MSMarita Skjuve

Marita Skjuve

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Marita Skjuve has a total of 28 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Who Uses Conversational AI and Why? Distinct User Profiles and Their Links to Personality, Attachment, and Well-Being

    by Nino Gugushvili, Petter Bae Brandtzaeg, Marita Bjaaland Skjuve

    1. Since the release of ChatGPT-3.5 in 2022, the use of commercial Conversational AI (CAI) agents such as ChatGPT, Claude, and Gemini has surged, with new models and regular updates. Despite this r...

  • Open Access

    Who Uses Conversational AI and Why? Uncovering Distinct User Profiles, Their Antecedents and Mental Health Outcomes

    by Nino Gugushvili, Petter Bae Brandtzæg, Marita Skjuve

    With the increasing use of conversational AI (CAI) agents, such as ChatGPT, public concern about their social implications has grown. Beyond instrumental reasons (e.g., work-related), many people use these agents for personal purposes, such as social support and entertainment. Yet, no study has examined the motivational profiles underlying CAI use, their personality and attachment antecedents, and their consequences for mental health outcomes. To address this gap, we conducted a cross-sectional study with 782 CAI users (Mage = 43.47, SD = 13.59). Drawing on Uses and Gratifications Theory, we measured key motivations underlying CAI use, including information seeking, task efficacy, personalization, social interaction, and playfulness, and subjected them to latent profile analysis to uncover distinct motivational profiles. To identify the antecedents of these profiles, we included Big Five personality traits (Extraversion, Neuroticism, Openness, Conscientiousness, Agreeableness) and attachment styles (Anxious and Avoidant) as predictors. Finally, we measured emotional, social, and psychological well-being as key outcome variables of the identified profiles. The analyses identified three distinct profiles of CAI users: Occasional Information Seekers, Utilitarians, and All-Round Engagers, most strongly differentiated by their levels of social interaction and playfulness motivations. Individual differences in attachment anxiety and extraversion significantly predicted the likelihood of belonging to a specific profile. Finally, profile membership was predictive of social well-being, with All-Round Engagers reporting higher levels of social well-being than Utilitarians. These findings advance our understanding of who engages with CAI and how, with implications for mental health outcomes.


  • Generative AI in the Workplace: Workers’ Experiences and Competence Development

    by Anna Grøndahl Larsen, Marita Bjaaland Skjuve, Asbjørn Følstad

  • Open Access

    Beyond Utility: Relational Conflicts in Human–AI Relationships

    by Marita Skjuve, Anna Grøndal Larsen, Asbjøn Følstad, Petter Bae Brandtzaeg

  • Open Access

    AI aversion? Effects of author disclosure on young people’s perceptions of mental health advice

    by Petter Bae Brandtzaeg, Marita Skjuve, Asbjørn Følstad

    The increasing use of large language models (LLMs), such as ChatGPT, is already impacting how young people seek mental health support online. However, AI aversion, the reluctance or resistance individuals feel toward AI, may influence individuals’ perceptions and willingness to engage with LLM-generated advice. In this mixed-method study, we investigated how 440 young people (aged 17–21) perceived mental health advice from ChatGPT compared with that of health professionals, emphasizing the effect of author disclosure. Participants assessed answers from ChatGPT and health professionals across four dimensions—Validation, Relevance, Clarity, and Utility—and were asked to recommend answers. The findings indicate a preference for AI-generated answers when participants were unaware of the author’s identity: ChatGPT’s answers scored significantly higher on Validation, Relevance, Clarity, and Utility. Conversely, when the author was disclosed, participants favored responses from health professionals and rated their answers significantly higher for Validation, indicating AI aversion. Qualitative data further revealed that participants became more critical when they knew the content was AI-generated, while responses from health professionals were viewed as more credible, empathetic, and tailored. These findings may indicate human favoritism. The study makes the key contribution of identifying how source awareness impacts the reception of AI-generated content in a sensitive domain. To address the potential for AI aversion within help-seeking, our findings suggest the importance of developing hybrid human–AI support models that combine the efficiency of AI with the relational legitimacy of human professionals, improving both the acceptance and impact of digital mental health support.


  • Open Access

    AI Individualism: What Are Social Relationships in the Age of Artificial Intelligence?

    by Petter Bae Brandtzaeg, Marita Skjuve, Asbjørn Følstad

    Seeking a hybrid social design approach that blends human and AI relationships in socially responsible ways.

  • Open Access

    Conversational AI and Transparency Statements: Design Options and Impact on User Perceptions

    by Marita Skjuve, Asbjørn Følstad, Knut Kvale, Anna Grøndal Larsen, Nena van As

    Conversational AI powered by large language models is increasingly seen as an attractive technology for customer service due to its ability to provide fluent, efficient support. However, the technology's humanlike qualities can make it hard for users to realize they're interacting with an AI rather than a human employee, and the unpredictability of LLMs increases the risk of misleading or inaccurate outputs. This creates a conversational AI transparency design challenge: How can we disclose conversational AI use and communicate its limitations in ways that users notice and understand? To address this, we conducted two complementary studies examining user perceptions and the effects of transparency statements (disclosures and disclaimers) in customer service contexts. In Study 1 (N = 300), participants evaluated multiple transparency design options presented as chat screenshots, varying statement wording and placement (e.g., top/bottom of the interface vs. message-level labels with and without tooltips). The results indicate that participants' preferences were driven primarily by visual prominence and low interaction cost; clear wording at the top or bottom of the interface was often seen as beneficial, tooltipbased designs were seen as easy to miss, and selective message-level labeling risked confusing users about which messages were AI-generated. In Study 2 (N = 208), participants interacted with a GPT4o-powered conversational AI prototype under three conditions (no disclosure, disclosure only, disclosure and disclaimer). The results show that many participants did not notice or retain the transparency statements, and we found no significant differences across conditions in regard to trust, willingness to use, or likelihood of following advice. Together, the results suggest that while transparency statements are necessary to comply with transparency requirements and have been thought to be beneficial for trust calibration, designing transparency statements that users notice and remember remains an important design challenge.

  • Open Access

    Predefined Rule-Based NLU Vs LLM Generated Responses in Conversational AI: A Study of Customer Service Dialogs in the Wild

    by Silje Granås, Anna Grøndal Larsen, Asbjørn Følstad, Gunn Marit Kinden, Marita Skjuve, Nena van As

    Conversational AI is increasingly used in customer service, yet little is known about how traditional systems based on predefined responses compare with conversational AI using large language models (LLM) in real-world use. This study analyses 575 customer-conversational AI dialogues from a Norwegian insurance provider to investigate how predefined and LLM-generated responses differ in performance across relevance, helpfulness, efficiency, and understandability. Users could choose between the two systems during live service interactions, enabling a unique comparison of their performance in the wild. Drawing on the framework of Følstad and Taylor (2021), we conducted a qualitative, two-level analysis of message sequences and full dialogues. The results demonstrate that the LLM-based system produced a substantially higher proportion of relevant responses, fewer false positives and false negatives, and more dialogues that likely led to help being used. Conversational AI with predefined responses, in contrast, showed higher rates of irrelevant answers and more frequent escalations, many of which were policy-driven but also triggered by misunderstanding. Few incidents of hallucinations and other LLM-related mistakes were detected. Overall, our findings indicate that LLM-based conversational AI can improve response quality and reduce conversational friction in customer service. The study contributes with the first in-the-wild comparison of these two conversational AI approaches within customer service, offering implications for system design, automation strategies, and future research on LLM-augmented customer support.

  • Open Access

    ChatGPT as a mental health advisory service: Comparing evaluations from youth and health professionals

    by Marita Skjuve, Asbjørn Følstad, Petter Bae Brandtzaeg

    Objective: Despite the growing use of artificial intelligence in youth mental health support, little is known about how either young people or health professionals perceive answers to mental health-related inquiries generated by large language models (LLMs). Therefore, we draw on Media Richness Theory to examine how these two user groups perceive the “richness” of text-based communication in this context and whether young people and health professionals differ in their assessment. Methods: A total of 123 young people and 31 health professionals evaluated answers to youth mental health inquiries. Each inquiry had two blinded answers: one generated by ChatGPT (GPT-4) and one written by a health professional. Participants rated the answers for validation, relevance, clarity, and utility and were asked to recommend one or both answers. Open-ended responses elaborating participant choices were also collected. Results: The quantitative findings show that young people and health professionals rated answers from both sources similarly on validation, clarity, and utility. However, young people rated ChatGPT's answers higher for relevance and utility, finding them “richer.” This was supported by qualitative data, where youth preferred ChatGPT's clear and actionable answers. Health professionals showed no strong preference and were more critical, often finding the answers too detailed or lacking empathy. Conclusion: This study is the first to compare youth and professional perspectives on ChatGPT's role in youth mental health advice within a blind evaluation design. We conclude by proposing a hybrid advisory model that combines professional expertise with LLMs to enhance the efficiency, scale, and accessibility of youth mental health advisory services.


  • Open Access

    Large language models in daily life: Users’ perceptions of beneficial assistance and undue reliance

    by Maria Nylund, Marita Skjuve, Anna Grøndahl Larsen, Asbjørn Følstad, Silje Bechmann Granås, Ophelia Prillard

    Large language models (LLMs), such as ChatGPT and Gemini, have rapidly become an integral part of daily life, driven by their advanced conversational capabilities and versatility. Despite widespread adoption, understanding of users’ perceptions regarding the balance between beneficial assistance and undue reliance on LLMs remains limited. To close this gap, we conducted a qualitative questionnaire study (N = 223) to investigate how users perceive the impact of LLMs on their daily lives. Results indicate that users find beneficial support through significantly enhanced productivity, creativity, learning, cognitive offloading, and quality of work. However, problematic reliance also emerges, leading to self-doubt, concerns on reduced efficiency, diminished skill confidence, guilt, and negative self-perceptions. Nevertheless, many users maintain a balanced perspective, viewing LLMs primarily as helpful tools rather than necessities and employing deliberate usage strategies to retain autonomy and competence. The findings are discussed through the lens of self-determination theory and contribute important insights into how LLM interactions may fulfill or compromise fundamental psychological needs of autonomy, competence, and relatedness. The findings offer implications for designing technologies that optimize user well-being while mitigating negative effects.