Digital mental health (DMH) tools have extensively explored personalization of interventions to users' needs and contexts. However, this personalization often targets what support is provided, not how it is experienced. Even well-matched content can fail when the interaction format misaligns with how someone can engage. We introduce generative experience as an approach to DMH support, where the intervention experience is composed at runtime. We instantiate this in GUIDE, a system that generates personalized intervention content and multimodal interaction structure through rubric-guided generation of modular components. In a preregistered study with N=237 participants, GUIDE significantly reduced stress (p=.02) and improved user experience (p=.04) compared to an LLM-based cognitive restructuring control. GUIDE also supported diverse forms of reflection and action through varied interaction flows, while revealing tensions around personalization across the interaction sequence. This work lays the foundation for interventions that dynamically shape how support is experienced and enacted in digital settings.
Figures & tables
Figure 1. GUIDE generates intervention experiences at runtime by composing interaction structures from user context and selected interventions. The system elicits context, selects an intervention, and constructs a multimodal experience using elements such as visual guidance and timed activities. The interfaces shown are illustrative snippets of generated UX.
Figure 2. System overview of GUIDE. The system first elicits user context through a structured conversation and produces a concise summary. It then generates multiple candidate interventions from this context and selects one using rubric-guided evaluation. Given user context and the selected intervention, the system composes candidate UX realizations from a set of interaction modules (e.g., audio, text, timer) and selects the final experience through a second rubric-based selection process.
Table 1. Representative interaction primitives used to construct generated intervention experiences. The full set of primitives is shown in Table 4 .
Outcome
Measure and Scoring
Primary outcomes
Stress reduction
“How stressful is the situation you are thinking about?” measured before and after the activity ( Bhattacharjee et al., 2024a ) . Rated 1–5; Stresspre−Stresspost .
User experience
UEQ-8 user experience scale measured after the activity ( Hinderks, 2017 ) . Mean of 8 items, rescaled to [−2,2] .
Exploratory outcomes
Stress mindset improvement
8-item stress mindset scale measured before and after the activity ( Crum et al., 2013 ) . Rated 0–4; Mindsetpost−Mindsetpre .
Perceived personalization
“The suggested activity felt personalized to my specific situation.” (post, 1–5 agreement)
Table 2. Outcome measures used in the study.
Outcome
GUIDE
Control
p
Cohen’s d
(M ± SD)
(M ± SD)
Primary outcomes
Stress reduction*
0.65 ± 0.7
0.35 ± 0.8
.02
0.39
User experience*
0.49 ± 0.6
0.33 ± 0.6
.04
0.27
Exploratory outcomes
Stress mindset improvement
1.44 ± 3.1
0.75 ± 3.5
.09
0.21
Table 3. Summary of outcome measures for participants in the GUIDE and Control conditions.
Figure 3. (a) Co-occurrence of interaction types across generated intervention experiences (diagonal = total usage; off-diagonal = pairwise co-occurrence). (b) Interaction type usage across intervention combinations. Values represent the proportion of sessions within each combination that include the interaction type; CR = Cognitive restructuring, BA = Behavioral activation, RS = Regulation strategies.
Appendix figures & tables13 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 4. Ablation study results across four conditions defined by the presence or absence of rubric guidance in intervention and UX generation
Table 4. Overview of interaction primitives, their parameters, and interaction types.
Figure 5. Illustrative examples of interaction types used to construct intervention experiences. In practice, each type can take many forms and be combined in different ways depending on the intervention and user context (See Figures 7 , 8 , and 9 ).
Rubric
Description
Rationale
Narrative Flow
Whether the reflective and action oriented steps form a clear and continuous experience where each step connects naturally to the next.
Supports coherence in multi-step interventions (prior work ( Persson et al., 2025 ) )
Small Progress
Whether the sequence ends with a concrete outcome such as a clarified sentence, named emotion, reframe, or small action.
Emphasizes actionable outcomes (expert feedback)
Safe Sequencing
Whether the steps remain low intensity, clearly bounded, and easy to pause, avoiding heavy emotional processing.
Reduces risk in brief interventions (prior work ( Persson et al., 2025 ) ; expert feedback)
Explicit Alignment with Psychology Principles
Whether the activity clearly names the psychological principle and demonstrates how the steps enact it.
Improves transparency and learning (prior work ( Mohr et al., 2014 ; Persson et al., 2025 ; Paredes et al., 2014 ) )
Specificity
Whether the activity reuses the user’s phrases, routines, or constraints so the activity clearly belongs to their situation.
Enhances perceived relevance (prior work ( Yardley et al., 2015 ) )
Non-retrievability
Whether the activity depends on the user’s specific context and would not easily apply to another person.
Avoids generic responses (expert feedback)
Appendix
Table 5. Intervention judge rubrics used to evaluate the quality of generated interventions.
Rubric
Description
Rationale
Intervention-Interface Alignment
Whether the overall structure of the interface reflects the user’s request and presents modules in a clear, stepwise progression.
Ensures alignment between user intent and interaction flow (prior work ( Nielsen, 1994 ; Hartmann et al., 2008 ) )
Task Efficiency
Whether the activity can be completed with minimal friction, limited typing, and within the intended time window.
Reduces effort and supports short, feasible interventions (expert feedback)
Usability
Whether interactive controls are clear, visible, and actionable, with consistent navigation and feedback.
Improves interaction reliability and ease of use (prior work ( Nielsen, 1994 ; Hartmann et al., 2008 ; Brooke and others, 1996 ) )
Information Clarity
Whether content is structured, concise, and easy to scan, reducing cognitive load.
Supports comprehension and reduces overload (prior work ( Nielsen, 1994 ; Hartmann et al., 2008 ) )
Interaction Satisfaction
Whether the experience ends with a clear sense of completion, visual consistency, and smooth transitions.
Reinforces completion and overall experience quality (prior work ( Nielsen, 1994 ; Hartmann et al., 2008 ) ; expert feedback)
Specificity
Whether the interface incorporates the user’s specific situation through wording, examples, or UI elements.
Enhances contextual relevance and personalization (prior work ( Yardley et al., 2015 ) )
Appendix
Table 6. UX judge rubrics used to evaluate the quality of generated interaction experiences.
Figure 6. Rubric-guided selection of intervention candidates. Given a user context, multiple candidate interventions are generated and evaluated across rubric criteria, and the highest-scoring intervention is selected. UX structure is also generated in a similar process.
Outcome
Condition (GUIDE)
Pre-stress
PSS
Stress mindset (pre)
Age
Race (White)
Race (PNA)
Gender (Male)
Post-stress
-0.28** (0.10)
0.59*** (0.08)
0.01 (0.01)
-0.01 (0.01)
0.08* (0.04)
-0.15 (0.16)
-0.28* (0.11)
-0.09 (0.13)
Appendix
Table 7. Regression predicting post-intervention stress controlling for pre-stress and demographic variables
Outcome
Condition (GUIDE)
Pre-stress
PSS
Stress mindset (pre)
Age
Race (White)
Race (PNA)
Gender (Male)
UEQ mean
0.12 (0.09)
0.08 (0.06)
-0.01 (0.01)
0.01 (0.01)
0.01 (0.03)
-0.03 (0.15)
0.01 (0.20)
-0.05 (0.10)
Appendix
Table 8. Regression predicting post-intervention user experience (UEQ) controlling for pre-stress and demographic variables
Outcome
Condition (GUIDE)
Pre-stress
Log-transformed time
Post-stress
-0.30** (0.10)
0.67*** (0.07)
0.02 (0.07)
Appendix
Table 9. Regression predicting post-intervention stress controlling for pre-stress and time spent
Outcome
Pre-stress
Audio
Visual
Temporal
Post-stress
0.22** (0.08)
-0.18 (0.18)
-0.08 (0.20)
0.06 (0.15)
Appendix
Table 10. Regression predicting post-intervention stress within the GUIDE condition using interaction type presence
Figure 7. Example of a generated support experience for a user stressed about an exam, composed of a timer-based activity and list entry input. The two screens show consecutive steps in the same interaction flow, illustrating how the experience unfolds through structured guidance and user input.
Figure 8. Example of a generated support experience for a user stressed about a recent meeting, composed of a list entry input, an audio message, and text-based questions. The two screens show consecutive steps in the same interaction flow.
Figure 9. Example of a generated support experience for a user stressed about a family member, composed of a visual card pair and text-based questions. The two screens show consecutive steps in the same interaction flow.
Mindfulness meditation is a widely accessible and evidence-based method for supporting mental health. Despite the proliferation of mindfulness meditation apps, sustaining user engagement remains a persistent challenge. Personalizing the meditation experience is a promising strategy to improve engagement, but it often requires costly and unscalable manual effort. We present MindfulAgents, a multi-agent system powered by large language models that (1) generates guided meditation scripts based on an expert-established mindfulness framework, (2) encourages users' reflection on emotional states and mindfulness skills, and (3) enables real-time personalization of the mindfulness meditation experience for each user. In a formative lab study (N=13), MindfulAgents significantly improved in-session engagement (p = 0.011) and self-awareness (p = 0.014), and reduced momentary stress (p = 0.020). Furthermore, a four-week deployment study (N=62) demonstrated a notable increase in long-term engagement (p = 0.002) and level of mindfulness (p = 0.023). Participants reported that MindfulAgents offered more relevant meditation sessions personalized to individual needs in various contexts, supporting sustained practice. Our findings highlight the potential of LLM-driven personalization for enhancing user engagement in digital mindfulness meditation interventions.
Mengyuan Millie Wu, Zhihan Jiang, Yuang Fan +9
Columbia University, New York, NY, USA · St. Margaret’s Episcopal School, San Juan Capistrano, CA, USA · Carnegie Mellon University, Pittsburgh, PA, USA +3
Artificial intelligence (AI)-enabled digital interventions, including Generative AI (GenAI) and Human-Centered AI (HCAI), are increasingly used to expand access to digital psychiatry and mental health care. This PRISMA-ScR scoping review maps the landscape of AI-driven mental health (mHealth) technologies across five critical phases: pre-treatment (screening/triage), treatment (therapeutic support), post-treatment (remote patient monitoring), clinical education, and population-level prevention. We synthesized 36 empirical studies implemented through early 2024, focusing on Large Language Models (LLMs), machine learning (ML) models, and autonomous conversational agents. Key use cases involve referral triage, empathic communication enhancement, and AI-assisted psychotherapy delivered via chatbots and voice agents. While benefits include reduced wait times and increased patient engagement, we address recurring challenges like algorithmic bias, data privacy, and human-AI collaboration barriers. By introducing a novel four-pillar framework, this review provides a comprehensive roadmap for AI-augmented mental health care, offering actionable insights for researchers, clinicians, and policymakers to develop safe, effective, and equitable digital health interventions.
Yang Ni, Fanli Jia
School of International and Public Affairs Columbia University New York · Department of Psychology Seton Hall University South Orange
User Experience Research (UXR) in a legal and regulatory contexts presents unique challenges that require specialised approaches to protect vulnerable populations whilst generating actionable insights. Digital consultation, appointment booking, and medication delivery platforms show promise for extending care access; however, their real-world effectiveness is curtailed by an absence of theoretically grounded user experience research (UXR) methodologies that adequately account for the psychosocial conditions of these populations. This paper introduces a Generative AI-augmented UXR methodology, grounded in the UXR Point of View (PoV) Playbook, to guide the design of psychologically safe, low-cognitive-load digital health interventions for MSM and transgender individuals living with HIV/AIDS in Nigeria. Drawing from empirical research involving co-design workshops, thematic analysis, and requirements engineering, the methodology is operationalised through a four-stage UXR process encompassing AI-supported hypothesis generation, foundational planning, insight generation via Building Blocks, and the construction of stakeholder-specific PoV narratives. This process results in ten theory-informed UXR Play Cards that translate psychological mechanisms and empirical findings into actionable design guidance. Each play contains actionable tasks, AI-augmented approaches, and ethical guardrails tailored for research with marginalised populations. The output is a set of ten theory-informed UXR Play Cards translating psychological insight and empirical evidence into actionable design guidance. The core contribution is a replicable, stigma-aware, and privacy-centred framework for responsible GenAI use in UXR practice, advancing human-centred digital health design for marginalised communities.