Exploring Expert Perspectives on Wearable-Triggered LLM Conversational Support for Daily Stress Management Conference: Interactive Health Conference; July 05–08, 2026; Porto, Portugal Interactive Health Conference (IH ’26), July 05–08, 2026, Porto, Portugal DOI: 10.1145/3786579.3804960 ISBN: 979-8-4007-2422-0/2026/07 CCS: Human-centered computing Empirical studies in ubiquitous and mobile computing CCS: Human-centered computing Ubiquitous computing CCS: Human-centered computing User studies
Poorvesh Dongre, Sameer Neupane, Priyanka Jadhav, Nikitha Donekal Chandrashekar, Christian Webb, Santosh Kumar, Denis Gračanin · 2026
Read Exploring Expert Perspectives on Wearable-Triggered LLM Conversational Support for Daily Stress Management Conference: Interactive Health Conference; July 05–08, 2026; Porto, Portugal Interactive Health Conference (IH ’26), July 05–08, 2026, Porto, Portugal DOI: 10.1145/3786579.3804960 ISBN: 979-8-4007-2422-0/2026/07 CCS: Human-centered computing Empirical studies in ubiquitous and mobile computing CCS: Human-centered computing Ubiquitous computing CCS: Human-centered computing User studies by Poorvesh Dongre, Sameer Neupane, Priyanka Jadhav, Nikitha Donekal Chandrashekar, Christian Webb, Santosh Kumar, Denis Gračanin (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-04915.txt.
Introduction
Dongre and five colleagues probe how wearable stress alerts could pair with chatbot support for daily stress care. Their EmBot app joins wearable-triggered stress detection to language-model conversation, then serves as a design probe in interviews with 15 mental health experts. The experts surface tensions in timing, trust, and tone when a device ping launches a supportive chat. Those early design lessons aim to guide future systems that link body signals to dialogue without overstepping clinical care.
Catalog record
| CREATOR | Poorvesh Dongre, Sameer Neupane, Priyanka Jadhav, Nikitha Donekal Chandrashekar, Christian Webb, Santosh Kumar, Denis Gračanin |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 20359 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.04915
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Exploring Expert Perspectives on Wearable-Triggered LLM Conversational Support for Daily Stress Management Conference: Interactive Health Conference; July 05–08, 2026; Porto, Portugal Interactive Health Conference (IH ’26), July 05–08, 2026, Porto, Portugal DOI: 10.1145/3786579.3804960 ISBN: 979-8-4007-2422-0/2026/07 CCS: Human-centered computing Empirical studies in ubiquitous and mobile computing CCS: Human-centered computing Ubiquitous computing CCS: Human-centered computing User studies Poorvesh Dongre, Sameer Neupane, Priyanka Jadhav, Nikitha Donekal Chandrashekar, Christian Webb, Santosh Kumar, Denis Gračanin Abstract . Wearable devices increasingly support stress detection, while LLMs enable conversational mental health support. However, designing systems that meaningfully connect wearable-triggered stress events with generative dialogue remains underexplored, particularly from a design perspective. We present EmBot, a functional mobile application that combines wearable-triggered stress detection with LLM-based conversational support for daily stress management. We used EmBot as a design probe in semi-structured interviews with 15 mental health experts to examine their perspectives and surface early design tensions and considerations that arise from wearable-triggered conversational support, informing the future design of systems for daily stress management and mental health support. † † cc-license: by 1. Introduction & Background Stress is a mental state arising from cognitive or emotional overload when demands exceed an individual’s capacity. While short-term stress can be helpful, chronic stress is associated with adverse mental health outcomes such as mood and eating disorders and depression ( 4 ; 30 ; 14 ) , and negatively affects well-being and productivity ( 10 ; 11 ) . Over the past decade, technological advances have enabled systems for daily stress management. Passive health sensing applications, particularly wearables such as wristbands, smartwatches, and rings, capture physiological and behavioral signals, which are used to detect and monitor stress ( 23 ; 13 ; 31 ; 34 ; 26 ; 5 ; 1 ; 17 ) . Beyond passive sensing, wearables can also provide stress interventions, including biofeedback ( 22 ; 27 ; 24 ; 16 ) and just-in-time adaptive prompting ( 6 ; 25 ; 29 ; 2 ) . However, prior work highlights challenges including false positives, notification fatigue, disengagement, and contextual ambiguity in wearable data ( 21 ; 22 ) . These challenges suggest that stress detection and monitoring alone is insufficient; a key unresolved challenge is how to translate wearable data into timely, meaningful, and user-appropriate interactions. LLMs introduce new opportunities for daily stress monitoring and intervention through their natural language inference and generative capabilities. Prior work explores LLMs for emotion logging ( 28 ) , therapy chatbots ( 15 ) , and summarizing wearable-derived insights for mental health support ( 7 ) . LLMs have also been used for conversational mental health support, including therapeutic-style dialogue ( 19 ; 18 ) and generative journaling or coaching systems ( 20 ; 33 ) . However, many LLM-based mental health systems operate independently of wearable sensing, relying primarily on user-provided text. As a result, they lack grounding in wearable data, limiting their ability to proactively and empathically support users in everyday contexts. Recent efforts have investigated using LLMs for sensemaking of wearable data for activity ( 12 ) , sleep ( 32 ) , and general health & behavioral health ( 7 ) . These systems primarily focus on post-hoc interpretation or summarization of sensed data rather than supporting real-time, user-facing interaction. In parallel, emerging systems have explored wearable-triggered conversational support for stress management ( 21 ; 9 ; 8 ) . While promising, these efforts largely demonstrate feasibility and do not yet provide a clear understanding of how such interactions should be designed in real-world mental health contexts. In particular, the role of mental health experts in shaping these interactions remains underexplored which is essential in ensuring safety, appropriateness, and clinical relevance in early-stage system design. To address this gap, we developed EmBot (short for Empathic Chatbot), a functioning mobile application that uses wearable-triggered stress events to initiate and ground LLM-based conversational support and used it as a design probe to ground discussions with 15 mental health experts (researchers and clinicians) on the design of wearable-triggered LLM conversational systems for daily stress management. Experts interacted with EmBot (either through guided hands-on use or a structured walkthrough) and evaluated its interaction flow, triggering mechanisms, feedback design, and potential real-world applicability. Our goal was to elicit expert perspectives to inform early-stage design decisions for wearable-triggered LLM conversational systems, with a focus on how such systems should be designed, triggered, and integrated into real-world mental health contexts. 2. System Design EmBot ( Figure 1 ) was developed as a functional mobile design probe to support grounded discussion during expert interviews. The system was implemented to allow experts to experience a plausible interaction flow of a wearable-triggered LLM conversational system for daily stress management. As a design probe, EmBot was not intended to evaluate model performance, but to elicit interaction-level feedback grounded in a concrete user experience. The mobile application is paired with a wearable device that monitors raw physiological signals and detects stress ( 1(a) ). For the interviews, stress events were simulated to ensure consistent scenarios across participants and to focus discussion on interaction design rather than model accuracy. When a notification appears, users can review and respond to the detected stress event ( 1(b) and 1(c) ). This interaction stage was designed to preserve user agency by allowing users to confirm, reject, or contextualize the detection before proceeding. Following user feedback, EmBot initiates a conversational interaction powered by an LLM ( 1(d) ). The conversation is grounded in the detected event and user input, offering reflective prompts and coping-oriented dialogue. Rather than presenting static advice, the probe demonstrates how wearable data can be translated into conversational engagement. To support longer-term reflection, users can revisit a history of stress detections and prior conversations ( 1(e) ). This feature allows exploration of how wearable-triggered LLM systems might support stress management and sustained engagement over time. (a) Detection: Notification when stress is detected. (b) Detection: Screen with real-time stress detection. (c) Feedback: User stress detection agreement scale. (d) Support: In-app chat support powered by LLM. (e) Reflection: History with all stress detections. Figure 1 . Interaction Stages in EmBot: Detection, Feedback, Support, and Reflection. Interaction stages when using EmBot. Detection stage includes a notification when stress is detected followed by real-time stress detection. It is followed by Feedback stage complemented by an LLM-based in-app chat (Support stage). Finally, Reflection stage provides a history of all stress detections. 3. Methodology We conducted semi-structured interviews with 15 mental health experts (18 interviews conducted; 3 excluded due to recording issues). These included licensed clinicians, clinical researchers, and computer scientists whose primary research focus is digital mental health. Experts were initially recruited via professional networks and direct outreach to researchers and clinicians in digital mental health, followed by snowball sampling through participant referrals. As discussed, EmBot was used as a functional design probe during interviews. All experts were exposed to the same four interaction stages: Detection, Feedback, Support, and Reflection. As discussed, stress detection events and associated data were simulated and experts were instructed to interpret these scenarios as plausible real-world situations, allowing them to evaluate system behavior and design implications independent of model accuracy. Interviews were conducted in-person and remotely, with both groups first viewing a