Latest / 1 More Paper / To hasten slowly: The prudence of slow AI implementation in public relations — Christensen & Andersson (2025)

Transcript

To hasten slowly: The prudence of slow AI implementation in public relations — Christensen & Andersson (2025) · 1 More Paper

• You’re listening to “To hasten slowly: The prudence of slow AI implementation in public relations,” by E. Christensen and R. Andersson. Published in 2025. To hasten slowly The prudence of slow AI implementation in public relations Christensen, Emma; Andersson, Rickard Published in: Public Relations Review Publication date: 2025 Document Version Publisher's PDF, also known as Version of record Creative Commons License CC BY Citation for published version (APA): Roskilde University Christensen, E., & Andersson, R. (2025). To hasten slowly: The prudence of slow AI implementation in public relations. Public Relations Review, 51, Article 102557. the linked source General rights and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. Document Removal If you believe that this document breaches copyright please contact the email address providing details, and we will remove access to the work immediately and investigate your claim. Contents lists available at ScienceDirect Public Relations Review journal homepage: the linked source To hasten slowly: The prudence of slow AI implementation in public relations Emma Christensen a,, Rickard Andersson b a Department of Communication and Arts Roskilde University Universitetsvej 1, Roskilde 4000, Denmark b Department of Communication, Lund University, Sweden ’Hasten slowly’.... ’That is done fast enough, which is done well enough’ (Emperor Augustus quoted in Suetonius, 121/1890, Caesar Augustus, Chapter XXV, Lines 34–38) 1. Introduction. Artificial intelligence (AI) has for years been predicted to transform public relations practices (e.g., Buhmann & White, 2022; Galloway & Swiatek, 2018; Soriano & Vald ́es, 2021; Swiatek & Galloway, 2022). AI systems’ ability to assist or augment professionals’ work is, for example, expected to allow professionals to solve operational tasks more efficiently and effectively, enabling them to focus on more strategic or creative tasks. However, despite transformative expectations, public relations professionals’ implementation of AI has been described as worrisomely slow, and several studies express fear that public relations risks lagging behind adjacent industries such as marketing and HR. Thus, rapid adoption and Available online 28 March 2025 implementation of AI tools is deemed crucial for public relations professionals since “It is only by being ahead of the game that it will grasp that most strategic of opportunities – to take an important seat at the governance table”. Thus, when professionals’ reported use of AI tools tripled during the second half of 2023 compared to the beginning of the year, it was perceived as a sign that professionals had finally woken up. While over 6000 AI tools relevant to public relations professionals were available at the beginning of 2023, it is highly suggestive that the generative AI (GenAI) system ChatGPT had a significant impact on professionals’ adoption rate. However, whereas the adoption of AI is gaining momentum, little is known about what this adoption looks like and how professionals have implemented and integrated GenAI systems into their workflows. AI implementation is a particular instance of the broader digital transformation of organizational processes. This transformation, particularly its consequences for public relations departments, is gaining increasing attention from public relations scholars (e.g., Brockhaus et al., 2023; Zerfass & Brockhaus, 2021, 2023; see also Arthur W. Page Society, 2019, 2021). Drawing on the socio-technical system view of 0363-8111/© 2025 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license (the linked source). technology implementation (STS), Brockhaus et al. (2023) conceptualize the digital transformation of public relations as a socio-technical change process. However, studies of such changes in communication departments are scarce. In this article, we address this lack of research, thereby contributing to a more profound understanding of AI and technology implementation in public relations. We do so by adopting the socio-technical change process view and exploring a one-year-long pre-implementation “learning process” conducted by a communication department (henceforth CoD) in a large Danish municipality. Specifically, our study answers the research question: How did CoD explore the potential integration of AI into their workflows? Following Brockhaus et al. (2023; see also Zerfass & Brockhaus, 2021), we adopt STS as our analytical lens, emphasizing the need to equally consider the social system (people and structures) and the technical system (technologies and tasks) during technology implementation. Our study revealed that the professionals approached AI implementation exploratively, emphasizing co-worker involvement and careful reflection. This slow and human-centered approach, where co-workers actively participated in framing, organizing, and executing the process, likely contributed to the fact that many of the challenges identified by previous research as “hidden costs” and barriers towards sustainable AI implementation were avoided. CoD’s approach contrasts prevailing recommendations on AI implementation in public relations, which often stress the need for rapid AI implementation if public relations professionals are to keep up. Based on our findings, we propose the concept and practice of Slow Implementation in public relations to underscore the importance of dedicating time to the (pre)implementation phase. A slower pace of AI implementation allows professionals the time to acquire the knowledge and skills necessary for a sound and sustainable AI implementation. It also increases the likelihood of co-workers’ active participation during the implementation. By introducing the concept and practice of slow implementation, we contribute to the emerging literature on AI and communication technology implementation in public relations. The remainder of the paper is structured as follows: First, the study is situated within the CommTech literature on the digital transformation of public relations. Second, the theoretical framework of STS is introduced, including previous studies on AI implementation in organizational settings. Third, the empirical study and the analytical procedure are outlined. The subsequent section presents the study’s findings, structured around the learning process’s main phases. Lastly, we discuss contributions and propose avenues for future research. 2. The digital transformation of public relations:. communication technology and AI implementation Although the introduction of GenAI systems such as ChatGPT and Microsoft’s Copilot in late 2022 and early 2023 has directed attention to the transformative potential of digital technology for public relations, the digital transformation of public relations has been ongoing for several decades and has attracted the attention of public relations scholars for about as long. Duh ́e’s (2015) overview of communication technology research in public relations between 1981 and 2014, for example, shows that scholarly conjectures about the digital transformation’s consequences for public relations date back to at least the 1980s. Moreover, as recent as 2019, thus predating the introduction of GenAI to the broader public, Lock (2019) pointed out that the ongoing digital transformation and new communication technology have already significantly impacted the daily work of public relations professionals. The recent introduction of GenAI into communication departments may thus be situated within this broader context of the digital transformation of public relations, where organizations increasingly use different communication technologies to strategically and tactically manage their communications activities. It is no exaggeration to state that the digital transformation of society and public relations has been exponential since the 1980s, and it is nowadays a prominent topic in public relations research. However, the public relations literature on the digital transformation of public relations has become significantly fragmented during the 21st century. To address this fragmentation, Kretschmer and Winkler review narratives about digital technology in public relations and its societal impact and identify five underlying narrative streams: cybernetics, connectivity, empowerment, transhumanism, and disruption. They suggest that placing these narratives in the limelight may enable public relations researchers to shift focus away from superficial and ephemeral technological trends and fads toward advancing the underlying and more enduring scientific narratives about the impact of the digital transformation of public relations. In this article, our primary aim is to engage in a conversation with and make a contribution to two of the research strands on the digital transformation of public relations identified by Kretschmer and Winkler, namely the strand they label as the “transhumanist”, that is, research exploring how digital technology such as AI may facilitate “the possibility of transcending the established dualisms between human and nonhuman agency” and “disruption”, that is, research investigating the “game-changing impact of new digital technologies on future business success”. In the public relations literature investigating the disruptive potential of communication technology, the term CommTech (communication technology) has been proposed to capture the presence and influence of digital technologies and infrastructures in communication departments and organizations. CommTech is increasingly used in a wide variety of communication management activities, including stakeholder communication, internal advising, and functional support activities (Pekkala & Erkkil ̈a, 2024; Weiner, 2021; Zerfass & Brockhaus, 2021, 2023). Brockhaus et al. (2023, p. 277; see also Zerfass & Brockhaus, 2021) define CommTech as “digital technologies provided or used by communication functions or departments to manage and perform primary activities, particularly stakeholder communications and internal advising, or functional support activities such as managing internal workflows for monitoring, content planning, or evaluation”. Digitalization and the introduction of (Gen)AI systems create new requirements for public relations departments and professionals in two main ways. First, as both objects and actors of change, departments must consider which systems to adopt and their implications on communication activities and management. Second, as a support function within organizations, public relations departments are likely to be involved in communicating digital transformation and AI implementation across other departments or throughout the entire organization as well as explaining the organization’s use of technology such as (Gen)AI to external publics. This requires professionals to be knowledgeable about their organization’s digital transformation in general and (Gen)AI systems in particular. In addition to providing a conceptual definition of CommTech, Brockhaus et al. (2023) also report the findings from a quantitative survey among public relations professionals in Europe. The findings show that while professionals acknowledge the importance of digitalization for public relations practice and professionals, the digital maturity of communication departments in Europe is relatively low. The findings indicate a lack of strategic approaches to CommTech, with ad hoc implementations being more prevalent. Lastly, inspired by STS, Brockhaus et al. stress the need for future research to adopt a broad view of CommTech implementation that embraces both technical and social dimensions. Despite CommTech’s growing presence in communication departments during the last decades, the increasing importance ascribed to CommTech in research and practice, and GenAI’s potential to further disrupt public relations practices, studies of the adoption and implementation of CommTech are still lacking, especially studies exploring actual implementation and use and not only practitioners’ perceptions. Through our study of AI pre-implementation, we contribute to the CommTech literature and thereby answer the call for additional research on CommTech implementation and its implications for public relations professionals, public relations departments, and organizations made by Brockhaus et al. (2023) and Zerfass and Brockhaus (2021, 2023; see also Virmani & Gregory, 2021). 3. A socio-technical lens on technology implementation. Inspired by previous CommTech research, we adopt the analytical lens of STS view of technology implementation developed within the field of Information Systems. Introduced by Trist and colleagues in the early 1950s, STS has since underscored the interdependent and recursive influence of social (people and structure) and technical (technology and tasks) subsystems in implementation and technology use within organizations. The theory suggests that for technology to be effectively integrated within an organization, the technical and social systems must be aligned and mutually supportive. The social element, key from the founding of STS, emphasizes the importance of a human-centered design perspective. This perspective includes involving co-workers (i.e., the users of technology) in (re)formulating goals, policies, and decisions regarding which technologies to implement and how to do so. Echoing the early days of STS, Mumford (2006, p. 338) asserts that, “employees should be allowed and encouraged to participate in, and influence, decisions that concern them”. However, much STS research has drifted away from the principle of equal interdependence, often treating the social subsystem as subordinate to the technical, requiring it to adapt to fit new technologies introduced by management. Indeed, since the 1990s, it has become commonplace for researchers to view technology implementation as a classical top-down process, where co-workers’ involvement often is limited to being educated about the technology they are expected to use (e.g., Benbya et al., 2020; B ́erub ́e et al., 2021; Zerfass & Brockhaus, 2023). Recently, scholars have begun to criticize the prioritization of the technical subsystem, arguing that it reveals how economic value, including enhanced productivity and efficiency gained from new technology, is considered more important than the social or humanistic aspects, such as co-workers’ well-being and satisfaction. When humanistic concerns are deprioritized, the likelihood of “hidden costs” or “post-decision surprises” of technology implementation increases, underscoring the significance of striking a balance between the social and technical subsystems in technology implementation. In the following, we define and discuss the four elements of the subsystems – structure, people, technology, and tasks – based on previous research on AI implementation. The social subsystem entails structure and people. Structure refers to organizational hierarchy, goals, roles, work practices, and communication channels, determining task coordination and information flow. Previous research indicates that failure to adapt organizational structure is often a critical factor in unsuccessful AI implementation (e.g., B ́erub ́e et al., 2021; Pumplin et al., 2019; Ångstr ̈om et al., 2023). To address this, researchers recommend creating new roles, cross-functional teams with representatives from the department(s) using the AI system and the IT department, and structures tailored to AI integration while also considering governance, control measures, and ethical concerns. The people element focuses on employees’ well-being, satisfaction, skills, and attitudes. Previous studies show that common barriers include a lack of understanding of AI among managers, employees, and external publics, and weak motivation to adopt new technology (B ́erub ́e et al., 2021; Christensen et al., 2022; Kolbjørnsrud et al., 2017; O’Neil et al., 2023; Zerfass et al., 2020), and fear of identity loss, changes in power relations, lower salary, or job loss (Christensen et al., 2022; Gregory et al., 2023; O’Neil et al., 2023; Virmani & Gregory, 2021; Zerfass et al., 2019; Zerfass et al., 2020; see also Arthur W. Page Society, 2019). The ethical dimension of AI, including issues such as lack of legal clarity, bias, opaqueness, and exclusion, also constitutes significant concerns that challenge implementation. According to the literature, these challenges can be mitigated by fostering a positive learning environment and offering hands-on AI experiences to enhance understanding and skills. The technical subsystem consists of technology and tasks. Technology refers to the digital tools, systems, hardware, and software implemented and used within the organization. While AI is often depicted as an independent object to be adopted and implemented, we assume the social constructivist position within STS and view technology as a social construct shaped by the context in which it is used. By implication, technologies entail interpretative flexibility, meaning that while technologies have inherent properties or affordances, their integration into workflows and organizational practices and routines will differ between organizations. Organizational members may thus interpret, appropriate, and use the same technology in various ways. Tasks encompass the specific activities, processes, workflows, procedures, and operations conducted within the organization. The raison d’ˆetre of new technology is to enhance the efficiency and effectiveness of task accomplishment, thereby inevitably impacting how work is conducted. In an AI context, this means that ”work practices must be re-engineered [...] as workflows are broken down into smaller tasks corresponding to individual AI algorithms’ relatively limited capabilities” (Ångstr ̈om et al., 2023, p. 7). Adopting the STS framework allows for sensitivity to the various elements of GenAI implementation, emphasizing the need to consider structure, people, technology, and tasks equally for a more sustainable outcome. 4. The study. To explore the implementation of AI in the public relations context, we analyze what we refer to as a “pre-implementation” process conducted by a communication department (CoD) in a large Danish municipality. The municipality regards digitalization as a critical tool to enhance efficiency and free up resources, facilitating better value creation for its citizens. Consequently, both the municipality at large and the specific administration to which CoD belongs undertake digitalization projects, typically initiated by top management and the IT department, and executed in a traditional top-down manner. However, the specific pre-implementation project explored in this study is unique in several respects, to the administration and CoD. Initiated by a project leader at CoD and the Head of Communication, with support from the IT department, its emphasis on learning rather than implementation, its involvement of co-workers, and its one-year timeframe is one of a kind. Additionally, it was the only digitalization project driven by co-workers at CoD, encompassing the entire department and involving all co-workers. While potentially unique also outside the municipality, CoD’s pre-implementation process constitutes a highly relevant case to explore. Due to its explorative and co-worker-centered approach, it has been described and used as a “best case” of AI implementation in Denmark and abroad. Such designation suggests that the case is highly informative for researchers and professionals, providing insights into how to approach AI implementation soundly. Additionally, a case study approach is particularly well suited for producing knowledge of AI implementation as this approach enables the production of context-dependent knowledge integral to an individual’s ability to develop from a mere rule-follower to a virtuoso expert. CoD is located under the municipality’s Technical and Environmental Administration. Twenty-eight professionals work at CoD. These are divided into four teams: SEO & Web, Social media, Video & Graphics, and Communication. The pre-implementation process started in February 2023 and was set to finish in January 2024. The first author learned about the project through a colleague contacted by the Head of Communication on LinkedIn to disseminate information about the pre-implementation process. The first author then contacted the Head of Communication, expressing an interest in studying the process. The initial meeting took place in September 2023. The meeting lasted an hour, during which the Head of Communication described the process and the background of it, including motivations and aims. The meeting was recorded using note-taking. As the project had been ongoing for about six months when the empirical part of this study commenced, the study relies on interviews with members of the two groups driving the process - namely the project group and the working group - as well as an observation of a working group meeting, and various documents as data sources (see below). In mid-November 2023, a semi-structured interview was conducted with the project leader, who initiated the project and acted as a member of the project group and the working group. The interview took place on the administration’s premises and lasted 41 minutes. The purposes of the interview were (i) to gather information and insights into the initiation of the pre-implementation process as well as the process as such, (ii) to get insight into the motivations behind the process, and (iii) to understand co-worker’s perceptions of the process. Thus, the interview themes included: the initiation of the project and motivations, the responsibilities of the project group and the working group, how activities were prepared, how frames and objectives were produced, and reception among co-workers. The interview was audio-recorded and transcribed verbatim. A formal semi-structured interview with the Head of Communication was conducted at the end of the learning process in January 2024 in the administration’s premises. The purposes of this interview were (i) to gather information and insights into the process, (ii) to understand the perceptions of the process, and (iii) to get information about the future adoption of GenAI. The interview thus covered the following main themes: clarification regarding the process, perceptions and experience, and future implementation. The interview lasted 59 minutes, was audio-recorded and transcribed verbatim. Furthermore, semi-structured interviews were conducted with two working group members in a working space at CoD. The purposes of these interviews were (i) to gather information and insights into the learning process and (ii) to understand interviewees’ perceptions of the process. Themes covered during the interviews included: approach to GenAI and personal experience of using it, perception and experience of the learning process, preparation and execution of activities, and the future implementation of GenAI. The interviews lasted 42–58 minutes and were audio recorded and transcribed verbatim. Additionally, the first author conducted a non-participant observation of a working group meeting in January to gain first-hand insight into their work. While she had been invited to observe two meetings that took place during the autumn, she had been hindered due to teaching obligations. In the January meeting, held in a meeting room at CoD, the Head of Communication, the project leader, and five working group members participated. The main topics of the meeting were the evaluation of the pre-implementation process thus far and what GenAI systems already were used or could be used. The meeting lasted 64 minutes and was audio-recorded and transcribed verbatim. The documents, nine in total, produced and used by the working group and Head of Communication, including PowerPoint presentations of the learning process, objectives, frames, and guidelines, as well as test lab material and the municipality’s Code for the Use of Artificial Intelligence were collected as these constitute relevant information sources for both the process and the output. Lastly, with his permission, the Head of Communication’s LinkedIn posts regarding the learning process have been gathered, comprising 31 posts and 33 Word pages. These were relevant to our understanding of the learning process as they comprise an additional source about the process and the Head of Communication’s perspectives and perceptions of GenAI implementation. 4.1. Analytical procedure. The empirical data, which included interviews, observations, documents, and LinkedIn posts, was analyzed using a four-step process. First, the entire material was read twice to gain a comprehensive overview of the data and the overall process. During this phase, the material was also organized chronologically. Second, a coding was performed. This coding was primarily guided by the operationalization of the theoretical framework and the following questions: What GenAI systems are being considered? (technology) What tasks are targeted for re-engineering? (tasks) What organizational reconfigurations are being considered or implemented? (structure) How are co-workers involved? (people) How do these elements (technology, task, structure, and people) influence one another? We also incorporated empirically derived codes to ensure a more nuanced analysis. Examples include “assessment”, “test labs”, and “carefulness”, which provided critical insight into the specifics of the pre-implementation process that a purely theory-based coding might have missed (see e.g., Lindlof & Taylor, 2002). This coding process led us to identify four main phases of the learning process: Constructing the foundation, Exploring and framing GenAI use, Probing the value of GenAI, and Considering future use of GenAI. In the fourth stage of the analysis, we examined which elements were present in each phase and how they were interconnected. In the next section of the paper, we present our findings. 5. Findings. In the following, we present the findings, organized around the four phases of the learning process. 5.1. Constructing the foundation of the learning process. The learning process was triggered by events in the external environment, namely the launch of ChatGPT, prompting a discussion of the potential implementation of GenAI and the creation of foundational structures to support this exploration. The first phase of the learning process thus entailed primarily three elements: people, technology, and structure. The seed of what eventually would become the learning process was planted around the time of the launch of ChatGPT in late 2022. The Head of Communication recounts that when ChatGPT was introduced, he was “very fascinated” and “definitely on the hype train”. Knowledgeable about the Head of Communication’s strong interest in GenAI, it was the project leader who first queried whether CoD should consider using GenAI (project leader). Three key motivations – being able to make informed decisions about the use of GenAI, a perceived responsibility to serve as organizational experts on communication technologies, and finding out whether GenAI could indeed offer the enhanced efficiency and productivity it is reputed to provide – led them to submit a request to the administration’s IT council for a ‘proof-of-concept’ assessment of integrating GenAI into CoD’s work processes. While ChatGPT was the primary technology that sparked the learning process, the project would explore the applicability of other GenAI systems as well, including MidJourney, Davinci, chatbase.co, and others. In March 2023, the council approved the project, which meant that resources were allocated to the digitalization department to support CoD’s project. The project’s approval also meant that new structures were constructed within CoD and the administration. These involved the formation of a project group responsible for leading and formulating the project’s framework, the establishment of communication channels across departments, and the creation of a working group that would ensure the involvement of co-workers in the process. The project group comprised the Head of Communication, the project leader, and a representative from the digitalization department, serving as the technical expert in the group. While the group first approached the project as a “traditional project”, impatiently looking to move quickly to GenAI adoption, the Head of Communication retells that they soon realized that such an approach would not be feasible and that there, in fact, was little reason to rush implementation. Instead, they decided to embrace what the Head of Communication referred to as a more “deliberate and thoughtful approach”, not driven by any perceived necessity to adopt GenAI but by an urge to build knowledge and understanding of entailed technologies to enable well-grounded decisions regarding their use. Therefore, they decided that this project would be conducted as a learning process. 5.1.1. Constructing the frames of the learning process. While technology projects typically aim to enhance efficiency, in their discussions, the project group explored other motivations for the project, too. These discussions revolved around values that the members of the project group described as already governing the work at CoD, including high-quality communication, efficiency and productivity, legitimacy among internal and external publics, employee well-being, and trustworthy communication. In addition, the group also investigated legal frames and regulations, including GDPR, the Danish National Strategy for AI, and the municipality’s Code for the Use of Artificial Intelligence. The latter two stipulate that “the public sector should use AI to offer world-class public services” (Agency for Digital Government, 2024). The municipality specifies that Table 1 Guiding principles of the learning process. AI should not compromise citizens’ trust but contribute to better service for citizens, release resources for core welfare support, and support professionals in efficiently addressing their tasks. Based on CoD’s values and the legal frameworks, the group formulated four guiding principles (see Table 1): The guiding principles did not only inform and shape activities in the project (see e.g., Section 5.4). According to the Head of Communication, they also had the effect of making the learning process more careful: “We used them to govern the process. They made us more cautious. We decided on these guiding principles because they’re important to us, right?” (Head of Communication). Most of the guiding principles regard ‘people’, pointing to how the project group sought to ensure that co-workers, but also external publics, would perceive the introduction and future use of GenAI as valuable and meaningful. For example, the second principle stipulates that those responsible for the learning process should be mindful of and take seriously the worries of some co-workers regarding GenAI’s potential impact on what competencies are valued. This principle also guided the handling of co-workers who were uninterested in or hesitant to delving into GenAI. For instance, those who were not “too happy about having to participate” (project leader) in test labs (see Section 5.3) only had to attend three of those nine arranged. In addition to the guiding principles, the project group also defined three objectives for the learning process, based on the motto to learn first by exploring, understanding, and developing skills, and only thereafter potentially implement GenAI (see Table 2): Around the time of the formulation of objectives, the Head of Communication contacted an expert on learning who assisted the project group in better grounding the learning process in learning theories. Based on this input, they began considering learning as a dynamic interplay between understanding and mastering. As presented in one of the PowerPoints, these two were seen to complement each other: “When we increase our knowledge, then we also strengthen the foundation to understand generative AI in communication practices. Concurrently, this makes us better equipped to master methods, principles, and technologies that, in turn, strengthen our understanding.” 5.1.2. Appointing a working group. The final part of the learning process’s foundation was the formation of a working group. This group was established in mid-spring and was responsible for planning, developing, and pursuing the activities that would ensure the attainment of the learning process’ objectives, namely workshops, the formulation of Frames and Guidelines for AI Use, test labs, and the assessment method. The project group thus assumed what Mumford (2006) refers to as a “democratic” approach to the implementation of technology, making co-workers central to the framing, organizing, and execution of the learning process. The working group consisted of two co-workers from each CoD team (SEO & Web, Social media, Video & Graphics, and Communication) whom the project group appointed based on their believed capacity to provide qualified input to the process. Both co-workers who were enthusiastic about the future adoption of GenAI and those with a more cautious or even critical approach were appointed. The project leader considered these latter Effect and quality: Generative AI should enhance our communication work. It should be advantageous for both the administration and the city. It should not degrade but create more engaging, timely, and comprehensible communication for the people of [municipality]. Well-being and security: Efficiency gained through automation is valuable, but not at the expense of our well-being. We must prioritize creative, quality-conscious, and meaningful work, fostering a sense of security regarding the use of technology. Openness: Our approach to using generative AI in communication is open and transparent. We are committed to sharing our experiences, successes, and challenges for the benefit of the administration, [municipality], and the wider world. Readiness: As communicators, we will prepare ourselves for the changes to better understand and navigate opportunities and risks. This is crucial to supporting the administration in the transformation that AI development entails. Adapted from Working group (2023b). Table 2 Objectives of the learning process. 1. We should enhance our knowledge. Practically and theoretically. 2. We should formulate relevant frames and guidelines for our work and be continuously adjust them as we gain more insights. 3. We should develop a method to assess how and to what extent generative AI can be integrated into all or part of our workflows. Adapted from Working group (2023b). voices essential for the working group’s work and the overall process, stating, “We must pose critical questions too. Often, these are the most important ones.” Furthermore, the project group’s decision to let the working group lead the process, including providing exercises and training for the entire department, resulted in the learning process being perceived as less top-down and more co-worker-centered (working group participant). This involvement also meant that the ability to more fully account for the uniqueness of the context, including individual co-workers, group, and work requirements, increased (cf. Bostrom & Heinen, 1977b). While there was a strong hype surrounding GenAI at the time of the Head of Communication and the project leader initiating the learning process, they did not succumb to it by hasty implementation. Instead, in the first phase of the learning process, they chose to assume an explorative and co-worker-centered approach to the use of GenAI that involved taking the time to construct a solid structure that both foregrounded and involved the people that the technology would impact. 5.2. Exploring and framing GenAI use. In the second phase of the learning process, the project and working groups started exploring GenAI through invited talks, and the working group formulated Frames and Guidelines for ethical GenAI use. Thus, this phase of the learning process was primarily focused on developing group members’ understanding, knowledge, and skills of the technology and further structure construction. 5.2.1. Developing understanding & knowledge of GenAI. While the Head of Communication, who had dedicated much time to developing knowledge and skills on ChatGPT on his own, gave an introductory presentation to the whole department in early spring, the other inputs were given by invited external guests and only to the working group. According to interviewees, these inputs proved fundamental for furthering the learning process, as the working group got educated about GenAI, the specific infrastructure they were to use in their hands-on explorations of GenAI systems, and ethical issues related to the technology. A data science consultant and advisor on GenAI talked about how the technology can be utilized in search engine optimization, sparking discussions among working group members about how potential changes in the administration’s website should be managed to enhance visibility and the implications of such adjustments. A representative for Microsoft Denmark, an expert on the cloud computing and infrastructure Microsoft Azure, which CoD would use when exploring GenAI, educated the working group on how to use Azure. Representatives for the working group later forwarded this information to co-workers during an “Azure workshop”. To ensure ethical GenAI use, an advisor on AI ethics was also invited. Echoing findings from previous studies on communication professionals’ attitudes towards AI and ethics, members of the working group identified this dimension as highly important, and the project leader noted that the talk significantly influenced many co-workers. The project leader recalls, “It was very clear that the ethical aspects were crucial”. Therefore, the advisor was re-invited to conduct a series of workshops on ethics based on a tool called “the digital ethical compass”. In short, the compass encourages professionals to surpass existing laws and regulations by fostering an “ethic consciousness”. By developing a common language that aids doubting and discussions among co-workers, the aim was to make professionals more attentive to ethical grey zones and dilemmas. The usefulness of these workshops for creating attentiveness to ethical GenAI issues is observable both in the Frames and Guidelines that the working group formulated (see next section) and in their assessment discussion of GenAI (see Section 5.4). 5.2.2. Formulating and formalizing ethical GenAI use. During this phase, the working group also engaged in further structure construction, by developing a governance tool titled Frames and Guidelines. This structure sought to define GenAI use, ensuring that neither the guiding principles nor legal frameworks were violated, thereby promoting ethical GenAI use. Frames and Guidelines commence with a list of fundamental rules that apply to all uses of GenAI, both internal and external. These rules encompass openness (always declare the use of GenAI), photodocumentary (refrain from using GenAI when documenting real situations), facts- and quality check (perform a fact and quality check whenever GenAI is used), and security (never provide models personal sensitive, assignable, or confidential data to models; Working group, 2023b). The frame categorizes how co-workers at CoD can use GenAI across four levels, each accompanied by guidelines specifying the use of GenAI (see Table 3): The usefulness of this categorization was initially planned to be discussed during test labs and revised based on these. However, during the labs, the working group realized that such evaluation was too early and should instead be conducted once GenAI has been integrated into CoD’s workflows. In the second phase of the learning process, the working group members deepened their understanding and knowledge of GenAI and established new structures. Both in later phases of the process and once GenAI is integrated into their daily work, such structures influence how the technology can be used and in which workflows it can be used. In other words, instead of simply adapting to technology, the project and working groups assumed control, demonstrating how people and structures effectively (can) frame and define what GenAI “is” within their specific organizational context. 5.3. Probing the value of GenAI. The third phase of the learning process primarily involved people, tasks, and technology. Through hands-on exploration of GenAI, co- Guidelines specifying the use of GenAI. Adapted from Working group (2023b). workers at CoD both developed their understanding and skills and assessed the applicability of GenAI integration into their workflows. This exploration took place during “test labs”. Test labs were dedicated to practically exploring if and how GenAI could assist co-workers in solving their work tasks. In addition, they also constituted a forum where co-workers could learn, develop their skills, and discuss issues and dilemmas related to the GenAI system they tested. In preparing the test labs, the working group developed several so-called “use cases”, defined as, “a general task area where we assess that generative AI can play a role” (Working group, 2023c) to be conducted during the labs. The working group had sought to formulate use cases that would be “as relevant as possible for co-workers” (working group participant). To that purpose, members of the group asked colleagues in their respective teams about what workflows or assignments they would find relevant to explore. Based on this input, the working group developed four to five assignments for each team. For instance, for the SEO/Web team, the working group created assignments that included: finding relevant keywords, creating a teaser for a web page, write a FAQ, and translating a Danish text into English. Each assignment detailed what GenAI systems to use (e.g., ChatGPT, Davinci, or MidJourney), described the specific assignment, and proposed helpful prompts. During the labs, co-workers were encouraged to write down comments, reflections, and questions sparked by the exercise. When finished, they were asked to evaluate the GenAI system by responding to questions such as: “Is it [the AI tool used] actually helpful?”; “Did you come up with some ideas you otherwise would not have?”; and “Describe your considerations and concerns about using [the AI tool used] in this way.” (Working group, 2023c). In groups, co-workers could immediately air and discuss their thoughts and reflections on the potential optimization gains from the explored GenAI systems, and the risks and ethical issues they perceived. Through the test labs, all co-workers had the opportunity to discuss what and how GenAI systems could be beneficial and what workflows might be reconfigured upon implementation. Their ability to contribute constructively to these discussions and evaluate the systems was grounded in the understanding and practical skills they developed during the labs. In essence, the test labs empowered co-workers by first building their knowledge of GenAI and then soliciting their assessments and opinions. By actively involving co-workers – who are most knowledgeable about workflows and potential users – in technology exploration, rather than merely educating them, the learning process likely decreased the risk of hidden post-implementation costs. Furthermore, test labs (re)affirmed some of those structures that had previously been constructed. For instance, several guiding principles were reinforced through test lab discussions focused on evaluating GenAI systems’ potential to enhance the efficiency and quality of CoD’s communication, increasing co-workers’ readiness to navigate opportunities and risks in the new GenAI landscape, and ensuring the meaningfulness of new work practices (see Section 5.1.1.). 5.4. Considering the future use of GenAI. The last phase of the learning process was planned to construct a final structure: the method enabling CoD to assess what GenAI systems to implement and into which workflows. When the working group reached this stage, they decided to integrate this method into an implementation plan that would guide decisions regarding future GenAI use. This plan would outline work tasks and workflows (e.g., translation, content for web, photographs) along with key parameters of concerns – such as legal, ethical, and efficiency considerations – that must be met before GenAI can be used for each specific task or workflow (Interview, Head of Communication). As the Head of Communication explained, a pivotal dimension of the implementation of GenAI involved the strategic motivation for its integration: Does it enhance the product? Is the task resolved more efficiently? Can it assist in reaching the right target groups? Moreover, most importantly, does it support co-workers in meaningful ways? During the evaluation meeting at the end of January, the working group discussed the implementation of GenAI, focusing on the tasks for which the technology could be directly utilized. The co-workers’ evaluations during test labs were used, and attending members shared their experience using GenAI in their workflows. The Head of Communication commented, “We [the working group as well as other co-workers] talked a lot about this. If we now should use it [GenAI], what do we want to use it for? Using the knowledge we now have. It was pretty amazing to hear how good understanding they [co-workers] have for related risks and quality issues.” The discussions revealed that GenAI was already being used at level 1 (for brainstorming, testing ideas, summarizing, etc.) and level 2 (for content creation; see Section 5.2.2.). For instance, there was a consensus that using GenAI for increased accessibility was an area where this technology was particularly effective and thus could be used immediately. One of the members shared that co-workers were already employing voice-to-text GenAI for subtitling videos and were now experimenting with various voices, accents, and languages. Image creation was another area where co-workers had begun to use GenAI for tasks previously performed with Adobe Photoshop. Interestingly, however, while the use of Photoshop had never required formal declarations, the use of GenAI prompted a discussion about the necessity of such declarations. In this discussion, the previously constructed structure Frames and Guidelines was explicitly invoked by one of the members to resolve the issue, contending that GenAI use at the second level, according to the document, should always be declared. The primary overall conclusion from the discussion was that the use of GenAI requires both preceding and succeeding tasks performed by humans. Indeed, none of the use cases that entirely replaced humans or generated communications from scratch demonstrated a quality, including risks and ethical considerations, sufficient for implementation. As the discussions and implementation plan imply, the implementation of GenAI is thus not happening as a “big bang” (Head of Communication) but rather through a gradual approach, ensuring a robust and sustainable integration into workflows for co-workers, the organization, and its publics, while minimizing hidden costs of technology implementation. Echoing the STS principle of incompletion (see Bednar and Welch, 2020, p. 15), this slow implementation aims to promote a continuous, deliberate, and mindful approach to co-workers’ use of GenAI. 6. Discussion. In this paper, we have explored how a communication department in a large Danish municipality approached the implementation of GenAI. Placing the study within the emerging CommTech literature and adopting an STS perspective, the study demonstrates how managers at the department chose to approach GenAI implementation in an explorative and co-worker-centered mode. Instead of hasty implementation, CoD spent a year on what they called a “learning process” through which managers and co-workers gained the knowledge and skills needed to collectively make informed decisions about what GenAI systems to use and in which workflows. Thereby, they increased the probability of meaningful use of GenAI and decreased the likelihood of post-implementation surprises. We contribute to the literature on AI implementation in public relations generally and to the CommTech literature specifically in two interrelated ways: (i) by foregrounding the importance of dedicating time to the exploration of GenAI, and (ii) by adopting a human-centered, democratic approach to this exploration. Previous research on (Gen)AI implementation in public relations often conveys a sense of urgency, urging professionals to accelerate the implementation to keep pace with rapid developments in the technology and with professionals in adjacent fields such as marketing and HR. The STS perspective adopted in this study and our findings highlight the value of devoting time to the pre-implementation phase and involving co-workers (i.e., the end users) in framing, organizing, and executing a learning process. Involving and empowering co-workers can not only enhance job satisfaction and well-being but also lead to more well-grounded decisions regarding (Gen)AI use. Although not all communication departments have the resources for a year-long learning process, the study suggests that slowing down implementation for thorough exploration can significantly reduce the risk of costly and burdensome issues. While Brockhaus et al. (2023) have shown that ad hoc approaches to CommTech implementation are common, this study strongly advises against such (Gen)AI adoption, as this can lead to greater resource expenditure on post-implementation troubleshooting and performance dysfunction. By offering an empirical case example of an AI implementation process and explicitly pointing out several concrete implementation measures ensuring that both technological and social aspects are considered and safeguarded, we contribute to the digitalization strand within public relations Kretschmer and Winkler (2024) label “transhumanist”. Previous literature within this strand has mainly offered conceptual treatises on AI’s possible consequences for public relations practices and the profession. Our empirically grounded analysis and subsequent findings thus inform and extend this, to date, predominately theoretical strand and offer avenues for further theorizing and exploration of the practical intricacies of AI implementation. By providing an in-depth breakdown of the different phases occurring during a real-case AI implementation initiative, we also contribute with a more profound understanding of the implementation of potentially disruptive communication technologies. Thereby, we also contribute to the digitalization strand Kretschmer and Winkler (2024) label “disruptive,” which seeks to make the ongoing digital transformation of public relations more intelligible and provide practical guidance to organizations and public relations departments looking to implement AI in their operations. Specifically, by providing an empirical case study, our study addresses the lack of studies on the implementation of CommTech, pointed out by previous research within this strand. To encapsulate the insights gained from our analysis, we suggest the concept and practice of Slow Implementation. Drawing inspiration from “Slow Movements” (e.g., Slow Food, Slow Science, Slow Education, and Slow Thought; see e.g., Andrews, 2008; Berg & Seeber, 2016; Frith, 2020), we contend that by deliberately slowing down implementations, professionals afford themselves the time to establish a solid foundation for AI implementation. Slow, as Honor ́e (2004) has noted, “is not about doing everything at a snail’s pace. It is about seeking to do everything at the right speed” (p. 15). This approach ensures the integration of suitable (Gen)AI systems into the appropriate workflows with a clear purpose, insulating professionals against hasty implementation stimulated, for example, by recent GenAI hype. In the following, we define Slow Implementation. Slow Implementation describes a cautious, thoughtful, and investigative approach to the (Gen)AI system(s) under consideration for adoption. It allows ideas and intentions to be thoroughly discussed, examined, rejected, or matured, ensuring sustainable implementations. As such, Slow Implementation demands reflexive skepticism, where professionals question the object’s relevance, scrutinize proclaimed benefits, and consider context-specific practicalities, intricacies, and ethical dimensions. Such questions may include: Is the (Gen)AI system necessary? Does it help us? Does it add tasks (e.g., quality, fact-checking)? Are we creating communication tasks solely to employ (Gen)AI? How might the implementation influence our communication with different audiences, influencing trust? Cultivation of deep focus is necessary to acquire the knowledge to pose critical, meaningful, and relevant questions about the (Gen)AI ́s place(s) in workflows, its consequences for co-workers, work structures, and organizing (Ångstr ̈om et al., 2023). Such deep focus entails a clear and long-term commitment dedicated to developing theoretical and practical knowledge about the (Gen)AI system(s) through a slow and steady process involving learning methods featured in the (Gen)AI implementation literature (e.g., Gregory et al., 2023; Ångstr ̈om et al., 2023) and guided by clear objectives. While these objectives are crucial to inform the process, they may change as participants gain more insight. However, in contrast to part of the implementation literature (e.g., Gregory et al., 2023; Ångstr ̈om et al., 2023), co-workers should be considered vital participants in the whole process, and representatives be included in the group responsible for it. Such participation not only enhances the quality of assessment of the (Gen)AI systems but is also likely to improve the quality of its implementation and use. Managers should, however, also carefully consider whether all co-workers must embrace and use (Gen)AI. A substantial and sustainable (Gen)AI implementation requires managers to balance between co-workers’ well-being and sense of security on the one hand and the adoption of (Gen)AI on the other. Again, this entails careful deliberation regarding the necessity of the (Gen)AI system(s) under consideration and the necessity of implementing (Gen)AI right away. Furthermore, Slow Implementation might attract more co-workers to engage with (Gen)AI, as the extended timeframe, explorative approach, and emphasis on learning can encourage co-workers to develop their knowledge and skills. This strategy may thus empower individuals by allowing them to adopt (Gen) AI in a more personalized manner, which traditional top-down implementation does not typically facilitate. Finally, Slow Implementation cultivates resilience against hype that can lead to hasty implementation (e.g., Huang et al., 2022; Nordstr ̈om, 2021). Previous studies in public relations suggest an interesting contradiction between researchers advocating for faster and broader adoption of (Gen)AI and professionals who, up to recently, have been cautious about implementing (Gen)AI. However, the massive hype that surrounded the launch of ChatGPT, as indicated by the tripled (Gen)AI adoption rate among public relations professionals, might have influenced not only the use of (Gen) AI but also the level of mindfulness – or lack thereof – in its use. Indeed, when hype hits, professionals in all sorts of industries may adopt a practice based on the imitation of others rather than the deliberate and careful consideration of its drawbacks and consequences. This was, for instance, initially the case with the Head of Communication at CoD. However, the thoughtful slowing down of the implementation process made the professionals involved more mindful of the complexities of implementation and the intricacies of (Gen)AI. Thus, amid the hype, Slow Implementation entails taking it slow or, after careful consideration, even saying no. While the cost for such practice may initially be perceived as high, fueled by a fear of falling behind (see Gregory et al., 2023; see also Rosa, 2013), those embracing it will benefit in the long run by significantly reducing the risk of post-implementation issues. Slow Implementation can, however, also be seen as what Rosa (2013) refers to as a “slowdown strategy”, aimed at “further acceleration of other processes” (p. 87). While Slow Implementation is an ongoing practice, one of its key advantages is that slowing down allows for the implementation of appropriate systems right away and a smoother implementation process. As a result, with minimal post-implementation issues, (Gen)AI systems can enhance efficiency, thereby accelerating activities, tasks, and workflows. 6.1. Limitations and future research. This study is one of the first in the field of public relations to explore the intricacies of GenAI implementation. However, more studies are needed, particularly in-depth longitudinal fieldwork that follows (pre) implementation processes on-site. Such studies should be conducted in different types (e.g., private, non-governmental, and political) and sizes of organizations to provide insight into various frames and challenges that influence adoption. In addition, the everyday use of (Gen)AI needs close investigation, focusing on its benefits and perils for public relations tasks and activities. These studies should also delve into the implications for work and power relations among professionals. Furthermore, while it is easy to assume that overcoming initial barriers will ensure smooth (Gen)AI implementation, such an assumption is rejected by Ångstr ̈om et al. (2023), who point out that: “Gaining experience does not lessen the trials and tribulations of AI implementation. Instead, increased maturity comes hand-in-hand with new challenges and, in some cases, exacerbates existing ones.” (Ångstr ̈om et al., 2023, p. 9). Thus, not only is the initial AI implementation vital to study, but also the everyday use of (Gen)AI. Together, such studies can significantly further knowledge and practices of sustainable GenAI implementation. Declaration of Competing Interest None. Acknowledgements. This paper was written whilst Emma Christensen was funded by the Velux Foundation, Denmark, funding no. 38916. Data availability. The data that has been used is confidential.