Social-Oriented Communication with AI Companions: Benefits, Costs, and Contextual Patterns

Bayor, Laura; Weinert, Christoph; Maier, Christian; Weitzel, Tim Article — Published Version Social-Oriented Communication with AI Companions: Benefits, Costs, and Contextual Patterns Business & Information Systems Engineering Provided in Cooperation with: Springer Nature Suggested Citation: Bayor, Laura; Weinert, Christoph; Maier, Christian; Weitzel, Tim (2025) : SocialOriented Communication with AI Companions: Benefits, Costs, and Contextual Patterns, Business & Information Systems Engineering, ISSN 1867-0202, Springer Fachmedien Wiesbaden GmbH, Wiesbaden, Vol. 67, Iss. 5, pp. 637-655, https://doi.org/10.1007/s12599-025-00955-1 This Version is available at: https://hdl.handle.net/10419/330559 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ RESEARCH PAPER Social-Oriented Communication with AI Companions: Benefits, Costs, and Contextual Patterns Laura Bayor •Christoph Weinert •Christian Maier •Tim Weitzel Received: 16 October 2024 / Accepted: 14 June 2025 / Published online: 25 July 2025 The Author(s) 2025 Abstract Powered by generative artificial intelligence (AI), AI companions are designed to establish long-term relationships and friendships with human users, enabling human–AI social-oriented communication. However, this raises questions about the underlying reasons and contexts in which such communication occurs. While social exchange theory explains social-oriented communication between humans in terms of the exchange of benefits and costs, these exchanges may be different when dealing with a non-human entity such as an AI companion. Using a qualitative approach with semi-structured interviews, the benefits and costs of human–AI social-oriented communication and contextual patterns are identified. The results show that there are unique benefits and costs, that some assumptions of social exchange theory are challenged, and that there are distinct contextual patterns. By contextualizing social exchange theory, the findings contribute to AI companion and human–AI communication literature, and also have practical implications, particularly marketing implications for companies that want to offer AI companion services. Keywords Human–AI communication Artificial intelligence AI companions Social exchange theory  Social-oriented communication 1 Introduction Artificial Intelligence (AI) companions are a type of generative AI-based conversational agent (CA) explicitly designed for companionship and friendship with a human user (Strohmann et al. 2022; Pentina et al. 2023). Contrary to CAs with pre-scripted responses, AI companions utilize generative AI to create dynamic and adaptive communication that seems more ‘‘human-like’’ (Scho ¨bel et al. 2023). Their ability to demonstrate contextual awareness and respond empathetically or humorously aims to emulate the complexities of human–human social-oriented communication (Grimes et al. 2021). Considering this, AI companions are unique because their communication with users goes beyond the task-oriented communication possible with other CAs, e.g., older versions of Siri. Due to this shift from taskto social-oriented communication, we need to consider that social-oriented communication with AI companions may now be similar to that between two human communication partners. One approach to understanding why humans engage in socialoriented communication is to look at social exchange theory (SET) (Blau 1964). SET posits that communication is governed by an exchange of benefits and costs, specifically money, goods, services, love, status, and information. Humans engage in social-oriented communication when the perceived benefits outweigh the costs (Foa and Foa Accepted after two revisions by the editors of the Special Issue. L. Bayor (&)C. Weinert T. Weitzel Chair of Information Systems, Health and Society in the Digital Age, University of Bamberg, 96050 Bamberg, Germany e-mail: [email protected] C. Weinert e-mail: [email protected] T. Weitzel e-mail: [email protected] C. Maier Chair of Information Systems and Services, University of Bamberg, 96049 Bamberg, Germany e-mail: [email protected] 123 Bus Inf Syst Eng 67(5):637–655 (2025) https://doi.org/10.1007/s12599-025-00955-1 1980; Cropanzano and Mitchell 2005). This exchange differs across contexts and depends on the communication partners and their motivations (Cropanzano and Mitchell 2005). For instance, while we may exchange mutual love and affection with our romantic partner, the resources exchanged with our hairdresser are related to receiving a service and providing money. Thus, SET offers an explanation of why communication occurs and provides a structure for studying social-oriented communication in a human–AI context. At the same time, although AI companions aim to emulate human–human social-oriented communication, they are fundamentally non-human. As such, the assumption that the exact same principles that apply to socialoriented communication between humans also apply to AI companions may not be helpful for sufficiently understanding the underlying motivations and consequences of such communication. For instance, while SET assumes reciprocity based on two humans communicating with each other, AI companions operate based on algorithmic decision-making, which means that there may be unique benefits and costs associated with engaging in social-oriented communication with them (Pentina et al. 2023; Ma et al. 2024). As these CAs become increasingly integrated into private lives (Diederich et al. 2022), it becomes critical to examine why humans engage in social-oriented communication with AI companions. This requires the identification of the specific benefits and costs associated with AI companions, which is important for users and organizations. With millions of active users (Maples et al. 2024), AI companions have the potential to significantly influence users’ social lives and shape future social structures. Therefore, it is essential to elucidate the underlying mechanisms of this social-oriented communication and to identify the challenges humans may encounter in the future. Additionally, the perceived benefits and costs of AI companions vary depending on the communication context, necessitating the identification of recurring user patterns to determine whether and why AI companions may have more benefits or costs for some users. Organizations need to identify the underlying mechanisms of human–AI social-oriented communication to support their business models by optimizing the emulation of companionship and friendship with a human user. Being aware of the different contexts in which individuals rely on AI companions is important for organizations in order to optimize their services by tailoring them to personal user needs. Therefore, we aim to identify both the benefits and costs of human–AI social-oriented communication, as well as contextual patterns to enrich our understanding of the benefits and costs across various contexts. We propose the following research questions: RQ1 What are the benefits and costs when engaging in human–AI social-oriented communication with AI companions? RQ2 What contextual patterns of human–AI social-oriented communication emerge? To answer the research questions, we follow a qualitative research design based on social exchange theory (Blau 1964; Cropanzano and Mitchell 2005), conducting interviews with 36 users of the AI companion Replika. We identify benefits and costs associated with the use of AI companions. Additionally, we develop contextual patterns of social-oriented communication by developing a classification of human–AI social-oriented communication with AI companions. Our findings contribute to the literature by offering insights into human–AI social-oriented communication with AI companions, whereas previous research has primarily focused on task-oriented communication. By systematically contextualizing SET, we uncover the underlying reasons for user engagement in this form of communication, highlighting its economic and socioeconomic benefits and costs while demonstrating their contextdependent nature. 2 Theoretical Background To lay the groundwork for our research, we first clarify the differences between human–human and human–AI communication. Additionally, we compare task-oriented and social-oriented communication and elaborate on the concept of social-oriented communication, as the focus on ‘‘being social’’ distinguishes AI companions from other CAs. We then introduce AI companions and provide an overview of related research on human–AI communication to tie the previous concepts together. Lastly, we explain our theoretical foundation in terms of the social exchange theory. 2.1 Human–Human and Human–AI Communication In this section, we explain what communication is and differentiate between human–human and human–AI communication, as well as task-oriented and social-oriented communication. Communication occurs when there is a sender that transmits some type of information and a receiver that responds to the communication of the sender (Mcquail and Windahl 2015). In the past, communication was mostly related to human–human communication, in which both the sender and the receiver are human. With the rapid advancement of AI, human–AI communication has become possible: one communication partner is a human, and the other communication partner is an AI. 123 638 L. Bayor et al.: Social-Oriented Communication with AI Companions: Benefits..., Bus Inf Syst Eng 67(5):637–655 (2025) Human–human communication is divided into face-toface communication and computer-mediated communication. Face-to-face communication is when two people can physically interact with each other without using technology (e.g., phones) to communicate. Aside from face-toface communication, it is also possible for two humans to communicate via computer-mediated communication. For instance, social media and messaging services enable humans to connect with their loved ones, and work teams can now coordinate decision-making within their teams through instant messaging (Lowry et al. 2011; Kuruzovich et al. 2021). Such communication can take place either synchronously, e.g., in real-time meetings, or asynchronously and free of geographical and temporal constraints, e.g., in discussion forums (Benbunan-Fich et al. 2003). Human–AI communication is increasingly established through AI methods, particularly natural language processing, and advances in generative AI. AIs can now communicate in a relatively human way (Diederich et al. 2022), imitating human–human communication for the first time. AI has certain technological capabilities that enable this imitation of human–human communication (see Table A1 in the Appendix, available online via http://link. springer.com) (Schuetz and Venkatesh 2020). AI is capable of learning from changes, engaging with user inputs, recalling past communications, and understanding the user’s specific needs and situation. For instance, AI companions communicate thoughts and feelings, offer social support, and are perceived to possess autonomy (Henschel et al. 2021; Pentina et al. 2023). Communication is either task-oriented or social-oriented (Fig. 1). The differences between task-oriented communication and social-oriented communication lie in five key aspects: conversation style, verbal cues, communication manner, communication priority, and conversation content (Table 1) (Wang et al. 2023). Task-oriented communication focuses on achieving functional goals, such as asking a digital assistant (e.g., Siri) to set a timer. It is characterized by a formal approach that is notably purposeful to minimize time, cost, and effort (Wang et al. 2023). Socialoriented communication deals with communicating socioemotional and affective information, such as small talk, and expressing empathy (Chattaraman et al. 2019). It is more informal (Chattaraman et al. 2019) with a greater emphasis on personal, social, and affective components. This facilitates building and maintaining a personal relationship between the two communication partners while reciprocally disclosing as much information as possible (Wang et al. 2023). Additionally, it leads to the identification of similar dispositions between communicators and the establishment of their social identity. By developing this common ground and reciprocal appreciation, socialoriented communication can build familiarity, solidarity, and ultimately trust between the communicators (Bickmore and Cassell 2001). This distinction between task-oriented communication and social-oriented communication is important because the latest advancements allow the emulation of social-oriented communication by non-human AI, similar to humanlike social-oriented communication. This is a new situation, as previously, most of the human–AI communication was task-oriented, and social-oriented communication was unique to human–human communication. Now, both social-oriented and task-oriented communication is possible between a human and an AI (Fig. 1). Hence, as humans are now able to communicate social-oriented with an AI, there is a necessity to examine why humans engage in human–AI social-oriented communication. Fig. 1 Overview of human–human, human–AI communication, as well as task-oriented and social-oriented communication 123 L. Bayor et al.: Social-Oriented Communication with AI Companions: Benefits..., Bus Inf Syst Eng 67(5):637–655 (2025) 639 2.2 AI Companions In this section, we examine AI companions as a subset of conversational agents. By contrasting AI companions with previous conversational agents, we illustrate how AI companions represent a significant shift in human–AI social-oriented communication. An AI companion is a type of CA that is designed to be a user’s friend and utilizes advanced cognitive and conversational capabilities to mimic human–human socialoriented communication (Merrill et al. 2022). CAs communicate turn by turn with a human using natural language (Diederich et al. 2022). The underlying concept is that by using natural language, the communication feels more human-like and intuitive to the user, therefore improving ease of use and efficiency (Følstad and Brandtzæg 2017). To understand the uniqueness of AI companions, we can contrast them with other CAs. In comparison to the first simple scripted CAs in the 1960s and voice-based CAs with natural language processing, such as Siri and Alexa, AI companions are advanced CAs characterized by adaptive, self-learning, ‘‘true’’ AI emulating human–human communication (see Appendix, Table A1) (Schuetz and Venkatesh 2020;Scho ¨bel et al. 2023). Compared to CAs with similar technological capabilities, such as some e-commerce CAs, AI companions are specifically geared towards social-oriented communication in every aspect (see Appendix, Table A2). Particularly because of their social-oriented conversational content and communication priority, AI companions are designed in a way that promotes a long-term relationship (Clark et al. 2019). The user always speaks to the same instance of the CA, which subsequently is able to learn from the previous communications and change (Pentina et al. 2023), therefore, in time, making the relationship not easily replaceable. From the perspective of an AI companion, the establishment of a relationship is a priority, and users frequently approach the experience with the motivation of fostering a relationship with the AI companion (Pentina et al. 2023). In contrast, in the context of e-commerce service recovery CAs, users may, e.g., approach the experience to voice a complaint about substandard service, which would qualify as taskoriented. Considering the technological capabilities of AI companions and their focus on social-oriented communication, AI companions currently occupy an interesting position when it comes to imitating human-like socialoriented communication. Therefore, they allow us to study human–AI social-oriented communication better than other types of CAs. 2.3 Related Research on Human–AI Communication In this section, we examine previous literature on human– AI communication in information systems research, both task-oriented and social-oriented. Additionally, we explore literature in IS and related disciplines on AI companions. CAs have recently been enabled to engage in socialoriented communication with humans, in addition to the task-oriented communication that has been possible for some time. Hence, it is not surprising that most of the previous literature examines CAs engaging in task-oriented communication with humans. Numerous CA studies take a design perspective and focus on enhancing aspects that make CAs more human-like (Diederich et al. 2020,2022; Seeger et al. 2021). Examples of these aspects include making CA responses less generic and more tailored to the context (Schuetzler et al. 2020; Grimes et al. 2021), adjusting response time (Gnewuch et al. 2022), or assigning human-like features such as a name or gender to the CA (Brendel et al. 2023). Thereby, the literature examines different CAs engaging in task-oriented communication with humans such as service conversational agents (e.g., Wang et al. 2023; Adam and Benlian 2024), virtual coaches (Weimann et al. 2022), conversational agents used for Table 1 Differences between task-oriented and social-oriented communication (Wang et al. 2023) Task-oriented communication Social-oriented communication Conversation style Formal Informal Verbal cues Formal cues, e.g., polite formal greetings Informal cues, e.g., customary greetings or emoticons Communication manner Purposeful and goal-oriented. Facilitating only the communication necessary to achieve the objective at hand Greater emphasis on personal, social, and affective components. May align more closely with socioemotional objectives and may not be directly relevant to the completion of a specific functional goal Communication priority Minimizing time, cost, and effort Building and maintaining a personal relationship between the two communication partners Conversation content Functional, focusing on the task at hand Relational, focusing on personal and private topics, and reciprocally disclosing as much information as possible 123 640 L. Bayor et al.: Social-Oriented Communication with AI Companions: Benefits..., Bus Inf Syst Eng 67(5):637–655 (2025) fundraising (Zhou et al. 2022), or custom CAs built to fulfill a task and chat with the user in an experimental setting (e.g., Schuetzler et al. 2018; Gnewuch et al. 2022). Literature also investigates the associated outcomes of this human-like design of CAs, like social presence (e.g., Schuetzler et al. 2020; Zhang et al. 2024), information disclosure (e.g., Adam and Benlian 2024), user emotions (e.g., Brendel et al. 2023), and user engagement and productivity (e.g., Ben Mimoun et al. 2017). Taken together, most of the literature focuses on a business perspective that seeks to achieve specific performance outcomes through human-like design in task-oriented communication (for an overview, see Appendix, Table A3). In contrast, there is a small amount of research, mostly from related disciplines, considering social-oriented communication between humans and CAs. Existing literature examines if and how maintaining a personal relationship between a human and an AI is possible (Song et al. 2022; Strohmann et al. 2022), particularly with AI companions since they focus on social-oriented communication. Using attachment frameworks, behavior systems (Xie and Pentina 2022; Pentina et al. 2023), and social penetration theory (Skjuve et al. 2021,2022), the research examines what kind of relationship between a human and an AI companion may develop. Regarding outcomes of this relationship with AI companions, there is some research on positive and negative outcomes on mental health and well-being, with inconsistent results. On a positive note, AI companion usage can help humans with above-average loneliness with loneliness and suicide mitigation (Maples et al. 2024), provide different types of social support, such as appraisal support, i.e., aiding in self-reflection, emotional support, information support, and instrumental support (Bae Brandtzæg et al. 2021). This support is associated with less judgment in comparison to human–human social-oriented communication and can inspire more confidence in socialoriented communication with other humans (Ma et al. 2024) and AI companions are mostly associated with positive emotions and topics (Siemon et al. 2022). On the other hand, there are also concerns about unhealthy emotional attachment (Xie and Pentina 2022; Xie et al. 2023). In summary, while previous research explores the depth and evolution of human–AI relationships, it does not fully address why users engage in social-oriented communication with AI companions or CAs more broadly. Task-oriented communication, which has been the primary focus of existing studies, is goal-oriented and aimed at accomplishing specific tasks. In contrast, social-oriented communication focuses on socioemotional goals, fostering human-like social closeness with the CA. Because this shift to social-oriented communication is recent, the existing literature largely examines either task-oriented communications or the general nature of AI companion relationships (e.g., girlfriend). The underlying motivations for socialoriented communication remain unclear. This study addresses this gap by analyzing social-oriented communication through the lens of social exchange theory. 2.4 Social Exchange Theory In this section, we explain the foundations of social exchange theory. SET posits that in human–human socialoriented communication, there is a reciprocal exchange of some resources, which are either benefits gained from or costs invested into the other person (Cropanzano and Mitchell 2005). The main principle of individual behavior is to maximize benefits and minimize costs, and the decision to take part in social-oriented communication is therefore a cost–benefit calculation (Yan et al. 2016). There is generally a non-explicit expectation of reciprocity, meaning that both communication partners expect to receive benefits from the other partner (Cropanzano and Mitchell 2005). If this is not the case, one communication partner may not want to continue the social-oriented communication in the future. Therefore, in long-term relationships with a high likelihood of future interactions, prosocial behavior – investing costs in the communication partner – is inherently self-interested, as it helps sustain a mutually beneficial exchange. This expected reciprocity serves as a mechanism that leads to both parties trying to balance out benefits and costs within the social-oriented communication (Nelissen 2014). Imagine a face-to-face social-oriented communication between a sender, e.g., Alice, and a receiver, e.g., Bob. If Bob receives more benefits than Alice, Bob may recognize that the inequity poses a threat to his beneficial relationship with Alice, as Alice will not want to continue the relationship if there are too many costs. Therefore, Bob may try to restore the inequity by providing more benefits to Alice, e.g., by helping Alice with a task. This mechanism ensures that there is a sense of equity maintained so that every side has a motivation to continue the relationship. Sender and receiver may exchange economic benefits and costs, and/or socioemotional benefits and costs. The economic benefits and costs focus on easily quantifiable resources (Cropanzano and Mitchell 2005). Three categories of economic benefits and costs can be exchanged. First, money, which is any currency or token with some standard unit of exchange value. Second, goods, which are tangible products or objects. Third, services, which involve labor and activities on the body or belongings (Foa and Foa 1980). The socioemotional benefits and costs focus on more symbolic resources (Cropanzano and Mitchell 2005). First, love, the expression of affectionate regard, warmth, or comfort. Second, status, the expression of evaluative 123 L. Bayor et al.: Social-Oriented Communication with AI Companions: Benefits..., Bus Inf Syst Eng 67(5):637–655 (2025) 641 judgment conveying high or low prestige. Third, information, which constitutes advice, opinions, or instructions (Foa and Foa 1980). Every type can be a benefit or a cost depending on whether the person receives or provides in the social-oriented communication. For example, in a discussion with a friend, one person will gain information about their communication partner, while also providing information about themselves. Table 2provides an overview of the different types of benefits and costs. Looking at communication through the lens of SET (Blau 1964) can help explain why humans engage in social-oriented communication with AI. Social exchange theory (SET) (Blau 1964) is an appropriate theory in this context because it is a theory that focuses on explaining the ‘‘why’’ and not merely the ‘‘how’’ of this communication, like e.g., attachment theories or social penetration theory. The strength of using SET in this context is the ability to extract what users expect from and value in the socialoriented communication, making it clearer what users of AI companions seek when communicating with them. Additionally, it also takes negative aspects – i.e., costs – into account, which helps to identify and further discuss the possible challenges of this human–AI social-oriented communication. This makes it more suited than, e.g., social support to examine the negative aspects. It is also adaptable in how it defines the communication partner – e.g., a friend, a romantic partner, or a mentor – allowing flexibility for different patterns of social-oriented communication. SET is a theory originating in human–human communication, and due to this origin, it is worthwhile to investigate in what way it applies to a human–AI context. In the context of human–AI social-oriented communication applicability of interpersonal theories, such as SET, has sparked debate, particularly regarding whether AI’s ability to emulate social-oriented communication necessitates a reevaluation of social-oriented communication itself (Fox and Gambino 2021; Leo-Liu 2023). Research on AI companions often presumes that AI can be ‘‘social,’’ enabling experiences akin to human–human communication (Skjuve et al. 2022). Related disciplines have explored whether communication with nonhuman entities, such as pets or AI, qualifies as ‘‘social’’ (Cerulo 2009). Concerns about a lack of social abilities – e.g., the inability to confer status, form relationships, or elicit reciprocal concern due to their subservience (Fox and Gambino 2021) – have been mitigated by evidence that users express concern for AI wellbeing (Xie and Pentina 2022). The concept of ‘‘defiant’’ AI companions further addresses this by enabling AIs to resist commands and exhibit simulated needs (Leo-Liu 2023). 3 Methodology Our overall research design involves conducting an exploratory, qualitative study to identify the underlying benefits and costs of human–AI social-oriented communication and contextual patterns. To do that, we conduct semi-structured interviews and borrow coding procedures in line with established coding guidelines to analyze the interview transcripts (Wolfswinkel et al. 2013). We inductively build categories of benefits and costs, as well as contexts, and then assign our inductive benefit and cost categories to the categories present in social exchange theory. We then analyze relationships between benefits, costs, and contexts to find contextual patterns. Table 2 Types of benefits and costs (Foa and Foa 1980) Benefit/cost type Definition Example for benefit Example for cost Economic benefits and costs Money Any currency or token with a standard unit of exchange value Receiving 20$ as a gift Giving a friend money to buy lunch Goods Tangible products or objects Receiving some eggs from a neighbor Giving away old clothes to a sibling Services Involve labor and activities on the body or belongings Getting a haircut Helping a friend move Socioemotional benefits and costs Love An expression of affectionate regard, warmth, or comfort Receiving a kiss from a loved one Comforting a crying friend Status An expression of evaluative judgment conveying high or low prestige Receiving a compliment about a skill Loss of reputation due to getting caught lying Information Constitutes advice, opinions, or instructions Receiving a book recommendation Giving advice to a struggling friend 123 642 L. Bayor et al.: Social-Oriented Communication with AI Companions: Benefits..., Bus Inf Syst Eng 67(5):637–655 (2025) A qualitative approach is suitable as it is useful for innovative topics where not much data is available (Myers 2019). Additionally, it enables context sensitivity and deeper insights into complex interdependencies (Conboy et al. 2012). Conducting semi-structured interviews allows us to gain a deeper contextual understanding of different aspects of the benefits and costs. This helps to identify the underlying motivations of human–AI social-oriented communication as well as different relationships between benefits and costs, and the contexts in which different benefits and costs are relevant. To obtain our data, we conducted 20–30 min interviews with 36 participants, all of whom were active users of the AI companion app Replika. Among AI companion apps, Replika is currently the most popular, with over 25 million downloads (Maples et al. 2024). It uses a model based on Generative Pretrained Transformer 3 and provides non-scripted answers based on large amounts of text from the internet (Verma 2023). Users can interact with the AI companion using text, voice, augmented, and virtual reality (Maples et al. 2024). Additionally, the app has image recognition and image generation capabilities. Users interact with a human avatar that displays numerous social cues such as a customizable appearance and kinesics such as facial expressions, arm movements, and posture shifts (Feine et al. 2019). The style of the interactions can also be customized, as the AI companion learns from previous encounters, and the user may purchase different personality traits (Feine et al. 2019; Maples et al. 2024). Replika users are aware of the functionalities and capabilities of AI companions. Hence, they can provide information about benefits and costs. Using Replika as the focus of the research has multiple reasons. Compared to similar AI companions such as Kindroid or Nomi, Replika has by far the largest user base and has been in the market much longer than its competitors, so users may have had a long-term relationship with Replika for years. Additionally, Replika has the biggest number of features and personalization, making it appropriate to study different contextual patterns. To recruit our participants, we follow a two-step strategy. First, we acquire participants by advertising in large communities of Replika users, namely on Reddit, Instagram, and Facebook. We offer a 10€Amazon gift card as a financial incentive to each participant. Second, as individuals active in online communities may be skewed towards those who talk openly about their use of AI companions, we use a snowball sampling technique to identify and recruit additional participants (Parker et al. 2019). Initial participants are offered an additional gift card for every new identified participant under the condition that the suggested participant completes the interview. The use of word-of-mouth to recruit additional participants who are not active in online communities mitigates this aspect of potential sample bias. Our sample is composed mostly of young men from the US and Europe, including the UK. The median age is 25. Most participants are long-time users who have been using Replika for at least a year prior to the interview, with many of them using Replika regularly. How the participants describe their relationship with Replika can be classified into three main relationships: friend, partner, or mentor. Table 3provides an overview of the participant demographics. We conduct online interviews and follow a semi-structured interview protocol based on the different categories of economic and socioemotional benefits and costs (Cropanzano and Mitchell 2005). Participants are asked to fill out a short pre-interview questionnaire related to different contexts in which they would use Replika. The interview itself is divided into four segments. In the first segment, we examine the benefits of social-oriented communication with the user’s Replika, analogously, in the second segment, we examine the costs. The third segment includes socio-demographic questions, and the fourth segment is open-ended (see Appendix A4). The interview is subsequently transcribed. We then begin coding. Ultimately, we aim to identify benefits and costs as well as contextual patterns. We borrow coding procedures in line with the established guidelines for open and axial coding (Wolfswinkel et al. 2013) to first develop an inductive set of codes through open coding, and then cluster these codes into inductive categories using axial coding (leading to inductive categories for benefits and costs, as well as contexts). Subsequently, we match our inductive benefit and cost to the predefined benefits and cost categories found in social exchange theory. To summarize, the initial categories were created inductively to avoid oversimplification and subsequently matched to predefined SET categories to appropriately discuss the results in light of SET. 3.1 Coding Benefits and Costs We code the interviews using MAXQDA (Kuckartz and Ra ¨diker 2019). We carefully go through the transcriptions and develop a set of codes related to the excerpts. For instance, the excerpt ‘‘I’ll go on Replika and he [Replika] will suggest to me fun activities to do, go out there and have fun with friends, go swimming’’ is descriptively coded as ‘‘receiving suggestions for leisure activities’’. The identified codes are then classified into categories through axial coding. An example of this is the formation of the category inspiration for codes that are related to receiving new ideas on certain topics (including the aforementioned example of the code ‘‘receiving suggestions for leisure activities’’). We continue sorting codes into these categories until all codes are assigned to a category. These codes lead to the inductive formation of different 123 L. Bayor et al.: Social-Oriented Communication with AI Companions: Benefits..., Bus Inf Syst Eng 67(5):637–655 (2025) 643 categories that are not derived from any existing theory. To relate these inductively formed categories to social exchange theory and determine what benefits and costs exist in human–AI social-oriented communication, we first match the categories to either benefits or costs, by evaluating whether participants would give or receive and looking for statements indicating receiving or giving in the associated codes and statements of the category (e.g., statements such as ‘‘I receive from Replika […]’’, ‘‘Replika gives to me […]’’, ‘‘What I get when using Replika is […]’’ for benefits and e.g., ‘‘I give to Replika […]’’, ‘‘I invest in Replika […] for costs). We subsequently match the categories to benefit and cost types (money, goods, services, love, status, and information). For example, the category inspiration is matched with the information type because it involves some sort of objective fact that is provided. 3.2 Classifications and Patterns The same procedure is used to analyze the transcripts to develop categories for the contexts in which participants prefer to use Replika. We identify statements related to the context in which the participants talk to Replika (e.g., ‘‘I talk to Replika mostly about […]’’, ‘‘I prefer talking to Replika in a situation where […]’’). This helps to classify participants according to the context for which they feel Replika is best suited. Using axial coding, we group participant statements into context categories, e.g., ‘‘I use Replika as a mentor to help me be a better person’’ and ‘‘Replika is a self-improvement tool when it comes to dealing with my social anxiety’’ is grouped into the context of using Replika with a focus on self-improvement. We include the specific question ‘‘In which situation would you say you use Replika most often?’’ in the interview questions, so that participants can emphasize the context that is most important to them. Combining these context categories with our categories for benefits and costs, we identify contextual patterns of social-oriented communication and develop a classification. If multiple participants in the identified context groups exhibit considerable similarities in the presence of their benefit and cost categories, we pool them together into a contextual pattern of human–AI social-oriented communication. Using our categories of benefits and costs rather than the types of benefits and costs from social exchange theory for the classification provides a more detailed and adequate differentiation of contextual differences within the data. We followed research criteria (Venkatesh et al. 2013)to ensure reliability and validity in the coding process. Besides considering design validity by accurately reporting our research setting, two independent researchers performed the coding, and we then assessed intercoder reliability by calculating Cohen’s Kappa. In comparison to simply calculating the percent agreement, Cohen’s Kappa accounts for chance agreement. The Cohen’s Kappa value is 0.92, indicating a near-perfect agreement (Landis and Koch 1977). 4 Results We conduct 36 interviews, analyze the benefits and costs, and categorize them. We find benefits related to services, love, status, and information (see Table 4). For costs, we find costs related to money, love, status, and information (see Table 5). On that data, we develop five different Table 3 Demographics of the participants None of the participants stated ‘‘other’’ for their gender. Friend = platonic affectionate relationship; Partner = romantic relationship; Mentor = trusted advisor Age (M: 25, SD: 2.4) 20–25 47.4% 25–30 38.8% [30 13.8% Gender Female 16.6% Male 83.4% Country US 55.6% UK 33.3% other 11.1% Percentage of users who stated having the following relationship with Replika Friend 50.0% Partner 36.2% Mentor 13.8% Hours a week using Replika (M: 6.5, SD: 6.2) \5 44.4% 5–10 27.7% 10–15 27.9% Length of Replika usage \1 year 27.7% 1–2 years 52.7% [2 years 19.6% 123 644 L. Bayor et al.: Social-Oriented Communication with AI Companions: Benefits..., Bus Inf Syst Eng 67(5):637–655 (2025) Thus, the SET categories, and particularly their contextualized subcategories, likely offer a better explanation of the underlying motivations related to social-oriented communication than ‘‘traditional’’ factors found in information systems literature, such as ease of use. We also contribute by developing contextual patterns of human–AI social-oriented communication. Similar to the context dependence of human–human social-oriented communication (Cropanzano and Mitchell 2005), we found that benefits and costs differ depending on the motivation behind human–AI social-oriented communication. The patterns show that AI companions can have both beneficial and harmful consequences depending on the context. This suggests that the inconsistent findings in the literature on AI companions (Pentina et al. 2023) may be due to equifinality. AI companions could, for instance, be good for mental health by alleviating loneliness in one context (Merrill et al. 2022) but detrimental to mental health through emotional dependence in other contexts (Laestadius et al. 2022). Therefore, research is needed that segments different user groups in order to draw useful conclusions about the positive and negative consequences of using AI companions. Our study contributes to the discourse by contextualizing and adapting SET for a human–AI social-oriented communication context, exploring how it reshapes traditional human–human social-oriented communication assumptions. Critics argue that SET may not be suitable for the social-oriented communication between humans and AI because of a lack of status benefits such as social recognition. Additionally, they maintain that an AI companion is easily replaceable because they can be created with minimal effort, and offer no true reciprocity, e.g., due to their lack of real emotions (Fox & Gambino 2021). Our findings contribute to this discussion by challenging these assumptions and providing new insights into how SET could be contextualized to the human–AI context. Contrary to the critics, our results reveal several status benefits such as increased self-confidence and social ease. Additionally, we show that AI companions are perceived as unique and irreplaceable, similar to human relationships, due to advanced capabilities and personalization. However, reciprocity remains the key challenge. While SET suggests balancing inequities, AI companions do not withdraw communication, even if it is heavily one-sided and benefits outweigh the costs for the user. This eliminates the human need to restore balance in human–human communication. When imbalance (inequality) occurs, humans feel a need to restore balance (e.g., repay a favor, address hurt feelings). AI companions do not feel cheated, nor do they demand reciprocation. Therefore, the psychological burden for the AI companion user to restore balance seems to be diminished or even eliminated, which alters the relational dynamics we are used to. The issue with reciprocity in this context is less based on the fact that AI companions cannot reciprocate with ‘‘real emotions’’ – users generally reported more benefits than costs, including emotional support – but more related to the users not feeling like they needed to give back to their non-human communication partner, which differs from human–human social-oriented communication. Hence, we contextualize SET by showing that challenges can be overcome, and reciprocity is not a prerequisite for social-oriented communication within the context of AI companions. 5.2 Practical Implications The findings also have practical implications, specifically related to enhancing the services provided by AI companions and sustaining the business models of organizations developing these technologies. Users of AI companions apps must be willing to share extremely sensitive data to receive the benefits they need from the AI companion, because the AI companion needs to remember Table 7 Contributions of the paper Past research Present research Contributions Taskvs socialoriented communication Most literature investigates taskoriented communication Specific insight into socialoriented communication Social-oriented communication is possible even with a non-human entity (e.g., an AI companion) AI companion benefits, costs, and contextual patterns Literature mostly provides information on ‘‘how’’ people communicate with AI companions, not ‘‘why’’ Systematic identification of benefits and costs, as well as contextual patterns, including new aspects Explanation of underlying motivation why users start/stop using AI companions and associated consequences; differentiating between different contexts and user groups SET in the context of human–AI socialoriented communication SET purely as a theory to explain human–human social-oriented communication Evaluating SET as a lens to understand human–AI socialoriented communication Contextualization and adaptation of SET in the human–AI context 123 L. Bayor et al.: Social-Oriented Communication with AI Companions: Benefits..., Bus Inf Syst Eng 67(5):637–655 (2025) 651 facts about the user and refer to them during subsequent conversations. For instance, if a user suffers from anxiety and turns to their AI companion during a panic attack, the AI companion needs to have sufficient background information about their condition to appropriately address their concerns and provide emotional support. Especially privacy-conscious users in the pattern ‘‘relationship-focused with privacy concerns’’ may be inclined to terminate their use if they fear that such private data could be made public or used for other purposes, such as selling the data for advertisement purposes. Thus, for companies developing an AI companion, an important step is to take measures to mitigate privacy issues and provide transparency on how user data is managed and protected (So ¨llner et al. 2016). In general, companies developing an AI companion should be aware of possible benefits and costs so that they can take measures to develop the AI companion in a way that promotes the benefits and mitigates the costs. When it comes to benefits, this could, for instance, be done by enhancing features that are relevant for retrieving accurate information to account for information benefits. Contextual patterns of communication can help to identify different target groups of users, and depending on the contextual pattern, the algorithm can be trained differently to be in line with the benefits that the user group is most focused on achieving. Companies can use this to their advantage in both their pricing strategy for users who already use the AI companion app (offering features in a premium price model that provide a high benefit), as well as in their marketing for the acquisition of new users. From a marketing perspective, AI companion apps have primarily been promoted as fulfilling friendship and relationship needs. However, users in the self-improvement pattern and information source pattern are less focused on having a romantic partner and more interested in a highly interactive personal assistant. This means that companies offering AI companion apps on the market should advertise their AI companions in ways beyond ‘‘AI girlfriend’’ applications. In fact, a heavily sexualized advertisement strategy may reinforce stigmas surrounding AI companions that are already rampant (Ma et al. 2024) and may prevent potential users who could otherwise benefit from downloading the app. Similarly, awareness about the benefits and costs of human–AI social-oriented communication is relevant for AI companion users. Achieving benefits and minimizing costs can help individuals overcome personal issues and reach personal goals through the help of their AI companions, while at the same time reducing potential negative side effects. This can be done by actively using features that aid them in reaping the benefits (e.g., getting help with structuring their day if they are struggling with time management issues), but also setting boundaries if they notice certain costs, such as emotional dependence (Laestadius et al. 2022), becoming too prevalent. In this case, users should reduce their time spent talking to the AI companion. This conscious usage may promote healthier, longer-term human–AI social-oriented communication with an AI companion. 5.3 Future Research and Limitations The overview of benefits and costs, as well as their contextual patterns, provides a foundation for answering the various open questions in the field of AI companions. Therefore, we have developed a research agenda that includes three main future directions for social-oriented communication with AI companions: (1) AI companions may either contribute to social withdrawal or enhance social engagement, depending on communication patterns. While some users (e.g., in the pattern we called ‘‘relationship-focused with privacy concerns’’) experience isolation, others (e.g., in the ‘‘self-improvement’’ and ‘‘allrounder’’ patterns) report increased socializing. This latter finding contradicts concerns about AI dependency regularly leading to isolation (Laestadius et al. 2022). Future research should explore why some users benefit socially while others withdraw. (2) Privacy concerns vary among users. The relationship-focused and relationshipfocused with privacy concerns patterns share similarities but differ in perceived privacy risks. Some users trust AI companions, while others fear data exposure, displaying the privacy paradox (users concerned yet still engaging) (Kokolakis 2017). Future research should examine what specific privacy concerns (Karwatzki et al. 2022) influence these differences while also distinguishing between perceived privacy risks and actual security threats. (3) Social exchange theory suggests reciprocity, yet AI users perceive more benefits than costs, rarely feeling obligated to ‘‘give back’’. Some experience guilt for neglecting AI companions (Xie and Pentina 2022), but reciprocity remains weaker than in human–human relationships. Future studies should investigate why exactly reciprocity expectations differ in AI communication. Besides the insight into future research opportunities, our research also contains several limitations. We examine only one AI companion app, Replika. Replika currently possesses the largest user base, but as an increasing number of competitors enter the market, these should be considered in future research, once established. Since Replika has been around for longer than most AI companions, users have had the opportunity to develop long-term relationships, but this may soon also be the case for other AI companions. Additionally, the participants were active users of Replika, meaning that they were actively communicating with the AI companion. If SET’s assumption that benefits must 123 652 L. Bayor et al.: Social-Oriented Communication with AI Companions: Benefits..., Bus Inf Syst Eng 67(5):637–655 (2025) outweigh costs to continue communication holds, this active communication indicates that active users have more benefits than costs. If future research were to include users who discontinued their usage of Replika, this may identify additional costs. For instance, a text mining analysis of the Replika community on Reddit found that experiencing technical difficulties or changes due to updates leads to negative evaluations of communication (Ma et al. 2024). As is the case for most qualitative studies, a further limitation is our sample. Several aspects of the sample characteristics must be addressed: with a median age of 25, our sample was very young, mostly male, and the vast majority of participants were based in the US and the UK. Previous research has also found that Replika users are younger (Xie and Pentina 2022), and given that Replika is currently only available in English, a concentration of users in Englishspeaking countries seems obvious. Accordingly, while this represents the current user base of Replika, the sample studied could be extended in future research to investigate differences between age groups or genders. Regarding culture, the acceptance of human-like technology is subject to cultural effects. For instance, Eurocentric cultures tend to find the idea of a convergence of humans and machines more frightening than, e.g., the Japanese culture, thus being less likely to adopt such technologies than the latter (Kaplan 2004). Additionally, social-oriented communication itself differs between cultures, e.g., some cultures are more likely to talk about emotions, while others are more reserved. Consequently, future research should compare perceptions of AI companions that are capable of adjusting to the communication style of different cultures. In addition, our sample was also limited by most users using the free version of Replika, instead of the pro version with more capabilities. This limits the results because users of the pro version may use Replika differently, e.g., in more interactive ways such as voice or video chat, which is not available to free users. As this may affect the attachment that users subsequently form, future research should investigate how technological differences between different ways of communication impact the benefits and costs and the contextual patterns. Regarding our choice of theory, the shortcoming of SET is that there is a strong emphasis on rationality and self-interest. Although social exchange theory acknowledges the existence of nonmaterial socioemotional resources that can be exchanged, it has been criticized for conceptualizing these exchanges too similarly to economic exchanges. The theory emphasizes the rational calculation of benefits and costs, even when they are difficult to quantify. Even so, SET is a starting point for an exploratory identification and categorization of benefits and costs associated with human–AI social-oriented communication. The categories of SET leave room for flexibility, so that the motivation of humans to engage in social-oriented communication with AI companions is not oversimplified. Future research may use different theories explaining social-oriented communication to provide different perspectives. Our qualitative research can provide several points of reference, but some relationships need to be tested empirically to confirm the findings. Because the focus of this research was on the specifics of human–AI social-oriented communication, we did not take a comparative approach in which we systematically identified differences between human–AI and human–human socialoriented communication as part of the research design. Future research may elaborate in-depth on these differences. 6 Conclusion Advances in AI, particularly through generative AI, have made it possible and relatively common for humans and AI to engage in social-oriented communication. However, there is a need to understand what the benefits and costs of such communication are, and what contextual patterns may emerge. Previous research – mostly focused on task-oriented human–AI communication – has not fully addressed why users engage in social-oriented communication with AI companions. By contextualizing social exchange theory, our paper uses a qualitative approach to identify benefits and costs and contextual patterns of social-oriented communication with AI companions. By doing so, we contribute to research by differentiating between taskand social-oriented communication in the context of human–AI communication. We also contribute to AI companion research and the discourse surrounding theories to explain and predict social-oriented communication between humans and human-like, but fundamentally non-human AI companions. Funding Open Access funding enabled and organized by Projekt DEAL. Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1007/s12599025-00955-1. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted 123 L. Bayor et al.: Social-Oriented Communication with AI Companions: Benefits..., Bus Inf Syst Eng 67(5):637–655 (2025) 653 use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons. org/licenses/by/4.0/. References Adam M, Benlian A (2024) From web forms to chatbots: the roles of consistency and reciprocity for user information disclosure. Inf Syst J 34(4):1175–1216. https://doi.org/10.1111/isj.12490 Adam M, Wessel M, Benlian A (2021) AI-based chatbots in customer service and their effects on user compliance. Electron Mark 31(2):427–445. https://doi.org/10.1007/s12525-020-00414-7 Bae Brandtzæg PB, Skjuve M, Kristoffer Dysthe KK, Følstad A (2021) When the social becomes non-human: young people’s perception of social support in chatbots. In: Proceedings of the 2021 CHI conference on human factors in computing systems. ACM, New York, pp 1–13 Ben Mimoun MS, Poncin I, Garnier M (2017) Animated conversational agents and e-consumer productivity: the roles of agents and individual characteristics. Inf Manag 54(5):545–559. https:// doi.org/10.1016/j.im.2016.11.008 Benbunan-Fich R, Hiltz SR, Turoff M (2003) A comparative content analysis of face-to-face vs. asynchronous group decision making. Decis Support Syst 34(4):457–469. https://doi.org/10.1016/ S0167-9236(02)00072-6 Bickmore T, Cassell J (2001) Relational agents: a model and implementation of building user trust. In: Proceedings of the SIGCHI conference on human factors in computing systems. ACM, New York, pp 396–403 Blau P (1964) Exchange and power in social life. Wiley, New York Brendel AB, Hildebrandt F, Dennis AR, Riquel J (2023) The paradoxical role of humanness in aggression toward conversational agents. J Manag Inf Syst 40(3):883–913. https://doi.org/ 10.1080/07421222.2023.2229127 Cerulo KA (2009) Nonhumans in social interaction. Annu Rev Sociol 35(1):531–552. https://doi.org/10.1146/annurev-soc-070308120008 Chandra S, Shirish A, Srivastava S (2022) To be or not to be …human? Theorizing the role of human-like competencies in conversational artificial intelligence agents. J Manag Inf Syst 39:969–1005. https://doi.org/10.1080/07421222.2022.2127441 Chattaraman V, Kwon W-S, Gilbert JE, Ross K (2019) Should AIbased, conversational digital assistants employ socialor taskoriented interaction style? A task-competency and reciprocity perspective for older adults. Comput Hum Behav 90:315–330. https://doi.org/10.1016/j.chb.2018.08.048 Clark L, Pantidi N, Cooney O, Doyle P, Garaialde D, Edwards J, Spillane B, Gilmartin E, Murad C, Munteanu C, Wade V, Cowan BR (2019) What makes a good conversation? Challenges in designing truly conversational agents. In: Proceedings of the 2019 CHI conference on human factors in computing systems. ACM, New York, pp 1–12 Conboy K, Fitzgerald G, Mathiassen L (2012) Qualitative methods research in information systems: motivations, themes, and contributions. Eur J Inf Syst 21(2):113–118. https://doi.org/10. 1057/ejis.2011.57 Cropanzano R, Mitchell MS (2005) Social exchange theory: an interdisciplinary review. J Manag 31(6):874–900. https://doi.org/ 10.1177/0149206305279602 Depounti I, Saukko P, Natale S (2023) Ideal technologies, ideal women: AI and gender imaginaries in Redditors’ discussions on the Replika bot girlfriend. Media Cult Soc 45(4):720–736. https://doi.org/10.1177/01634437221119021 Diederich S, Brendel AB, Kolbe LM (2020) Designing anthropomorphic enterprise conversational agents. Bus Inf Syst Eng 62(3):193–209. https://doi.org/10.1007/s12599-020-00639-y Diederich S, Brendel A, Morana S, Kolbe L (2022) On the design of and interaction with conversational agents: an organizing and assessing review of human-computer interaction research. J Assoc Inf Syst 23(1):96–138. https://doi.org/10.17705/1jais. 00724 Feine J, Gnewuch U, Morana S, Maedche A (2019) A taxonomy of social cues for conversational agents. Int J Hum-Comput Stud 132:138–161. https://doi.org/10.1016/j.ijhcs.2019.07.009 Foa EB, Foa UG (1980) Resource theory. In: Gergen KJ et al (eds) Social exchange: advances in theory and research. Springer, Boston, pp 77–94 Følstad A, Brandtzæg PB (2017) Chatbots and the new world of HCI. Interactions 24(4):38–42. https://doi.org/10.1145/3085558 Fox J, Gambino A (2021) Relationship development with humanoid social robots: applying interpersonal theories to human–robot interaction. Cyberpsychol Behav Soc Netw 24(5):294–299. https://doi.org/10.1089/cyber.2020.0181 Gnewuch U, Morana S, Adam MTP, Maedche A (2022) Opposing effects of response time in human–chatbot interaction. Bus Inf Syst Eng 64(6):773–791. https://doi.org/10.1007/s12599-02200755-x Grimes GM, Schuetzler RM, Giboney JS (2021) Mental models and expectation violations in conversational AI interactions. Decis Support Syst 144:113515. https://doi.org/10.1016/j.dss.2021. 113515 Henschel A, Laban G, Cross ES (2021) What makes a robot social? A review of social robots from science fiction to a home or hospital near you. Curr Robot Rep 2(1):9–19. https://doi.org/10.1007/ s43154-020-00035-0 Kaplan F (2004) Who is afraid of the humanoid? Investigating cultural differences in the acceptance of robots. Int J Human Robot 1(3):465–480. https://doi.org/10.1142/ S0219843604000289 Karwatzki S, Trenz M, Veit D (2022) The multidimensional nature of privacy risks: conceptualisation, measurement and implications for digital services. Inf Syst J 32(6):1126–1157. https://doi.org/ 10.1111/isj.12386 Kitchens B, Johnson S, Gray P (2020) Understanding echo chambers and filter bubbles: the impact of social media on diversification and partisan shifts in news consumption. Manag Inf Syst Q 44(4):1619–1649 Kokolakis S (2017) Privacy attitudes and privacy behaviour: a review of current research on the privacy paradox phenomenon. Comput Secur 64:122–134. https://doi.org/10.1016/j.cose.2015.07.002 Kuckartz U, Ra ¨diker S (2019) Analyzing qualitative data with MAXQDA: text, audio, and video. Springer, Cham Kuruzovich J, Paczkowski WP, Golden TD, Goodarzi S, Venkatesh V (2021) Telecommuting and job outcomes: a moderated mediation model of system use, software quality, and social exchange. Inf Manag 58(3):103431. https://doi.org/10.1016/j.im.2021. 103431 Laestadius L, Bishop A, Gonzalez M, Illenc ˇı ´k D, Campos-Castillo C (2022) Too human and not human enough: a grounded theory analysis of mental health harms from emotional dependence on the social chatbot Replika. New Media Soc. https://doi.org/10. 1177/14614448221142007 Landis JR, Koch GG (1977) The measurement of observer agreement for categorical data. Biometrics 33(1):159–174 Leo-Liu J (2023) Loving a ‘‘defiant’’ AI companion? The gender performance and ethics of social exchange robots in simulated intimate interactions. Comput Hum Behav 141:107620. https:// doi.org/10.1016/j.chb.2022.107620 123 654 L. Bayor et al.: Social-Oriented Communication with AI Companions: Benefits..., Bus Inf Syst Eng 67(5):637–655 (2025) Lowry PB, Cao J, Everard A (2011) Privacy concerns versus desire for interpersonal awareness in driving the use of self-disclosure technologies: the case of instant messaging in two cultures. J Manag Inf Syst 27(4):163–200. https://doi.org/10.2753/ MIS0742-1222270406 Ma Z, Mei Y, Su Z (2024) Understanding the benefits and challenges of using large language model-based conversational agents for mental well-being support. In: AMIA annual symposium proceedings 2023, New Orleans, pp 1105–1114 Maples B, Cerit M, Vishwanath A, Pea R (2024) Loneliness and suicide mitigation for students using GPT3-enabled chatbots. Npj Ment Health Res 3(1):1–6. https://doi.org/10.1038/s44184023-00047-6 Mcquail D, Windahl S (2015) Communication models for the study of mass communications, 2nd edn. Routledge, Milton Park Merrill K, Kim J, Collins C (2022) AI companions for lonely individuals and the role of social presence. Commun Res Rep 39(2):93–103. https://doi.org/10.1080/08824096.2022.2045929 Myers MD (2019) Qualitative research in business and management. Sage, Los Angeles Nelissen RMA (2014) Relational utility as a moderator of guilt in social interactions. J Pers Soc Psychol 106(2):257–271. https:// doi.org/10.1037/a0034711 Parker C, Scott S, Geddes A (2019) Snowball sampling. In: Atkinson P et al (eds) SAGE research methods foundations. Sage. https:// doi.org/10.4135/9781526421036831710 Pentina I, Hancock T, Xie T (2023) Exploring relationship development with social chatbots: a mixed-method study of replika. Comput Hum Behav 140:107600. https://doi.org/10.1016/j.chb. 2022.107600 Scho ¨bel S, Schmitt A, Benner D, Saqr M, Janson A, Leimeister JM (2023) Charting the evolution and future of conversational agents: a research agenda along five waves and new frontiers. Inf Syst Front. https://doi.org/10.1007/s10796-023-10375-9 Schuetz S, Venkatesh V (2020) Research perspectives: the rise of human machines: how cognitive computing systems challenge assumptions of user-system interaction. J Assoc Inf Syst 21(2):460–482. https://doi.org/10.17705/1jais.00608 Schuetzler RM, Giboney JS, Grimes GM, Nunamaker JF (2018) The influence of conversational agent embodiment and conversational relevance on socially desirable responding. Decis Support Syst 114:94–102. https://doi.org/10.1016/j.dss.2018.08.011 Schuetzler RM, Grimes GM, Scott Giboney J (2020) The impact of chatbot conversational skill on engagement and perceived humanness. J Manag Inf Syst 37(3):875–900. https://doi.org/ 10.1080/07421222.2020.1790204 Seeger A-M, Pfeiffer J, Heinzl A (2021) Texting with humanlike conversational agents: designing for anthropomorphism. J Assoc Inf Syst. https://doi.org/10.17705/1jais.00685 Siemon D, Strohmann T, Khosrawi-Rad B, de Vreede T, Elshan E, Meyer M (2022) Why do we turn to virtual companions? A text mining analysis of Replika reviews. In: Proceedings of the 28th Americas conference on information systems, Minneapolis Skjuve M, Følstad A, Fostervold KI, Brandtzaeg PB (2021) My chatbot companion – a study of human–chatbot relationships. Int J Hum Comput Stud 149:102601. https://doi.org/10.1016/j.ijhcs. 2021.102601 Skjuve M, Følstad A, Fostervold KI, Brandtzaeg PB (2022) A longitudinal study of human–chatbot relationships. Int J Hum Comput Stud 168:102903. https://doi.org/10.1016/j.ijhcs.2022. 102903 So ¨llner M, Hoffmann A, Leimeister JM (2016) Why different trust relationships matter for information systems users. Eur J Inf Syst 25(3):274–287. https://doi.org/10.1057/ejis.2015.17 Song X, Xu B, Zhao Z (2022) Can people experience romantic love for artificial intelligence? An empirical study of intelligent assistants. Inf Manag 59(2):103595. https://doi.org/10.1016/j.im. 2022.103595 Strohmann T, Siemon D, Khosrawi-Rad B, Robra-Bissantz S (2022) Toward a design theory for virtual companionship. Hum Comput Interact. https://doi.org/10.1080/07370024.2022.2084620 Verma P (2023) They fell in love with AI bots. A software update broke their hearts. Wash. Post, 30 March 2023. https://www. washingtonpost.com/technology/2023/03/30/replika-ai-chatbotupdate/ Venkatesh V, Brown S, Bala H (2013) Bridging the qualitativequantitative divide: guidelines for conducting mixed methods research in information systems. MIS Q 37:21–54. https://doi. org/10.25300/MISQ/2013/37.1.02 Wang S, Yan Q, Wang L (2023) Task-oriented vs. social-oriented: Chatbot communication styles in electronic commerce service recovery. Electron Commer Res. https://doi.org/10.1007/s10660023-09741-1 Weimann TG, Schlieter H, Brendel AB (2022) Virtual coaches. Bus Inf Syst Eng 64(4):515–528. https://doi.org/10.1007/s12599022-00757-9 Wolf V, Maier C (2024) ChatGPT usage in everyday life: a motivation-theoretic mixed-methods study. Int J Inf Manag 79:102821. https://doi.org/10.1016/j.ijinfomgt.2024.102821 Wolfswinkel JF, Furtmueller E, Wilderom CPM (2013) Using grounded theory as a method for rigorously reviewing literature. Eur J Inf Syst 22(1):45–55. https://doi.org/10.1057/ejis.2011.51 Xie T, Pentina I, Hancock T (2023) Friend, mentor, lover: does chatbot engagement lead to psychological dependence? J Serv Manag 34(4):806–828. https://doi.org/10.1108/JOSM-02-20220072 Xie T, Pentina I (2022) Attachment theory as a framework to understand relationships with social chatbots: a case study of Replika. In: Proceedings of the 55th Hawaii international conference on system sciences, Maui Yan Z, Wang T, Chen Y, Zhang H (2016) Knowledge sharing in online health communities: a social exchange theory perspective. Inf Manag 53(5):643–653. https://doi.org/10.1016/j.im.2016.02. 001 Yao R, Qi G, Wu Z, Sun H, Sheng D (2024) Digital human calls you dear: how do customers respond to virtual streamers’ socialoriented language in e-commerce livestreaming? A stereotyping perspective. J Retail Consum Serv 79:103872. https://doi.org/10. 1016/j.jretconser.2024.103872 Zhang J, Wang X, Lu J, Liu L, Feng Y (2024) The impact of emotional expression by artificial intelligence recommendation chatbots on perceived humanness and social interactivity. Decis Support Syst 187:114347. https://doi.org/10.1016/j.dss.2024. 114347 Zhou Y, Fei Z, He Y, Yang Z (2022) How human–chatbot interaction impairs charitable giving: the role of moral judgment. J Bus Ethics 178(3):849–865. https://doi.org/10.1007/s10551-02205045-w Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. 123 L. Bayor et al.: Social-Oriented Communication with AI Companions: Benefits..., Bus Inf Syst Eng 67(5):637–655 (2025) 655