Publications

In digital mental health, various persuasive applications have employed artificial intelligence (AI) to offer scalable support for vulnerable users. However, sustaining user trust and engagement remains a challenge, as existing persuasive design frameworks, such as the Persuasive Systems Design (PSD) framework, lack context sensitivity. This study addresses this gap by developing a novel framework for prioritizing persuasive design principles in AI-driven mental health interventions. Using a design science research (DSR) approach, we synthesized findings from a systematic literature review and a mixed-methods study (surveys and interviews) to identify user- and expert-driven design priorities. The primary result is a two-tiered prioritization framework that distinguishes between foundational “core principles” (e.g., Trustworthiness) and context-dependent “strategic enhancers” (e.g., Praise) within the PSD framework. We demonstrated its applicability in a proof-of-concept prototype. This framework provides researchers and practitioners with actionable, user-centered recommendations, mapping specific principles to a six-stage user journey to enhance trust and engagement.

Chatbots are being introduced to fill the gap between consultations, with the motivation of supporting adherence to treatment, especially in the case of chronic diseases. Such a task requires chatbots to have a persuasive ability to influence patient attitudes or behaviour, which is achievable through the advancement of Large Language Models (LLMs). By means of prompting engineering, we aim to accomplish the design goal of persuasive chatbots powered by LLMs that can systematically apply persuasion strategies and dynamically adapt such strategies to the user. We have converted well-founded persuasion strategies into guidelines for prompt crafting to enhance the persuasiveness of LLM-based chatbots. This paper contributes to the design of persuasive chatbots by providing a list of guidance on prompt crafting, meta-prompts, and a proof-of-concept of the developed guidance from a digital health application. This paper also initiates the discourse on prompt guidance as a new type of design artefact within Design Science Research (DSR).

Background: Mental health care systems worldwide face critical challenges, including limited access, shortages of clinicians, and stigma-related barriers. In parallel, large language models (LLMs) have emerged as powerful tools capable of supporting therapeutic processes through natural language understanding and generation. While previous research has explored their potential, a comprehensive review assessing how LLMs are integrated into mental health care, particularly beyond technical feasibility, is still lacking. Objective: This systematic literature review investigates and conceptualizes the application of LLMs in mental health care by examining their technical implementation, design characteristics, and situational use across different touchpoints along the patient journey. It introduces a 3-layer morphological framework to structure and analyze how LLMs are applied, with the goal of informing future research and design for more effective mental health interventions. Methods: A systematic literature review was conducted across PubMed, IEEE Xplore, JMIR, ACM, and AIS databases, yielding 807 studies. After multiple evaluation steps, 55 studies were included. These were categorized and analyzed based on the patient journey, design elements, and underlying model characteristics. Results: Most studies assessed technical feasibility, whereas only a few examined the impact of LLMs on therapeutic outcomes. LLMs were used primarily for classification and text generation tasks, with limited evaluation of safety, hallucination risks, or reasoning capabilities. Design aspects, such as user roles, interaction modalities, and interface elements, were often underexplored, despite their significant influence on user experience. Furthermore, most applications focused on single-user contexts, overlooking opportunities for integrated care environments, such as artificial intelligence--blended therapy. The proposed 3-layer framework, which consists of the L1: LLM layer, L2: interface layer, and L3: situation layer, highlights critical design trade-offs and unmet needs in current research. Conclusions: LLMs hold promise for enhancing accessibility, personalization, and efficiency in mental health care. However, current implementations often overlook essential design and contextual factors that influence real-world adoption and outcomes. The review underscores that the self-attention mechanism, a key component of LLMs, alone is not sufficient. Future research must go beyond technical feasibility to explore integrated care models, user experience, and longitudinal treatment outcomes to responsibly embed LLMs into mental health care ecosystems.

Designing effective conversational agents for healthcare requires methods grounded in expert interaction that scale to deployable agents. We present a structured, large language model-driven approach that transforms authentic expert-user dialogues into modular, patient-centered agents. Our three-step pipeline (elicitation, structure extraction, modular implementation) yields interpretable interaction phases implemented with the lightweight agent framework. In a case study on hearing-loss support for young adults, a communication-vulnerable and underserved group, the resulting agents foster purposeful engagement and surface well-being information aligned with WHOQOL domains. Compared with a fat-prompt baseline, the structured agents produced shorter, more focused exchanges and broader coverage of well-being topics. We position this method within context-aware and personalized healthcare systems and argue it offers a reproducible path toward adaptive, transparent, and trustworthy conversational agents.

The advent of increasingly powerful language models has raised expectations for conversational interactions. However, controlling these models is a challenge, emphasizing the need to be able to investigate the feasibility and value of their application. We present PROMISE (Available at: https://github.com/zhaw-iwi/promise), a framework that facilitates the development of complex conversational interactions with information systems. Its use of state machine modeling concepts enables model-driven, dynamic prompt orchestration across hierarchically nested states and transitions. This improves the control of language models’ behavior and thus enables their effective and efficient use. We show the applications of PROMISE in health information systems and demonstrate its ability to handle complex interactions.

Purpose. University career services offer individual counseling as an institutional response to career-related procrastination, but personnel limitations and the loss of motivational momentum between sessions constrain this effort. Digital interventions typically view procrastination as a mere scheduling problem or provide conversational AI as an autonomous substitute for institutional counseling. This study addresses the resulting follow-through gap by utilizing generative AI to support adherence to expert advice between counseling sessions. Design/methodology/approach. Guided by an Action Design Research (ADR) approach, this study designed, developed, and evaluated an AI-augmented persuasive system intended to mitigate procrastination by closing the loop between counseling sessions and everyday student activities. The artifact was implemented in collaboration with a university career services and evaluated with 15 students in a mixed-methods pre- and post-intervention study over 3 to 4 weeks. Findings. Results show a statistically significant reduction in state procrastination scale (APSI, −16.8%, 𝑝 = .032, 𝑑 = 0.62), large gains in career decision-making self-efficacy (CDSES-SF, +14.6%, 𝑝 = .001, 𝑑 = 1.07), and a 79.2% goal attainment rate. Qualitative insights reveal these effects were driven primarily by the system’s structural scaffolding, with the AI functioning as a contextual boundary object rather than a primary persuader. We further abstract these findings into three design principles, and discuss practical implications for system designers, university career services, and students. Originality/value. This paper challenges general practice in persuasive technology by repositioning generative AI from an autonomous coaching substitute to a context-aware, sociotechnical bridge supporting human-in-the-loop institutionalworkflows.

Persuasive chatbots powered by large language models (LLMs) are increasingly deployed in digital health to address the challenge of adherence to treatment, especially for chronic conditions. Based on established research in human-computer interaction (HCI), a common assumption is that an empathetic persona increases a chatbot's persuasiveness. Following a Research through Design (RtD) approach, we designed, implemented, and evaluated a persuasive chatbot for treatment adherence support. We assessed its effectiveness and perception in a five-week field study with 29 patients undergoing weight management, combining qualitative interviews with a computational analysis of conversation logs. Our findings reveal a paradox: while some patients appreciated the supportive tone, a large portion perceived the chatbot's empathy as passive, artificial, and ultimately unpersuasive. We conclude that a static empathetic persona is insufficient, and instead propose a sociotechnical approach where calibrated empathy and personalization are achieved through human-AI collaboration.

Efforts to influence human behaviors through digital technologies are ubiquitous, ranging from sustained lifestyle modification to personalized e-commerce. The effectiveness of such efforts relies on persuasion, which aims to achieve behavior change by leveraging psychological and communicative knowledge without resorting to coercion or deception. As these persuasive efforts increasingly incorporate generative artificial intelligence (GenAI), current information systems (IS) research lacks a robust conceptualization to address its novel challenges and opportunities, especially from a sociotechnical perspective. In this discussion paper, we address this gap by reflecting on the conceptual boundaries of persuasive artificial intelligence (persuasive AI). We propose viewing the persuasive AI system through a sociotechnical lens, where a technical subsystem (the “AI module”) and a social subsystem (the human “persuadee” and optional “persuader”) interact. Its defining process is the dynamic cycle of mutual adaptation between these subsystems to serve a persuasive purpose. Building on this, we propose to conceptualize persuasive AI as a multi-faceted concept that can be understood from four complementary views: as a (1) scientific discipline, (2) capability, (3) engineering practice, and (4) technology and system. We ground our analysis by examining GenAI’s role as a digital platform, which forms the technological foundation for these systems. Additionally, we propose an illustrative, architectural model to guide the development of these systems and derive a research agenda based on the four views. This paper aims to support practitioners in gaining a deeper understanding of persuasive AI, while encouraging researchers to challenge our proposition and explore the potential of this emerging field through collaborative, interdisciplinary efforts.