Purpose Human–agent interaction (HAI) is increasingly influencing our personal and work lives through the proliferation of conversational agents (CAs) in various domains. As such, these agents combine intuitive natural language interactions by also delivering personalization through artificial intelligence capabilities. However, research on CAs as well as practical failures indicates that CA interaction oftentimes fails miserably. To reduce these failures, this paper introduces the concept of building common ground for more successful HAIs. Design/methodology/approach Based on a systematic literature analysis, we identified 38 articles meeting the eligibility criteria. We critically reviewed this body of knowledge within a formal narrative synthesis structured around the use of common ground in the interaction with CAs. Findings Based on the systematic review, our analysis reveals five mechanisms for achieving common ground: embodiment, social features, joint action, knowledge base and mental model of conversational agent. We point out the relationships between these mechanisms as they are related to each other in directional and bidirectional ways. Research limitations/implications Our findings contribute to theory with several implications for CA research. First, we provide implications about the organization of common ground mechanisms for CAs. Second, we provide insights into the mechanisms and nomological network for achieving common ground when interacting with CAs. Third, we provide a broad research agenda for future CA research that centers around the important topic of common ground for HAI. Originality/value We offer novel insights into grounding mechanisms and highlight the potentials when considering common ground in different HAI processes. Consequently, we secure further understanding and deeper insights of possible mechanisms of common ground to shape future HAI processes.

As large language models (LLMs) increasingly serve as a technology basis for conversational health behaviour interventions, understanding how users evaluate and respond to these systems becomes critical for their acceptance. When LLM-based coaches are anthropomorphically designed - e.g. through human names, avatars, and messaging interfaces - they may trigger interpersonal trust expectations traditionally reserved for human relationships. This study investigates whether and to what extent the interpersonal trust antecedents of ability, benevolence, and integrity influence trust-related user responses in LLM-based health coaching for fitness and nutrition. Using a vignette-based survey design and combining partial least-squares structural equation modelling (PLS-SEM) with necessary condition analysis (NCA), we identify which antecedents are essential for perceived usefulness and intention to adopt. Results show that integrity is a necessary condition for both outcomes, with a medium to large effect, and that ability significantly influences perceived usefulness but is not necessary. Benevolence, however, was found to be neither a necessary nor a significant predictor in this context. These findings offer theoretical insights into the selective transfer of interpersonal trustworthiness beliefs to AI and provide practical guidance for the design of trustworthy anthropomorphic coaching systems.

Background Argument evaluation, including recognizing argument structures and fallacies, is crucial for academic success, yet many university students struggle with these abilities. Drawing on the motivational model of integrative processing and expectancy-value theory, this study examined how distal motivational factors (topic involvement, need to evaluate, academic self-concept in argumentation) relate to proximal appraisals (expectancy of success, task value, emotional cost) and, in turn, to argument evaluation. Method Data from 675 preservice teachers were analyzed using structural equation modeling to test these assumptions within a path model. Results The recognition of argument structures was positively related to recognizing fallacies. Emotional cost was negatively related to both abilities, whereas task value was positively related to the recognition of fallacies. Proximal motivational factors were systematically related to distal motivational factors such as topic involvement, need to evaluate, and academic self-concept. Relations involving topic involvement varied depending on the presumed prior topic knowledge. Conclusions The findings highlight the importance of fostering argument evaluation and underscore the role of recognizing argument structures for the recognition of fallacies. In addition, the findings indicate the need to reduce emotional cost in instructional settings and to focus on topics that combine high topic involvement with sufficient prior knowledge.

ABSTRACT Global crises and persistent uncertainty have exposed the vulnerability of supply chains. Procurement departments, traditionally focused on cost optimization, are increasingly required to act as strategic orchestrators of supply chains. This paper examines a data‐driven transformation within a European automotive procurement department, revealing how contextual frictions and organizational realities shape transformation outcomes. Based on a qualitative case study, our findings show that poor internal data quality, limited access to external supply chain data, multiple operational challenges from recurring disruptions, shortages or supplier insolvencies and insufficient institutionalization of data‐driven activities critically impede the value realization of data‐driven transformations. Building on these findings, we propose six actionable recommendations that address the identified barriers and position procurement as both a catalyst for effective data management and a strategic enabler of data‐driven supply chains. Key best practices include structured data negotiation with suppliers, embedded qualification programs and sustained leadership engagement to foster a data‐driven mindset. The study contributes to information systems research by advancing a domain‐specific understanding of data‐driven transformation and highlighting procurement as a strategic enabler of intelligent, data‐based decision‐making under conditions of crisis and uncertainty. Our findings provide actionable guidance for procurement practitioners and their data‐driven activities.

The infusion of generative AI (GenAI) is already disrupting established services. This technology's generative and agentic nature challenges the design and management of service routines, which have been previously handled primarily by frontline service employees. Guided by organizational routines theory, our longitudinal study (2020-2024) examines how the infusion of GenAI changes routines in customer support services. We gathered interview data from 41 employees, managers, and AI experts in two phases, pre- and post-GenAI. Based on the analysis of the qualitative data, we revealed seven recurring micro-level augmentation patterns, illustrating how GenAI-infused service routines function. The results show that GenAI is primarily embedded in the backstage of knowledge-intensive services, from which it then permeates the frontstage. We contribute to the literature on hybrid human-AI service delivery by identifying augmentation patterns and conceptualizing service permeation via two mechanisms: (1) simultaneous service permeation, which unfolds as employees leverage GenAI in real-time and integrate GenAI's responses, recommendations, and adaptations into the frontstage; (2) sequential service permeation, which emerges as employees perform new routines of documentation and AI feeding to facilitate GenAI's adaptability in frontstage and backstage operations. The MAPs and service permeation mechanisms guide practitioners in integrating GenAI into service routines and managing novel employee-GenAI collaborations.

AI agents are increasingly adopted in higher education, yet current systems handle requests uniformly, promoting cognitive offloading over sustained skill development. With the rise of orchestrated multi-agent systems, learning goals can be targeted by specialized AI agents that each address a distinct scaffolding mechanism: conceptual, procedural, strategic, or metacognitive. This study follows a Design Science Research approach to derive design requirements from 32 student interviews, iteratively refine a prototype with 22 IS experts, 6 educators, and 34 students, and computationally analyze scaffold effectiveness. We contribute three design principles: (1) multi-agent coordination through a lead orchestrator that delegates to mechanism-specific sub-agents, (2) adaptive learner profiling that enables cross-session scaffolding fading, and (3) continuous institutional knowledge integration grounding scaffolds in verified course materials. Our effectiveness analysis reveals that delivery order predicted learning behavior, positioning orchestration as a key design concern for AI-assisted educational systems.

Although conversational agents are successfully applied in teaching, it is largely unclear which communication principles should be employed to optimise learning. We examine the influence of common ground (i.e. shared knowledge on which to build during conversation) on learning. In an in-class experiment, students studied with one of two pedagogical conversational agents. The control version provided information without emphasising grounding, whereas the common ground version emphasised grounding, for example, by encouraging students to monitor and repair common ground. After the learning unit, students evaluated their learning experience and the pedagogical conversational agent, after which they were tested on the studied material. Students in the common ground (vs. the control) condition performed better in a post-study knowledge test and engaged longer with the pedagogical conversational agent. Thus, the common ground emphasis facilitated learning with a conversational agent, indicating that grounding principles should be incorporated when designing conversational agents.

Providing argumentation feedback is considered helpful for students preparing to work in collaborative environments, helping them with writing higher-quality argumentative texts. Domain-independent natural language processing (NLP) methods, such as generative models, can utilize learner errors and fallacies in argumentation learning to help students write better argumentative texts. To test this, we collect design requirements, and then design and implement two different versions of our system called ALure to improve the students' argumentation skills. We test how ALure helps students learn argumentation in a university lecture with 305 students and compare the learning gains of the two versions of ALure with a control group using video tutoring. We find and discuss the differences of learning gains in argument structure and fallacies in both groups after using ALure, as well as the control group. Our results shed light on the applicability of computer-supported systems using recent advances in NLP to help students in learning argumentation as a necessary skill for collaborative working settings.

In this study, we explore the potential of AI-generated podcasts as an educational tool in the evolving landscape of learning media. Podcasts have grown increasingly relevant in education due to their accessibility and ability to integrate learning into everyday life. With the advent of generative artificial intelligence (AI), there is a unique opportunity for scalable and adaptable creation of learning media. However, with novel technology, there also come new challenges. Thus, we developed fine-tuned AI-generated podcasts using Google NotebookLM, our course materials, and a custom prompt. We conducted a one-month explorative evaluation in the field using a qualitative diary study. Our study reveals that students find the podcasts beneficial for flexible everyday learning but also point toward challenges like a lack of emotional engagement and technical non-English language issues. In sum, our study highlights the current benefits and challenges of AI-generated podcasts and presents an agenda for future research.

This research applies the Design Science Research (DSR) methodology to investigate how self-referencing in Large Language Model (LLM)-based health coaching influences user trust and perceptions of anthropomorphism. We synthesized theory-driven design principles to guide the integration of self-referencing and demonstrated them in a vignette-based prototype. Through a single-factorial between-subjects experiment, analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM) and qualitative feedback, we identified a dual effect of self-referencing: while professional self-referencing enhances trust via increased anthropomorphism, overly personal references can directly undermine trust. Based on these findings, we refined our design principles to optimize trust-building in LLM-based coaching. Our contributions provide actionable design guidelines for creating more effective and trustworthy AI-driven health interventions, advancing the understanding of anthropomorphic design in digital coaching contexts.

This paper investigates the impact of large language model (LLM)-based coaching interventions on enhancing physical activity and nutritional habits, following motivational interviewing guidelines. By exploring user perceptions through qualitative research involving eight interviews, five key themes emerged: the tension between the need for authenticity and reservations about AI humanization, the desire for personalized coaching and autonomy, the necessity of simplifying daily tasks, the aspiration for self-development, and the need for perceived privacy and trust. The findings reveal that perceptions of LLM-based coaching are multifaceted and cannot be easily classified as purely beneficial or concerning; they vary based on the specific implementation. This complexity indicates that certain aspects can simultaneously present both benefits and risks. The paper discusses theoretical implications and offers practical recommendations to enhance the advantages and mitigate the risks associated with LLM-based coaching interventions.

Pedagogical conversational agents (PCAs) like chatbots are a novel approach to technologymediated language learning with artificial intelligence. They convey learning content interactively and accompany students in their education. However, many users find conversations with PCAs unmotivating. Gamification is a suitable solution to these motivational hurdles due to its playful nature. Given the difficulty of selecting the appropriate game elements and the scarcity of design recommendations for gamified PCAs, we propose the GNPL framework including a cohesive set of four design principles: goal-setting and reflection, novice-expert relationship, performance-related motivation, and learning story narration. In two design cycles, the article shows the application of the design principles in English learning - a domain commonly associated with motivational challenges - by implementing and evaluating a gamified PCA. The results show that the design principles significantly foster learners' motivation and that learners perceive a solid language learning experience, expressed by higher perceived value and social factors. They highlight the relevance of aligning the PCA's social role, the motivational impact of gamification, and the educational goals of the learning context. The design principles guide educators and developers in gamified PCA design. The paper contributes to the theory stream of PCAs by investigating learners' motivation enhancement when using PCAs. In addition, the paper provides new knowledge on meaningful gamification in an unexplored context and practical insights to solve the design challenges of selecting game elements in this context. Furthermore, it shows how language education can be supported by educational technology.

Artificial intelligence technologies are rapidly advancing. As part of this development, large language models (LLMs) are increasingly being used when humans interact with systems based on artificial intelligence (AI), posing both new opportunities and challenges. When interacting with LLM-based AI system in a goal-directed manner, prompt engineering has evolved as a skill of formulating precise and well-structured instructions to elicit desired responses or information from the LLM, optimizing the effectiveness of the interaction. However, research on the perspectives of non-experts using LLM-based AI systems through prompt engineering and on how AI literacy affects prompting behavior is lacking. This aspect is particularly important when considering the implications of LLMs in the context of higher education. In this present study, we address this issue, introduce a skill-based approach to prompt engineering, and explicitly consider the role of non-experts' AI literacy (students) in their prompt engineering skills. We also provide qualitative insights into students’ intuitive behaviors towards LLM-based AI systems. The results show that higher-quality prompt engineering skills predict the quality of LLM output, suggesting that prompt engineering is indeed a required skill for the goal-directed use of generative AI tools. In addition, the results show that certain aspects of AI literacy can play a role in higher quality prompt engineering and targeted adaptation of LLMs within education. We, therefore, argue for the integration of AI educational content into current curricula to enable a hybrid intelligent society in which students can effectively use generative AI tools such as ChatGPT.