Understanding GenAI Teammates in the Workplace: A Sensemaking and Sensegiving Analysis of User Reviews

The increasing integration of GenAI into organizational settings is transforming how work is performed. Unlike earlier forms of automation or general-purpose AI models, GenAI models, particularly large language models (LLMs) like ChatGPT, introduce qualitatively different capabilities. These models are built on deep learning architectures such as the Transformer and trained on vast amounts of textual data using self-supervised learning (Goodfellow et al., 2014; LeCun et al., 2015). It enables them to learn linguistic patterns, semantics and contextual relationships. ChatGPT, based on OpenAI’s GPT architecture, utilizes billions of parameters to generate coherent and contextually appropriate responses to user prompts. It operates through token-based prediction, where each word or subword token is generated sequentially by estimating the probability distribution over the vocabulary, conditioned on previous tokens (Feuerriegel et al., 2024; Goodfellow et al., 2014). Modern LLMs are fine-tuned using techniques such as reinforcement learning from human feedback to align their outputs with human preferences and ethical guidelines (Ouyang et al., 2022). This general-purpose design enables ChatGPT to perform diverse language tasks such as summarization, translation, question answering and creative writing. These models often do not require task-specific training. Their effectiveness comes from their ability to generalize knowledge learned during training. GenAI tools generate human-like content, handle complex queries and engage in adaptive communication. These features position GenAI not only as a support tool but as a potential collaborator or “AI teammate” in the workplace (Flathmann et al., 2023; Siemon, 2022; Ulfert et al., 2024). Prior research related to human-AI interaction highlights foundational factors such as task-technology fit, system quality, and user trust for effective collaboration at the workplace (Goodhue & Thompson, 1995). These studies underscore the importance of clarity of purpose, adequate user skills, and contextual alignment in driving adoption (Figueroa-Armijos et al., 2023; Gkinko & Elbanna, 2023; Hasija & Esper, 2022). Emerging GenAI-specific research emphasizes its potential to enhance innovation, productivity, and job autonomy (Brynjolfsson et al., 2025; Kong et al., 2024), while simultaneously raising concerns about hallucinations, algorithmic bias, accountability, transparency, and intellectual property (Anthony et al., 2023; Ooi et al., 2023). Moreover, GenAI introduces unique socio-cognitive demands by increasing the need for new skills, creating job crafting opportunities, and sometimes contributing to cognitive overload (Hai et al., 2025; Ritz et al., 2025). Psychological barriers such as algorithmic aversion also persist, especially among experts (Wang et al., 2024). From an organizational perspective, GenAI may function either as an “algorithmic cage” that constrains autonomy (Bankins et al., 2024) or as an “algorithmic colleague” that enhances decision-making, depending on the surrounding culture. In team settings, GenAI systems are increasingly framed as potential collaborators, though their acceptance depends on being perceived as competent yet limited in scope, rather than omniscient authorities (Maedche et al., 2019). Prior research in crisis-mapping demonstrates that volunteers construct plausible interpretations from sparse cues (Valecha et al., 2025). Similarly, employees in contemporary workplaces now engage in joint human-AI sensemaking, where LLM outputs meaningfully influence how tasks, problems, and decisions are framed. This interpretive interdependence amplifies both the opportunities and the risks associated with GenAI adoption.

The use of GenAI in knowledge work also raises persistent challenges around data privacy and organizational governance. While these systems can analyze and synthesize structured and unstructured data (e.g., documents, emails, video transcripts), they may expose sensitive knowledge to external parties via feedback loops (Benbya et al., 2024). This has led firms such as Samsung and JP Morgan to restrict GenAI use internally (Mok Aaron, 2023; Ray Siladitya, 2023). At the individual level, adoption patterns vary by gender, expertise, and professional context (Bankins et al., 2024; Kolbjørnsrud, 2024). For example, in healthcare, GenAI is seen as having the potential to support professionals by improving diagnostic accuracy and enabling self-assessment through mobile technologies (Kolbjørnsrud, 2024). However, even when GenAI is perceived to improve performance, its adoption depends on whether users believe the technology is easy to understand and operate (Chatterjee et al., 2021). Academic literature highlights that the successful use of GenAI in the workplace depends on more than its technical capabilities. Human factors, such as trust, cognitive load, organizational culture, and perceived usefulness, play a critical role in shaping outcomes (Cheng et al., 2017). Theoretical perspectives such as hybrid intelligence and human-AI symbiosis suggest that combining human intuition and emotional intelligence with AI’s computational power can produce superior outcomes (Bartelheimer et al., 2025; Siemon, 2022). In conclusion, we present a summary of recent research, along with the methodologies and theoretical frameworks used in academic literature, to offer insights into the evolving future of work enabled by technology (see Table A1 in the Appendix).

The concepts of sensemaking and sensegiving offer a valuable lens for understanding how employees interpret and adapt to the introduction of the latest technology and tools like GenAI in the workplace. Sensemaking, as introduced by Weick (1995), refers to the process through which individuals construct meaning when facing novel, uncertain or complex situations. It involves interpreting cues, forming mental models, and acting based on those interpretations. This process occurs not only at the individual level but also at group and organizational levels (Weick, 2005). In organizations, sensemaking helps employees understand how new technologies, like GenAI, fit into their roles and workflows. Sensegiving, on the other hand, is the intentional effort by individuals, leaders, managers, or system designers to influence how others understand a situation or technology (Gioia & Chittipeddi, 1991). It involves sharing visions, framing new initiatives and guiding others toward desired interpretations. In the context of GenAI adoption, sensegiving plays a key role in shaping how employees perceive the usefulness, reliability and purpose of AI teammates. Together, sensemaking and sensegiving form a dynamic, cycle through activities such as envisioning, signaling, revisioning, and energizing (Gioia & Chittipeddi, 1991). For example, during technology implementation, users continuously revise their understanding based on new experiences and leadership narratives (Vaara et al., 2016). GenAI systems are inherently equivocal; their outputs can be interpreted in multiple ways, making sensemaking essential (Weick, 1995). Technological design features, such as categorization mechanisms, structure collective sensemaking by improving information quality (Valecha et al., 2025). Applying this framework allows us to explain how users continuously adjust their expectations of GenAI tools like ChatGPT, while managers and designers shape these interpretations through training, narratives, and system affordances.

This aligns with information systems research showing that technology sensemaking helps users develop beliefs and practices around new tools (Mesgari & Okoli, 2019; Urquhart et al., 2024). Managers support this through sensegiving practices such as training, storytelling and vision-setting (Tong et al., 2015). Applying the sensemaking-sensegiving framework enables a deeper understanding of how individuals and organizations co-construct meaning during GenAI adoption. This perspective highlights that human-AI collaboration is not only technical but also social and interpretive, evolving through communication, experience, and reflection. Building on this, we identified several stages and factors shaping sensemaking and sensegiving in user engagement with GenAI. While sensemaking and sensegiving offer a strong overarching framework, the stages often manifest through specific psychological and technological constructs. To capture these nuances, we draw on established theories such as Task-Technology Fit (TTF), IT Mindfulness, Social Influence Theory (SIT), Adaptive Structuration Theory (AST), and IS Continuance Theory. These theories provide validated constructs that link abstract sensemaking and sensegiving dynamics to measurable factors. For example, Task-Technology Fit and Perceived Usefulness help to operationalize the cognitive evaluation of how well GenAI aligns with individual goals, while constructs such as Trust and AI Mindfulness illuminate the emotional and behavioral adaptations that shape subsequent sensegiving. Table 1 summarizes these constructs and definitions, adapted from the theoretical foundations.

Integrating these constructs ensures theoretical rigor, enables finer-grained explanations of user engagement, and fits the sensemaking-sensegiving framework within the broader IS literature. In conclusion, we draw on prior studies to demonstrate the diverse applications of the theory in explaining technology use across contexts. Additionally, we show how the theory has been applied in this study to generate insights into how GenAI technologies are shaping the future of work (see Table 2).

This study has extracted public reviews of GenAI-based teammate applications, ChatGPT, from the Google Play Store. Among the range of GenAI tools increasingly integrated into workplace settings, including Microsoft Copilot, Google Gemini, Grammarly, and GitHub Copilot, this study focuses on ChatGPT due to its general-purpose design, high accessibility and widespread adoption. ChatGPT was the most downloaded app worldwide in April 2025, with 52 million downloads (Backlinko, 2024). As a large language model, ChatGPT enables users to generate human-like text, code, images, and reasoning-based outputs, and can even suggest follow-up prompts in real time. It operates across multiple platforms: mobile, desktop, and via API and is being progressively embedded into enterprise software systems. These features make ChatGPT a highly versatile digital assistant, applicable across various roles and industries. Also, the mobile version of ChatGPT has been downloaded millions of times globally, providing a rich and diverse dataset for exploring real-world engagement with GenAI in professional contexts. The Google Play Store was selected over the Apple App Store due to the significantly larger user base of Android devices. As of 2024, Android commands a global market share of 72% with over 3 billion active users, while iPhone holds 28% market share with a total of 1.382 billion consumers (Backlinko, 2024; Webb Maria & Pankratyeva Alexandra, 2024). This selection allows for the incorporation of a more diverse range of perspectives. Reviews represent a potential source of data, capturing public perceptions regarding the nuances and artifacts of the ecosystem. The “Google_Play_Scraper”, a Python package, was utilized to extract reviews from the ChatGPT Android applications. A total of 887,567 reviews were collected for these applications up to May 31, 2025.

Full size image

Following the extraction of user reviews, we implemented NLP-based techniques to refine the reviews and establish the theoretical lexicons (Kar & Dwivedi, 2020). Subsequent to the elimination of duplicate reviews and those lacking textual content, our dataset comprised 443,338 reviews. We conducted a comprehensive text pre-processing process, which involved removing special characters, numbers, regular expressions and standard stop words. Generic stop word lists commonly used in studies often fail to account for domain-specific language (Alshanik et al., 2020). Techniques used in prior studies to define corpus-specific stopwords, such as author-based keywords from a database or frequency pruning (Baradad & Mugabushaka, 2015), are less suitable for user-generated reviews. Scopus terms rarely appear in informal user texts, while frequency thresholds risk removing semantically meaningful words. Hence, we created a custom stop word list including corpus-specific terms such as “chatgpt,”, ‘chat_gpt’, “app,” “application,” “AI,” “OpenAI,”, “android”, “tool”, “version”, “software”, “chatbot”, “update”, ”iOS”, “mobile” and similar frequently occurring but uninformative words. Removing these terms helped improve the quality of the analysis by reducing noise and focusing on contextually meaningful content. Subsequently, the text was tokenized into individual words and we applied Named Entity Recognition (NER) and part-of-speech filtering were then applied using the spaCy library to retain meaningful entities and linguistically significant terms before lemmatization. The tokens were then lemmatized, and bigram and trigram models from Gensim were employed to finalize the lexicons for analysis. Effective text preprocessing is critical in text mining, as it reduces dimensionality and enhances computational efficiency.

Textual data can be summarized using a variety of analytical methods. Prior studies have utilized principal component analysis for such summarization (Yadav et al., 2023). Topic modeling is widely recognized as a suitable method (Kar & Dwivedi, 2020; Miranda et al., 2022). Additionally, clustering techniques (see Fig. 3) can be employed to complement topic modeling by verifying and enriching the interpretation of the derived topics (Eryilmaz et al., 2022).

Following tokenization and lemmatization, we applied a clustering approach that organizes lexicons based on semantic similarity. Specifically, we used a combination of K-means and hierarchical clustering. K-means helped determine the optimal number of clusters (refer to Fig. 2), which was then used to guide hierarchical clustering (Fig. 3). The lexicons were progressively grouped into broader clusters, supporting the development of preliminary theoretical insights based on lexical similarity.

We employed Latent Dirichlet Allocation (LDA) to extract theoretical lexicons by identifying patterns in word co-occurrence and probabilistic distributions across the corpus (Blei et al., 2003). In recent years, advanced approaches such as Non-negative Matrix Factorization (NMF) (Vangara et al., 2021), BERTopic (Xu et al., 2022), and embedding-based models using Doc2Vec or transformer architectures have gained popularity (Budiarto et al., 2021). While these methods produce semantically rich outputs, they often lack transparency, are difficult to replicate, and pose challenges for theory-driven interpretation which are an essential focus in Information Systems research.

In contrast, LDA is one of the most widely used probabilistic topic models, offering interpretable topic-word distributions that align closely with theory-driven inquiry (Hannigan et al., 2019). This approach enabled us to identify underlying latent themes within the textual data. To determine the optimal number of topics, we tested values ranging from 1 to 40 and selected four topics based on the highest observed coherence score of 0.5397. To enhance rigor, we complemented LDA with k-means and hierarchical clustering, ensuring a balance between robustness, transparency and interpretability. Finally, we conducted a detailed analysis of each topic to identify the primary themes and derive relevant subtopics.

Integrating clustering with topic modeling offers a cohesive analytical strategy to summarize large volumes of text and identify theoretically relevant lexicons. The preliminary groupings generated through similarity-based clustering support the interpretation of latent themes extracted via topic modeling. However, the thematic outputs require systematic alignment with the conceptual constructs outlined in Table 2. To ensure theoretical rigor, we assessed the semantic correspondence between topic keywords and the items within established constructs (Basu et al., 2024). Prominent keywords associated with each factor are illustrated in Fig. 4 and Figure A4 in appendix through word clouds, while the complete mapping of topics to theoretical constructs is detailed in Appendix Table A2 and Table A3. The wordcloud library in Python is used to visualize textual data by generating frequency-based graphical representations, where word size reflects its relative importance in the corpus.

The hierarchical clustering output is visualized in Fig. 3, where lexicons are grouped based on semantic proximity. For instance, terms such as “highly_versatile,” “timely_response,” and “save_time_efforts” appear within the same cluster. However, clustering alone cannot identify hidden thematic structures or estimate the topic prevalence across individual reviews. To address this, we utilized an LDA-based topic modeling technique. Figure 4 presents key themes and associated keywords as identified by this model. For example, a cluster featuring terms like “accurate,” “outstanding,” and “exceptional” reflects users’ perceptions of GenAI’s reliability and trustworthiness. This process was repeated to identify all relevant factors.Drawing from the topic modeling outcomes and guided by Sensemaking and Sensegiving theories, the study introduces the Envision-Evolve-Engage (3E) framework as shown in Fig. 5. This model conceptualizes user engagement with GenAI applications (e.g., ChatGPT) as a dynamic process structured around two mechanisms:

Based on the study’s findings, we map the four activities of sensemaking and sensegiving: envisioning, signaling, revisioning, and energizing onto the Envision-Evolve-Engage (3E) framework. This integration connects micro-level user actions with the broader process of technology adoption. Envisioning corresponds to the envision stage, where users articulate potential applications of ChatGPT (e.g., drafting emails, brainstorming ideas). Within the Evolve stage, signaling occurs when users share experiences that shape others’ expectations, while revisioning emerges in reviews where users iteratively reinterpret the tool’s capabilities in light of successes or limitations. Finally, energizing aligns with the Engage stage, where user’s express enthusiasm, satisfaction, and motivation for continued use, often encouraging others to adopt the tool. A detailed explanation of each construct is provided below.

This study offers several theoretical and methodological contributions to the academic literature. First, the study abductively derives the 3E framework (Envision-Evolve-Engage) from large-scale user data. It shows how sensemaking and sensegiving processes unfold in the context of GenAI as an ongoing cycle of meaning construction. Users not only interpret the capabilities and limitations of GenAI for personal cognitive framing but also participate in sensegiving activities such as sharing usage experiences, recommending strategies and signaling credibility. These activities shape the perceptions and adoption behaviors of their peers. In doing so, the study advances framing theory (Davidson, 2002) and signaling theory (Connelly et al., 2011) by positioning users as both interpreters and communicators of AI-related signals, thereby influencing how technological affordances are socially constructed and understood in organizational contexts. To further highlight the novelty of the 3E framework, we provide a comparative table (see Appendix Table A6) with traditional technology adoption models (e.g., The Unified Theory of Acceptance and Use of Technology (UTAUT), TTF, Information system success model (ISS). Whereas prior models conceptualize adoption largely as a linear process of cognitive evaluation (e.g., perceived usefulness, ease of use, behavioral intention), the 3E framework foregrounds recursive cycles of envisioning, evolving and engaging. This positions adoption as a dynamic socio-cognitive process that integrates affective, ethical and communal dimensions alongside individual cognition.

Second, the study contributes to the literature on Human-AI Interaction by exploring the social and emotional dynamics that characterize users’ engagement with GenAI tools. While prior research has largely focused on task efficiency and performance outcomes, our findings reveal that user interactions with GenAI are deeply shaped by emotional resonance, peer influence and evolving community norms. Drawing on AST, we show that GenAI tools are not simply technical artifacts but become socially embedded systems, adopted through processes of interpretation, experimentation and collaborative learning. Users develop affective commitment to these tools, using them not only for task completion but also for affirmation, uncertainty reduction and co-construction of new norms of technology use. These insights broaden existing understandings of human-AI collaboration by offering a holistic view of how human actors adapt, co-evolve and assign meaning to AI-enabled systems over time. Moreover, the study advances the future of work discourse by illustrating how GenAI adoption fosters unlearning, relearning and the cultivation of new skillsets (Hönigsberg et al., 2025). Factors such as AI mindfulness, trust, technological affordances and task-technology fit emerge as critical to sustained GenAI use, highlighting the need for adaptive human capabilities and socio-cognitive restructuring in AI-mediated work environments. AI mindfulness is more context-sensitive. It involves deliberate, ethical, and reflective engagement with GenAI, where users assess responses for accuracy, contextual appropriateness, and social implications before acting upon them. This distinction highlights the unique role of ethical awareness and adaptive evaluation in AI-mediated work.

Finally, this study contributes methodologically by illustrating how computational techniques can facilitate inductive theory development. Leveraging large-scale user reviews as a form of electronic word-of-mouth (eWoM), we applied NLP methods to identify salient lexical patterns related to GenAI use. These patterns were clustered using topic modeling and other clustering approaches to explore emergent constructs and relational themes. This approach responds to recent calls for computationally grounded theorizing (Berente et al., 2019, 2025) and demonstrates how text analytics can yield socially embedded insights that reflect the lived experiences of users. Through this method, the study bridges qualitative richness with computational rigor, offering a scalable pathway for theorizing from digitally native data.

The findings of this study provide several actionable insights for organizational leaders, system designers and policy-makers engaged in the integration of GenAI in the workplace. First, organizations should recognize employees as co-creators in shaping GenAI use and outcomes. Beyond technical fluency, targeted training that develops interpretive and sensemaking skills can enable employees to engage more meaningfully with GenAI systems. Embedding continuous feedback loops, such as employee evaluations of AI-generated outputs (e.g., draft performance reviews) can foster legitimacy, ownership, and trust. Designing for both cognitive and emotional user responses can further reduce resistance and identity threats, thereby supporting sustained engagement. Moreover, organizational learning mechanisms, including AI literacy programs, peer mentoring initiatives, and reflective training workshops, can facilitate the integration of the proposed 3E framework into everyday work practices. Such initiatives enable employees to internalize the interpretive, ethical, and collaborative dimensions of GenAI use, allowing organizations to institutionalize adaptive learning and responsible innovation. This participatory approach aligns with emerging calls for human-centric and relational AI governance frameworks (Janssen, 2025). Second, organizations should reframe the role of GenAI from a tool of automation to one of augmentation. Positioning GenAI as a collaborative partner strengthens employee creativity, decision-making and higher-order problem solving. For instance, a marketing team may use GenAI to generate campaign drafts, while managers guide critical evaluation of AI suggestions and their integration into final strategies. System designers can reinforce this by embedding affordances that promote transparency and user agency. Features such as “explanation prompts” that clarify how outputs are produced and annotation tools that allow employees to critique responses can enhance sensemaking and align system evolution with organizational needs. Such framing fosters psychological empowerment and supports role redefinition, reducing resistance often associated with technological change. Third, responsible GenAI deployment requires inclusive governance strategies that address ethical risks and access inequalities. Policy frameworks should prioritize equitable onboarding, scaffolded learning, and clear guidelines around data use, bias and algorithmic transparency. For instance, organizations rolling out enterprise-level GenAI systems could design tiered access and structured onboarding tailored to different professional roles. This should ensure that frontline workers and knowledge professionals benefits equitably from AI augmentation. Regular explainability audits and incident reporting systems for problematic or biased outputs further ensure accountability and prevent digital marginalization. As GenAI tools like ChatGPT increasingly shape knowledge work, proactive regulation is essential to ensure the safeguarding of fairness, inclusivity, and organizational integrity. Collectively, these implications emphasize the importance of aligning technological affordances with social, ethical and cognitive dimensions of employee engagement to harness the transformative potential of GenAI in the future of work.

This study contributes to advancing SDG 8, which promotes inclusive economic growth, productive employment and decent work for all (United Nations, 2015). GenAI tools such as ChatGPT can enhance workplace accessibility by offering real-time task assistance, knowledge support and productivity enhancement. These capabilities help users navigate complex tasks more efficiently and foster broader participation among individuals with varying levels of digital fluency. The time saved through GenAI-enabled efficiencies can be reallocated to high-value tasks or personal pursuits, thereby supporting improved work-life integration and employee well-being. Through the Envision-Evolve-Engage (3E) framework, this research highlights GenAI’s role in enabling continuous learning, adaptive decision-making, and skill enhancement, aligning closely with SDG 8’s emphasis on lifelong upskilling and sustainable employment. The proposed framework supports “decent work” by encouraging organizations to “envision” inclusive roles for GenAI, “evolve” job designs that balance augmentation with human agency, and “engage” employees in participatory learning pathways. GenAI-generated task recommendations can be assessed and prioritized by team members, reinforcing decision-making capacity and equitable participation. Such practices help ensure that technology adoption strengthens fairness, inclusivity, and long-term employability. However, to ensure that these benefits are equitably distributed, ethical oversight and inclusive implementation practices are essential. Without such safeguards, GenAI may inadvertently reinforce digital divides or foster over-reliance on algorithmic outputs. Organizations and policymakers must therefore prioritize participatory governance, responsible AI design and equitable access to GenAI tools. Importantly, this study highlights that GenAI not only enhances task performance but also supports psychological well-being by reducing cognitive load, increasing perceived autonomy and reinforcing users’ sense of control over work processes. When integrated thoughtfully, GenAI functions as a collaborative augmentation tool that enhances human capability rather than replacing it. While concerns over job displacement persist, our findings suggest that GenAI can enable more meaningful work as an AI Teammate by reshaping job roles, fostering creativity and reinforcing the need for ongoing reskilling. The proposed 3E framework offers a systematic foundation for national and institutional AI policy formulation by integrating principles of ethical accountability, participatory governance, and sustained human development within GenAI adoption processes. This shift toward augmentation-oriented adoption models and forward-looking policy design can help ensure that technological advancement translates into equitable growth, resilient labor markets, and the realization of “decent work for all” as envisioned under SDG 8.

This study provides valuable insights related to the use of GenAI in the workplace by analyzing user reviews of ChatGPT. However, several limitations must be acknowledged. First, the data were collected solely from the Google Play Store. Although this source offers authentic user experiences, it may not capture the full diversity of user interactions. Future research should consider including data from other platforms, such as the Apple App Store, as well as additional sources like social media and grey literature to improve data diversity and reliability. Second, the study focuses on a single GenAI application, namely ChatGPT. While ChatGPT is widely adopted and reflects broader trends in GenAI usage, its application spans personal, educational, and professional contexts. Such diversity provides valuable insights into generalized user behaviors, but some reviews may not represent workplace use. To ensure the relevance of the dataset for workplace analysis, reviews were categorized using keywords related to work, productivity, task completion, and professional collaboration. Subsequently, a subset of reviews was manually examined to validate the dataset. Findings showed that the majority of reviews, even if general in wording, provided insights applicable to workplace contexts. In addition, reliance on a single tool may limit the generalizability of the findings. Future research could employ stratified sampling to distinguish workplace-specific reviews from broader usage contexts and extend the proposed model across other GenAI platforms, including enterprise-level and specialized applications. Comparative research involving tools such as Samsung SDS FabriX and Amazon Q could reveal how organizational scale, customization, and data governance influence GenAI adoption and effectiveness. Third, the study is based on self-reported public reviews. Reliance on user-generated reviews may introduce biases, as such reviews often reflect extreme experiences - either highly positive or highly negative and may also be influenced by recall bias or subjective interpretation. While these insights highlight critical affordances and barriers, they risk skewing interpretations. Future research could strengthen findings by employing longitudinal designs and objective behavioral data (e.g., system usage logs, multimodal interaction patterns) to track how human-AI interactions evolve. Fourth, cultural and occupational differences in GenAI use remain underexplored. Users in different professional roles, such as educators, frontline workers or creative professionals, may engage with GenAI tools in diverse ways. Studying these variations would help develop more inclusive and generalizable insights. Finally, the study highlights the need for further research into the long-term social and ethical implications of GenAI integration. Issues such as overtrust, emotional reliance and the potential erosion of human skills require critical attention. Future information systems research should examine these risks through socio-ethical frameworks and complement exploratory insights with confirmatory quantitative methods to support responsible GenAI adoption.