavior patterns to make sustainability messages more relatable and tailored to users’ daily sustainability challenges. (4) Multi-strategy persuasion analysis. While single-strategy approaches (e.g., feedback strategy only) are still popular, integrating various persuasion strategies (e.g., social comparison, rewards, self-monitoring) can improve effectiveness in having more positive environmental behaviours. Researchers can create adaptable algorithms that swap between persuasive methods as consumers advance ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. Persuasive Conversational Agents for Environmental Sustainability: A Survey 139:23 through different levels of involvement or motivation. This method is supported by recognized behavior-change frameworks (e.g., the Transtheoretical Model [ 137 ]), which provide indicators for when a particular user may benefit the most from various persuasive techniques. (5) Design for multiple users and collective impact. Individual behaviors have been analyzed and are crucial; however, they only reflect one aspect of long-term development. This specific research domain should look at how PCAs can help with collective decision-making, such as in multi-person families or workplace teams in which they are usually placed. Such a branch of research also opens up the possibility of advancing dialogue management capabilities to handle concurrent prompts while addressing group dynamics (e.g., consensus-building, decision-making). (6) Generative AI explainability, ethical, and privacy concerns. To strike a balance between flexibility, trust, and transparency, PCAs for environmental sustainability powered by LLMs or advanced AI approaches should include explainable AI (XAI) [ 122 ]. Short, user-friendly explanations of why certain suggestions are made or how personal data opens research to understand if there is an increase in user understanding, trust, and acceptance. Such an aspect is particularly sensitive in sectors such as sustainability, where ethical, privacy, and social concerns might influence motivations. This technique can also help reduce the hazards associated with LLM hallucinations and provide people with access to the data or reasoning process under a specific PCA suggestion or prompt. (7) Investigate the environmental footprint of LLM-based PCAs. As LLMs become increasingly prevalent, future research should explicitly critically reflect [ 15 ] and evaluate [ 63 ] both the direct and indirect energy costs associated with training, deploying and integrating these models in new PCAs. Emphasis should be placed on ensuring that the design of emerging LLM-based persuasive conversational systems in the field remains aligned with broader sustainability objectives. 6.2 Reporting Checklist This section describes a reporting checklist (Table 2) for describing future works on Persuasive Conversational Agents for environmental sustainability. Such a checklist aims at guiding future work in this domain by reporting all the information on the design, the main feature, and their empirical evaluation of such Conversational Agents. The reporting checklist is based on the findings and the difficulties encountered in reporting variables for the current review, as well as reconnecting with the conceptual framework described in Section 2.2. Even if it is not exhaustive, it is a good starting point for researchers when they want to report studies on green Persuasive Conversational Agents effectively. (1) From the conversational agent design perspective, we recommend future researchers specify the requirements and design choices adopted to build such Persuasive Conversational Agents. Among the possible features of PCA are input and output modalities, embodiment and shape, gender, conversational design, additional capabilities (e.g., emotion recognition), and integration with the external environment. In the previous work of Giudici et al . [61] , researchers could find a framework to support the definition of the main feature of a Conversational Agent for sustainability. (2) From a persuasion point of view, we recommend that future researchers pinpoint the agent’s persuasion strategies (possibly referring to Michie et al . [114] ), This specification allows ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. 139:24 M. Giudici et al. Table 2. Checklist to Systematically Report Empirical Studies on Persuasive Conversational Agents for Environmental Sustainability Category List of Items Example Conversational Agent Design Input Interaction Modalities Output Interaction Modalities Embodiment and Shape of the Agent Gender of the Agent Conversation Design Additional Capabilities Integration with the External Environment speech-based, text-based, etc. speech-based, text-based, etc. physical robot, virtual agent, chatbot, etc. female, male, neutral, etc. rule-based, knowledge-based, etc. emotion recognition, etc. smart plugs connection, transportation API, etc. Persuasion Persuasion Strategy(ies) Persuasion Theory (if any) Feedback and Monitoring, Goal Setting, etc. Sustainable Topic SDG Detailed Description Affordable and Clean Energy (7), etc. e.g., decreasing overall home electric consumption Demographics Sample Size Age Educational Level Gender Distribution Prior Experience with agents reporting number of people involved in the study reporting mean and SD primary school, university, etc. equal, percentage, etc. zero, little, expert, etc. Empirical Study Setting Study design Study duration Nature of the Prototype Collected Data Type Data Collection Method lab setting, home, etc. within-subject, between-subject, etc. hours, days, months, etc. Wizard of Oz, coded, etc. objective, subjective, etc. questionnaires, interview, etc. readers a better understanding of the theoretical feature underlying the agent and improves the replicability of the scientific work. In addition, they also have to consider if the approach adopted is grounded and associated with a specific Persuasion Theory. (3) Addressing the sustainable topic, researchers should identify their conversational agents’ corresponding SDGs and provide a detailed description of how they contribute to reaching the specific goal(s). For example, in the case of a persuasive conversational agent that helps users optimize and avoid the food waste in their fridge, it must be classified into the Responsible Consumption and Production (SDG 12). In addition, researchers should specify that the agent contributes to the goal by using food waste reduction in the household context. (4) Reporting the data about participants in empirical evaluation, researchers have to brief the demographics of the poll. It must include sample size, age of participants, educational level, gender distribution, and prior experience with conversational agents. (5) Finally, empirical studies must report key information for the scientific community evaluation. In the case of Persuasive Conversational Agents for sustainability, it is crucial to state the study’s setting. In addition, researchers should report the study design, its duration, and the type and method of data collection. Finally, it is also relevant to report information on the nature of the prototype, useful to deeply understand the conversational agent design (in the first point). 7 Limitations The value of our survey lies in its review of the current landscape of an emerging topic and its provision of a road map for future research and practice. However, we acknowledge that the study ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. Persuasive Conversational Agents for Environmental Sustainability: A Survey 139:25 may have some limitations. Some relevant articles may have been unintentionally excluded from our review despite our efforts to follow a systematic and standardized method of selection and analysis. We exhaustively identified articles that met the inclusion/exclusion criteria, employed backward and forward chaining of citations for included articles to track additional relevant studies, and involved multiple researchers in the search process, concluding with a final discussion on the relevance and completeness of the results. One of the difficulties was evaluating the scientific quality of some selected articles, partly due to the scarcity of information on various aspects and the different (sometimes limited) reporting methods. Incomplete information may have left some relevant features unrevealed or obscure, resulting in their exclusion from our analysis. For instance, several works used persuasive techniques without explicitly grounding these strategies in persuasive theories. The reporting checklist discussed in Section 6.2 provides a framework for better reporting future work in the field and can help mitigate these issues in future research. Finally, the reported evaluation studies were very heterogeneous regarding empirical research methods, settings, participants, tasks, and durations, making it difficult to derive insights that can be generalized to other evaluation contexts. Finally, given the nascent nature of this research stream, there is a scarcity of works focused on sustainability, specific SDGs, and associated or mix of such themes. Consequently, the results about the “topic” variable are limited and do not provide significant insights. As the field evolves and the range of sustainability topics addressed by future Persuasive Conversational Agents expands, it will become possible to perform more focused analyses of this specific design dimension. 8 Conclusion We presented a survey on Persuasive Conversational Agents for environmental sustainability as an innovative interaction paradigm to move toward a green world and more aware of energy consumption. The 52 articles reviewed provide a comprehensive overview of the main results achieved in the last decade of scientific work in the field. They highlight that various conversational technologies have been explored and a wide range of UX design features have been adopted. However, the articles also indicate some common trends: — the use of Feedback and Monitoring as the primary persuasion strategy, with a focus on Responsible Consumption and Production (SDG 12) as the Sustainable Development Goal addressed. — the integration of conversational solutions for persuasive sustainability with external “smart” environments (i.e., ubiquitous computing). — the evaluation of conversational solutions through short-term studies typically involves interaction with the tool and questionnaires. The survey discusses the key takeaways from the reviewed works and offers several suggestions (referred as recommendations) for further research and development in the area of Persuasive Conversational Agents for environmental sustainability. Some of these suggestions could also inspire practitioners and researchers in other application domains for conversational agents. In addition, for future studies on Persuasive Conversational Agents for environmental sustainability, we offer a reporting checklist. The goal of the checklist is to serve as a guide for researchers reporting the factors we believe are essential to fully describe such agents. References [1] Ifeoma Adaji and Mikhail Adisa. 2022. A review of the use of persuasive technologies to influence sustainable behaviour. In Adjunct Proceedings of the 30th ACM Conference on User Modeling, Adaptation and Personalization. 317–325. ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. 139:32 M. Giudici et al. [132] Matthew J. Page, Joanne E. McKenzie, Patrick M. Bossuyt, Isabelle Boutron, Tammy C. Hoffmann, Cynthia D. Mulrow, Larissa Shamseer, Jennifer M. Tetzlaff, Elie A. Akl, Sue E. Brennan, et al . 2021. The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. BMJ 372 (2021). https://www.bmj.com/content/372/bmj.n71.full.pdf DOI:10.1136/bmj.n71 [133] Georgia Panagiotidou, Enrico Costanza, Michael J Fell, Farhan Samanani, and Hannah Knox. 2023. Supporting solar energy coordination among communities. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 7, 2 (2023), 1–23. [134] Georgia Panagiotidou, Enrico Costanza, Kyrill Potapov, Sonia Nkatha, Michael J. Fell, Farhan Samanani, and Hannah Knox. 2024. SolarClub: Supporting renewable energy communities through an interactive coordination system. In Proceedings of the CHI Conference on Human Factors in Computing Systems. 1–19. [135] Alessandro Petruzzelli, Cataldo Musto, Michele Ciro Di Carlo, Giovanni Tempesta, and Giovanni Semeraro. 2024. Recommending healthy and sustainable meals exploiting food retrieval and large language models. In Proceedings of the 18th ACM Conference on Recommender Systems. ACM, Bari Italy, 1057–1061. DOI:10.1145/3640457.3688193 [136] James Pierce and Eric Paulos. 2012. The local energy indicator: Designing for wind and solar energy systems in the home. Proceedings of the Designing Interactive Systems Conference (2012), 631–638. DOI:10.1145/2317956.2318050 [137] James O. Prochaska and Carlo C. DiClemente. 1983. Stages and processes of self-change of smoking: Toward an integrative model of change. Journal of Consulting and Clinical Psychology 51, 3 (1983), 390. [138] Mohaimenul Azam Khan Raiaan, Md Saddam Hossain Mukta, Kaniz Fatema, Nur Mohammad Fahad, Sadman Sakib, Most Marufatul Jannat Mim, Jubaer Ahmad, Mohammed Eunus Ali, and Sami Azam. 2024. A review on large language models: Architectures, applications, taxonomies, open issues and challenges. IEEE Access 12 (2024), 26839–26874. [139] Dhineshkumar Ramasubbu, Krishnamoorthy Baskaran, and Grynberg Yann. 2018. Intrusive plug management system using chatbots in office environments. In Proceedings of the 2018 Asian Conference on Energy, Power and Transportation Electrification. IEEE, Singapore, 1–4. DOI:10.1109/ACEPT.2018.8610869 [140] Minjin Rheu, Ji Youn Shin, Wei Peng, and Jina Huh-Yoo. 2021. Systematic review: Trust-building factors and implications for conversational agent design. International Journal of Human–Computer Interaction 37, 1 (2021), 81–96. [141] Matthias C. Rillig, Marlene Ågerstrand, Mohan Bi, Kenneth A. Gould, and Uli Sauerland. 2023. Risks and benefits of large language models for the environment. Environmental Science and Technology 57, 9 (2023), 3464–3466. [142] Yvonne Rogers. 2011. Interaction design gone wild: Striving for wild theory. Interactions 18, 4 (2011), 58–62. DOI:10.1145/1978822.1978834 [143] Yvonne Rogers, Kay Connelly, Lenore Tedesco, William Hazlewood, Andrew Kurtz, Robert E. Hall, Josh Hursey, and Tammy Toscos. 2007. Why it’s worth the hassle: The value of in-situ studies when designing ubicomp. In Proceedings of the UbiComp 2007: Ubiquitous Computing. Lecture Notes in Computer Science, Vol. 4717, Springer Berlin, 336–353. DOI:10.1007/978-3-540-74853-3_20 [144] Manuela Sanguinetti and Maurizio Atzori. 2024. Conversational agents for energy awareness and efficiency: A survey. Electronics 13, 2 (2024), 401. [145] Manuela Sanguinetti, Alessandro Pani, Alessandra Perniciano, Luca Zedda, Andrea Loddo, and Maurizio Atzori. 2024. Assessing italian large language models on energy feedback generation: A human evaluation study. In Proceedings of the 10th Italian Conference on Computational Linguistics (CLiC-it 2024). 880–887. [146] Supraja Sankaran, Chao Zhang, Mathias Funk, Henk Aarts, and Panos Markopoulos. 2020. Do i have a say? using conversational agents to re-imagine human-machine autonomy. In Proceedings of the 2nd Conference on Conversational User Interfaces. 1–3. [147] Clara Scheutz, Theresa Law, and Matthias Scheutz. 2021. EnviRobots: How human–robot interaction can facilitate sustainable behavior. Sustainability 13, 21 (2021), 12283. DOI:10.3390/su132112283 [148] Daniel Schulman and Timothy Bickmore. 2009. Persuading users through counseling dialogue with a conversational agent. In Proceedings of the 4th International Conference on Persuasive Technology. 1–8. [149] Tobias Schwartz, Gunnar Stevens, Leonardo Ramirez, and Volker Wulf. 2013. Uncovering practices of making energy consumption accountable: A phenomenological inquiry. ACM Transactions on Computer-Human Interaction 20, 2 (2013), 1–30. [150] Alex Sciuto, Arnita Saini, Jodi Forlizzi, and Jason I. Hong. 2018. ” Hey Alexa, what’s up?” A mixed-methods studies of in-home conversational agent usage. In Proceedings of the 2018 Designing Interactive Systems Conference. 857–868. [151] Katie Seaborn, Norihisa P. Miyake, Peter Pennefather, and Mihoko Otake-Matsuura. 2021. Voice in human -agent interaction: A survey. ACM Computing Surveys (CSUR) 54, 4 (2021), 1–43. [152] Nataliya Shevchuk and Harri Oinas-Kukkonen. 2016. Exploring green information systems and technologies as persuasive systems: A systematic review of applications in published research. In Proceedings of the International Conference on Information Systems - Digital Innovation at the Crossroads, ICIS 2016, Dublin, Ireland, December 11-14, 2016, Association for Information Systems. ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. Persuasive Conversational Agents for Environmental Sustainability: A Survey 139:33 [153] Judith B. Strother and Zohra Fazal. 2011. Can green fatigue hamper sustainability communication efforts?. In Proceedings of the 2011 IEEE International Professional Communication Conference. IEEE, 1–6. [154] Pannee Suanpang and Pattanaphong Pothipassa. 2024. Integrating generative AI and IoT for sustainable smart tourism destinations. Sustainability 16, 17 (2024), 7435. DOI:10.3390/su16177435 [155] Robert Sugden. 2009. On nudging: A review of nudge: Improving decisions about health, wealth and happiness by Richard H. Thaler and Cass R. Sunstein. International Journal of the Economics of Business 16, 3 (2009), 365–373. DOI:https://doi.org/10.1080/13571510903227064 [156] Qiming Sun and I-Han Hsiao. 2023. Waste genie: Learning environmental sustainability from waste sorting and interactive feedback. In 2023 IEEE Global Humanitarian Technology Conference (GHTC). 310–317. DOI: https://doi.org/ 10.1109/GHTC56179.2023.10354701 [157] Bernardo Tabuenca, Wolfgang Greller, Davinia Hernández Leo, Carlos Gilarranz-Casado, Vicente Garcıa-Alcántara, and Edmundo Tovar. 2021. Talking to plants: An IoT system supporting human-plant interactions and learning. Workshop LAS4SLE @ EC-TEL 2021: Learning Analytics for Smart Learning Environments. [158] Sandeep A. Thorat and Vishakha Jadhav. 2020. A review on implementation issues of rule-based chatbot systems. In Proceedings of the International Conference on Innovative Computing and Communications. [159] Anna K. Trukenbrod and Manfred Thüring. 2020. Subjectively experienced time in human-computer interaction: The role of passage of time in user experience. In Advances in Usability, User Experience, Wearable and Assistive Technology: Proceedings of the AHFE 2020 Virtual Conferences on Usability and User Experience, Human Factors and Assistive Technology, Human Factors and Wearable Technologies, and Virtual Environments and Game Design, July 16-20, 2020, USA. Springer, 121–128. [160] Wan-Hsiu Sunny Tsai, Yu Liu, and Ching-Hua Chuan. 2021. How chatbots’ social presence communication enhances consumer engagement: The mediating role of parasocial interaction and dialogue. Journal of Research in Interactive Marketing 15, 3 (2021), 460–482. [161] George A. Tsihrintzis, Elissaios Sarmas, Vangelis Marinakis, Dimitrios Panagoulias, Evangelia-Aikaterini Tsihrintzi, and Maria Virvou. 2024. Evaluating non-expert stakeholder interaction with artificial intelligence on energy urban domain using VIRTSI: The case of ChatGPT. In Proceedings of the 2024 15th International Conference on Information, Intelligence, Systems & Applications. IEEE, Chania Crete, Greece, 1–10. DOI:10.1109/IISA62523.2024.10786650 [162] UK Government. 2023. Community Energy. Retrieved April 15th, 2025 from https://www.gov.uk/guidance/communityenergy Accessed: 2024-08-29. [163] Simon Thomas van Baal, Suong Thi Thanh Le, Farhad Fatehi, Antonio Verdejo-Garcia, and Jakob Hohwy. 2022. Effecting behaviour change using an artificial intelligence chatbot: A pilot randomised controlled study. PsyArXiv 10 (2022), 31234. [164] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in Neural Information Processing Systems. Curran Associates, Inc. [165] Viswanath Venkatesh, Michael G. Morris, Gordon B. Davis, and Fred D. Davis. 2003. User acceptance of information technology: Toward a unified view. MIS quarterly 3 (2003), 425–478. [166] Suzanne Vossen, Jaap Ham, and Cees Midden. 2009. Social influence of a persuasive agent: The role of agent embodiment and evaluative feedback. In Proceedings of the 4th International Conference on Persuasive Technology. 1–7. [167] Xuewei Wang, Weiyan Shi, Richard Kim, Yoojung Oh, Sijia Yang, Jingwen Zhang, and Zhou Yu. 2019. Persuasion for good: Towards a personalized persuasive dialogue system for social good. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, Association for Computational Linguistics, Florence, Italy, 5635–5649. DOI:https://doi.org/10.18653/v1/P19-1566 [168] Joseph Weizenbaum. 1966. ELIZA–A computer program for the study of natural language communication between man and machine. Communications of the ACM 9, 1 (1966), 36–45. [169] Katherine White, Rishad Habib, and David J. Hardisty. 2019. How to SHIFT consumer behaviors to be more sustainable: A literature review and guiding framework. Journal of Marketing 83, 3 (2019), 22–49. [170] Yifang Xu, Siyao Zhu, Jiannan Cai, Jianli Chen, and Shuai Li. 2025. A large language model-based platform for real-time building monitoring and occupant interaction. Journal of Building Engineering 100 (2025), 111488. DOI:10.1016/j.jobe.2024.111488 [171] Mizuki Yamawaki, Kimi Ueda, Yoshiki Sakamoto, Hirotake Ishii, Hiroshi Shimoda, Kyoko Ito, Takuya Fujioka, Qinghua Sun, Yasuhiro Asa, and Takashi Numata. 2020. An experimental study on promotion of pro-environmental behavior focusing on “vanity” for interactive agent. In Proceedings of the Social Computing and Social Media. Participation, User Experience, Consumer Experience, and Applications of Social Computing. Gabriele Meiselwitz (Ed.), Vol. 12195, Springer International Publishing, Cham, 247–258. DOI:10.1007/978-3-030-49576-3_19 [172] Yifan Yao, Jinhao Duan, Kaidi Xu, Yuanfang Cai, Zhibo Sun, and Yue Zhang. 2024. A survey on large language model (llm) security and privacy: The good, the bad, and the ugly. High-Confidence Computing 4, 2 (2024), 100211. ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. 139:34 M. Giudici et al. [173] Ray Yun, Peter Scupelli, Azizan Aziz, and Vivian Loftness. 2013. Sustainability in the workplace: Nine intervention techniques for behavior change. In Proceedings of the International Conference on Persuasive Technology. Springer, 253–265. [174] Emma L. Zaal, Gregory J. Mills, Afke Hagen, Carlijn A. Huisman, and John C. J. Hoeks. 2017. Convincing conversations: Using a computer-based dialogue system to promote a plant-based diet. In Proceedings of the Annual Meeting of the Cognitive Science Society. [175] Suifang Zhou, Latisha Besariani Hendra, Qinshi Zhang, Jussi Holopainen, and Ray Lc. 2024. Eternagram: Probing player attitudes towards climate change using a ChatGPT-driven text-based adventure. In Proceedings of the CHI Conference on Human Factors in Computing Systems. ACM, Honolulu HI USA, 1–23. DOI:10.1145/3613904.3642850 [176] Christian Zurbrügg, Silke Drescher, Almitra Patel, and HC Sharatchandra. 2004. Decentralised composting of urban waste–an overview of community and private initiatives in Indian cities. Waste Management 24, 7 (2004), 655–662. ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. Persuasive Conversational Agents for Environmental Sustainability: A Survey 139:35 Appendix A List of Surveyed Articles This appendix lists the 52 articles included in this survey and cited in Table 3. In addition, the supplementary material of this manuscript provides an Excel file that reports all the analyses carried out on the surveyed articles. Table 3. Table Reporting the Articles Included in this Survey on Persuasive Conversational Agents for Environmental Sustainability # Title Author(s) Year Reference 1 A robot that says “bad!”: Using negative and positive social feedback from a robotic agent to save energy Jaap Ham, Cees Midden 2009 [71] 2 Social influence of a persuasive agent: The role of agent embodiment and evaluative feedback Vossen S., Ham J., Midden C. 2009 [166] 3 The illusion of agency: The influence of the agency of an artificial agent on its persuasive power Midden C., Ham J. 2012 [115] 4 Designing motivation using persuasive ambient mirrors Tatsuo Nakajima, Vili Lehdonvirta 2013 [124] 5 A Persuasive Robot to Stimulate Energy Conservation: The Influence of Positive and Negative Social Feedback and Task Similarity on Energy-Consumption Behavior Ham J., Midden C.J.H. 2014 [72] 6 Live capture of energy-related knowledge into BIM systems Motawa I., Janarthanam S., Almarshad A. 2014 [120] 7 Conversations with my washing machine: an in-the-wild study of demand shifting with self-generated energy Jacky Bourgeois, Janet van der Linden, Gerd Kortuem, Blaine A. Price, and Christopher Rimmer 2014 [14] 8 Conforming to an artificial majority: persuasive effects of a group of artificial agents C Midden, J Ham, J Baten 2015 [116] 9 Tariff Agent: Interacting with a Future Smart Energy System at Home Alper T. Alan, Enrico Costanza, Sarvapali D. Ramchurn, Joel Fischer, Tom Rodden, and Nicholas R. Jennings 2016 [5] 10 Buildings with persona: Towards effective building-occupant communication Khashe S., Lucas G., Becerik-Gerber B., Gratch J. 2017 [90] 11 Convincing Conversations: Using a Computer-Based Dialogue System to Promote a Plant-Based Diet Zaal E.L.; Mills G.J.; Hagen A.; Huisman C.A.; Hoeks J.C.J. 2017 [174] 12 Designing conversational agents for energy feedback Gnewuch U., Morana S., Heckmann C., Maedche A. 2018 [65] 13 Intrusive Plug Management System Using Chatbots in Office Environments Ramasubbu, D; KrishnamoorthyBaskaran; Yann, G 2018 [139] 14 Virtual assistant for IoT process management, using a middleware Chilcañán D., Navas P., Escobar M. 2018 [27] 15 Promoting sustainable mobility beliefs with persuasive and anthropomorphic design: Insights from an experiment with a conversational agent Diederich S., Lichtenberg S., Brendel A.B., Trang S. 2019 [45] 16 Zero Food Waste: Food wastage sustaining mobile application M. D. C. J. Gunawardane; H. A. N. Pushpakumara; E. N. M. R. L. Navarathne; S. Lokuliyana; K. T. I. Kelaniyage; N. Gamage 2019 [69] (Continued) ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. 139:36 M. Giudici et al. Table 3. Continued # Title Author(s) Year Reference 17 Design and implementation of a novel environmental friendly chatbot based on VSM and raspberry PI with environmental protection knowledge Ouyang L., Luo X., Xiong L. 2020 [131] 18 Factors influencing adoption intention of ai powered chatbot for public transport services within a smart city Kuberkar S., Singhal T.K. 2020 [95] 19 Greening food consumption using chatbots as behavioral change agent Cacanindin N.M., Palaoag T.D. 2020 [19] 20 GreenLife: A Persuasive Social Robot to Enhance the Sustainable Behavior in shared Living Spaces Beheshtian N., Moradi S., Ahtinen A., Väänanen K., Kähkonen K., Laine M. 2020 [10] 21 Let’s Learn Biodiversity with a Virtual Robot? Ferreira, MJ; Oliveira, R; Olim, SC; Nisi, V; Paiva, A 2020 [52] 22 “Meet, greet and learn: introducing Bern!” Saving energy through voice interaction with your smart thermostat S de Vos, B Rolvink, N Güneş, S San Nguyen… 2020 [40] 23 An experimental study on promotion of proenvironmental behavior focusing on “vanity” for interactive agent M Yamawaki, K Ueda, Y Sakamoto, H Ishii… 2020 [171] 24 EnviRobots: How Human-Robot Interaction Can Facilitate Sustainable Behavior Scheutz, C; Law, T; Scheutz, M 2021 [147] 25 PeppeRecycle: Improving Children’s Attitude Toward Recycling by Playing with a Social Robot Castellano, G; De Carolis, B; D’Errico, F; Macchiarulo, N; Rossano, V 2021 [24] 26 Talking to plants: An IoT system supporting human-plant interactions and learning Tabuenca B., Greller W., Hernández-Leo D., Gilarranz-Casado C., García-Alcántara V., Tovar E. 2021 [157] 27 Chatbot as a Persuasive Technology to Promote Responsible Recycling in the City of Lima Flores K.A.F.; Perez J.J.G.; Sanchez L.M.C. 2022 [54] 28 Arabic Mini-ClimateGPT: A Climate Change and Sustainability Tailored Arabic LLM Mullappilly S.S.; Shaker A.; Thawkar O.; Cholakkal H.; Anwer R.M.; Khan S.; Khan F.S. 2023 [121] 29 Leafy: Enhancing Home Energy Efficiency through Gamified Experience with a Conversational Smart Mirror Giudici M.; Crovari P.; Garzotto F. 2023 [62] 30 Assessing LLMs Responses in the Field of Domestic Sustainability: An Exploratory Study Giudici M.; Abbo G.A.; Belotti O.; Braccini A.; Dubini F.; Izzo R.A.; Crovari P.; Garzotto F. 2023 [59] 31 CANDY: A framework to design Conversational AgeNts for Domestic sustainabilitY Giudici M.; Crovari P.; Garzotto F. 2023 [61] 32 Integrating Generative AI and IoT for Sustainable Smart Tourism Destinations Suanpang P.; Pothipassa P. 2024 [154] 33 Eternagram: Probing Player Atitudes in Alternate Climate Scenarios Through a ChatGPT-Driven Text Adventure Zhou S.; Hendra L.B.; Zhang Q.; Holopainen J.; Lc R. 2024 [175] 34 A novel approach to sustainable behavior enhancement through AI-driven carbon footprint assessment and real-time analytics Jasmy A.J.; Ismail H.; Aljneibi N. 2024 [82] 35 Healthy and Sustainable Meals Recommendation Exploiting Food Retrieval and Large Language Models Petruzzelli A.; Musto C.; Di Carlo M.C.; Tempesta G.; Semeraro G. 2024 [135] (Continued) ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025. Persuasive Conversational Agents for Environmental Sustainability: A Survey 139:37 Table 3. Continued # Title Author(s) Year Reference 36 Generative AI and Retrieval-Augmented Generation (RAG) in an Agent-Based Simulation Framework for Urban Planning Bruzzone A.; Giovannetti A.; Genta G.; Cefaliello D. 2024 [17] 37 Study of Promoting Energy Saving Behavior in Homes with Personalized Adaptive Interaction Hirai S.; Okamoto H.; Nakata T.; Chen S.; Saiki S.; Nakamura M. 2024 [75] 38 AI Insights: Unveiling UK Energy Consumption with Langchain Powered Chatbots Justice K.; Vakaj E.; Dridi A. 2024 [87] 39 Promoting pro-environmental behaviour spillover through chatbots Majid G.M.; Tussyadiah I.; Kim Y.R.; Chen J.L. 2024 [107] 40 Evaluating Non-Expert Stakeholder Interaction with Artificial Intelligence on Energy Urban Domain Using VIRTSI: The Case of ChatGPT Tsihrintzis G.A.; Sarmas E.; Marinakis V.; Panagoulias D.; Tsihrintzi E.-A.; Virvou M. 2024 [161] 41 Assessing Italian Large Language Models on Energy Feedback Generation: A Human Evaluation Study Sanguinetti M.; Pani A.; Perniciano A.; Zedda L.; Loddo A.; Atzori M. 2024 [145] 42 Care-Based Eco-Feedback Augmented with Generative AI: Fostering ProEnvironmental Behavior through Emotional Atachment Berney M.; Ouaazki A.; Macko V.; Kocher B.; Holzer A. 2024 [11] 43 Prompt-Gaming: A Pilot Study on LLMEvaluating Agent in a Meaningful Energy Game Isaza-Giraldo A.; Bala P.; Campos P.F.; Pereira L. 2024 [81] 44 Misrepresentation or inclusion: promises of generative artificial intelligence in climate change education Nguyen H.; Nguyen V.; Ludovise S.; Santagata R. 2024 [126] 45 Responsible Configuration Using LLMbased Sustainability-Aware Explanations Lubos S.; Felfernig A.; Hotz L.; Tran T.N.T.; Polat-Erdeniz S.; Le V.-M.; Garber D.; ElMansi M. 2024 [101] 46 Learning Waste Management from Interactive Quizzes and Adaptive GPT-guided Feedback Sun Q.; Chien S.-Y.; Hsiao I.-H. 2024 [156] 47 Delivering Green Persuasion Strategies with a Conversational Agent: a Pilot Study Giudici M.; Abbo G.A.; Crovari P.; Garzotto F. 2024 [60] 48 RoboRecycle Buddy: Enhancing Early Childhood Green Education and Recycling Habits Through Playful Interaction with a Social Robot Mahmud S.; Kamel Z.; Singh A.; Kim J.-H. 2024 [105] 49 Exploring the Potential of Chatbots in Extending Tourists’ Sustainable Travel Practices Majid G.M.; Tussyadiah I.; Kim Y.R. 2024 [106] 50 Designing Home Automation Routines Using an LLM-Based Chatbot Giudici M.; Padalino L.; Paolino G.; Paratici I.; Pascu A.I.; Garzotto F. 2024 [63] 51 Value-sensitive design of chatbots in environmental education: Supporting identity, connectedness, well-being and sustainability Nguyen H.; Nguyen V.; Ludovise S.; Santagata R. 2025 [127] 52 A large language model-based platform for real-time building monitoring and occupant interaction Xu Y.; Zhu S.; Cai J.; Chen J.; Li S. 2025 [170] Received 7 March 2023; revised 6 August 2025; accepted 29 October 2025 ACM Comput. Surv., Vol. 58, No. 6, Article 139. Publication date: December 2025.