SEED FUNDING Conversational AI and social support in daily life
Conversational artificial intelligence (CAI), defined as chatbots that enable interaction via natural spoken or written language (e.g., ChatGPT or Google Gemini), is widely and increasingly used by many people in everyday life (Bühler, 2025; Rainine, 2025). By providing adaptive and personalized responses, CAI can offer informational (e.g., giving advice), instrumental (e.g., helping draft messages or organize activities), and emotional social support (e.g., expressing empathy). Such communication not only resembles interaction with other humans but may, in certain aspects and contexts, even surpass it. For example, individuals tend to evaluate CAI-generated responses as more empathic than human responses (Liu et al., 2026) and report greater interpersonal closeness following emotionally engaging interactions (Kleinert et al., in press), which may contribute to a reduced psychological burden following human-CAI interactions (Gabriels et al., 2026). Despite this growing relevance, the psychological and biological consequences of CAI-human social interaction remain largely unexplored, particularly with respect to its potential buffering effects on stress- related systems such as the hypothalamic-pituitary-adrenal (HPA) axis and the sympatho-adrenomedullary (SAM) system. Dysregulation within these systems precedes stress-related mental and physical diseases (Connor et al., 2021; George et al., 2025). Positive and supporting social interactions can buffer these effects, whereas loneliness or social loss are associated with adverse health outcomes (Aguilar-Raab et al., 2025; Holt-Lunstad, 2024; Hopf et al., 2022; Stoffel et al., 2021a; Stoffel et al., 2026). In light of the increasing prevalence of stress-related psychiatric disorders (e.g., DGPPN e. V., 2025) and the rapid rise in the use of CAI systems, the present project examines whether CAI-human interaction can buffer psychobiological stress in everyday life in a manner comparable to human interaction. Moreover, the collected data will be used to train a CAI system to optimize the timing, frequency and qualitative characteristics of supportive social interactions as a form of momentary intervention in daily life. The project employs an intensive ecological momentary assessment (EMA) design, with assessments conducted during participants’ everyday life routines. A total of 100 participants will be assessed over five consecutive workdays, with up to 15 assessments per day (~ 3,500 data points). Based on our prior work demonstrating associations between social support and psychobiological parameters in everyday life (e.g., Stoffel et al., 2021a), we expect that, with this sample size, the study will be sufficiently powered. At each assessment, participants will be asked to provide self-reports and/or saliva samples. Self- reported data will be collected via an EMA-App installed on participants’ smartphones. Core variables include salivary cortisol (assessed seven times daily), continuous heart rate (HR) and heart rate variability (HRV), as well as self-reports on human and CAI-based social interactions, mood, and subjective stress.
Aims of the project
Aim 1: Investigate associations of CAI-based social support with stress-related psychobiological parameters
To examine how CAI-based social support is associated with psychobiological parameters (salivary cortisol as indicator of HPA axis activity, HR and HRV as indicators of the SAM system, mood, and subjective stress) and to test whether these associations differ between CAI-based and human interactions.
Aim 2: Develop a CAI optimized for social support
To allow identification of time points, contextual factors and qualitative characteristics of social interactions that optimally affect psychobiological parameters (see the first aim for a definition) in daily life. Effects will be examined separately for CAI-based and human interactions.
Interdisciplinary aspects
In a radically new approach, this project integrates subjective and psychobiological measures, EMA assessments and computational network models to analyze differences between CAI-based and human social support in everyday life. To realize this endeavor, the project integrates expertise from clinical science, computer science and human–AI interaction, neuroscience, psychoneuroendocrinology, and psychology and generates findings relevant across these disciplines. In the long run, the results are expected to inform the prevention and treatment of preclinical stress-related syndromes (and related mental disorders) through CAI-based interventions; therefore, they are also relevant for the fields of mental health, prevention, and psychiatry.
Project Duration
03/2026 – 12/2026