Paper deep dive
Personalizing Personal Health Interfaces: Co-Design with Generative AI
Karthik S. Bhat, Vidhi Shah, Vedika Agnihotri, Dong Whi Yoo, Koustuv Saha
Intelligence
Status: not_run | Model: - | Prompt: - | Confidence: 0%
Entities (0)
Relation Signals (0)
No relation signals yet.
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:Personal health interfaces present wellbeing data through standardized dashboards that rarely fit how people interpret or act on it. Personalizing them to what people would like to see for themselves often requires design and technical expertise, a barrier that generative AI may potentially lower. Therefore, we ask what designs emerge and how it enables and constrains the design process. We conducted a co-design study where 14 participants redesigned Google and Apple Health interfaces using Figma Make. Participants reimagined interfaces that supported personal context, future planning, and interactive experiences, yet conversational AI designs converged around chat-window conventions. AI helped materialize loosely articulated ideas, but model defaults and generation latency shaped iteration. The process more readily operationalized interpretability and accountability than privacy, trust, and emotional safety. Generative co-design let participants create interfaces directly, blurring the boundary between intentions and model defaults. We discuss implications for preserving agency and flexible user-directed interfaces.
Tags
Links
- Source: https://arxiv.org/abs/2609.15046v1
- Canonical: https://arxiv.org/abs/2609.15046v1
Trouble viewing inline? Open PDF directly â
Full Text
105,657 characters extracted from source content.
Expand or collapse full text
Personalizing Personal Health Interfaces: Co-Design with Generative AI Conference: ; ; CCS: Human-centered computing Empirical studies in collaborative and social computingCCS: Applied computing Psychology Karthik S. Bhat Affiliation: Drexel University, Philadelphia, PA, USA email: ksbhat@drexel.edu , Vidhi Shah Affiliation: Drexel University, Philadelphia, PA, USA email: vps43@drexel.edu , Vedika Agnihotri Affiliation: Drexel University, Philadelphia, PA, USA email: va432@drexel.edu , Dong Whi Yoo Affiliation: Indiana University Indianapolis, Indianapolis, IN, USA email: dy22@iu.edu and Koustuv Saha Affiliation: University of Illinois Urbana-Champaign, Urbana, IL, USA email: ksaha2@illinois.edu 2026 Abstract. Personal health interfaces present wellbeing data through standardized dashboards that rarely fit how people interpret or act on it. Personalizing them to what people would like to see for themselves often requires design and technical expertise, a barrier that generative AI may potentially lower. Therefore, we ask what designs emerge and how it enables and constrains the design process. We conducted a co-design study where 14 participants redesigned Google and Apple Health interfaces using Figma Make. Participants reimagined interfaces that supported personal context, future planning, and interactive experiences, yet conversational AI designs converged around chat-window conventions. AI helped materialize loosely articulated ideas, but model defaults and generation latency shaped iteration. The process more readily operationalized interpretability and accountability than privacy, trust, and emotional safety. Generative co-design let participants create interfaces directly, blurring the boundary between intentions and model defaults. We discuss implications for preserving agency and flexible user-directed interfaces. Keywords: self-reflection, personal health informatics, conversational AI 1. Introduction Personal health technologies increasingly shape how people monitor and understand their health and wellbeing (Piwek et al., 2016). Mobile applications and wearable devices allow individuals to continuously track physical activity, sleep, heart rate, menstrual cycles, and other aspects of their everyday lives (Piwek et al., 2016). Many of these technologies present personal data through dashboards, goals, and feedback intended to encourage healthier behavior (Consolvo et al., 2006; Patel et al., 2015). However, recording and presenting more data does not necessarily help people understand what the data means or translate it into meaningful action (Patel et al., 2015). Prior HCI research has further shown that interface design influences what people notice, how they reflect on their experiences, and which actions they consider taking (Fleck and Fitzpatrick, 2010; Bentvelzen et al., 2022). This has motivated the need for personal health interfaces that help individuals interpret those measurements in relation to their own lives and priorities (Cecchinato et al., 2019). One strand of HCI research has examined how personal informatics technologies can support self-reflection. Li et al. described how people move through different stages of preparing, collecting, integrating, and acting on personal data (Li et al., 2010). Rooksby et al. and Epstein et al. further showed that self-tracking is embedded within everyday routines and changes alongside individualsâ lives (Rooksby et al., 2014; Epstein et al., 2015). Prior work has also identified difficulties in collecting, integrating, and interpreting personal data (Choe et al., 2014). Baumer therefore argued that reflective technologies should help people question and interpret their experiences rather than merely present information back to them (Baumer, 2015). Recent research has further explored how interfaces can help people recognize discrepancies in their behavior and interpret wellbeing data in context (Bhat et al., 2026; Zhu et al., 2026a). A second strand of research has examined how people can participate in designing the technologies intended for them. Participatory design positions individuals as experts in their own experiences and involves them in shaping technological decisions (Spinuzzi, 2005; Harrington et al., 2019). Sanders and Stappers described co-design as an approach in which people participate directly in the creative process rather than only responding to completed designs (Sanders and Stappers, 2008). Steen characterized co-design as a process of joint inquiry and imagination through which participants can explore possible futures (Steen, 2013). Prior work has applied these approaches to the design of health and wellbeing technologies (van Hierden et al., 2021; Poot et al., 2023). Value-sensitive design has further emphasized examining how peopleâs values become reflected in particular technological features and decisions (Friedman et al., 2013). These approaches provide ways for people to articulate what they want from personal health technologies and why those preferences matter. Together, these strands raise an important question: what if individuals could move beyond discussing or evaluating personal health interfaces and directly design alternatives for themselves? In many co-design studies, participants articulate needs through interviews, workshops, sketches, or low-fidelity prototypes, after which researchers or professional designers translate their contributions into functional interfaces. Generative interface tools such as Figma Make11 1 https://w.figma.com/make/ create an opportunity for participants to perform more of this translation themselves. However, introducing generative AI into co-design may also influence which ideas are pursued and how participantsâ intentions become represented. Li et al. found that UX professionals viewed generative AI as an assistive tool but continued to emphasize the importance of human agency over design outcomes (Li et al., 2024). Wadinambiarachchi et al. found that AI-generated examples could increase design fixation and reduce the variety of ideas produced (Wadinambiarachchi et al., 2024). Recent work has similarly shown that AI design tools can support rapid interface production while constraining how designers explore and develop their ideas (Kobiella et al., 2026). Conceptually, it therefore remains unclear how generative AI mediates the translation of participantsâ values into design choices. Practically, we need to understand how generative prototyping can be incorporated into co-design while preserving participantsâ agency over the resulting interfaces. To address these gaps, we examine how individuals use generative AI-based co-design to reimagine personal health interfaces which they are already familiar with (e.g., Google Health and Apple Health). Our study is guided by the following research questions (RQs): RQ1: What design patterns emerge through re-imagining personal health interfaces during generative co-design? RQ2: How does incorporating generative AI into a co-design method create opportunities and limitations for designing personal health interfaces? We conducted a co-design study with 14 participants who used mobile or wearable technologies to understand their health. Participants reflected on their experiences with familiar platforms, and identified aspects of these interfaces that they considered important or wanted to change. Then, using Figma Make, participants translated these ideas into new interface designs and iteratively revised the generated outputs. During this activity, participants thought aloud as they formulated prompts, evaluated generated features, and explained their design decisions. Finally, participants reflected on their designs and their experiences using Figma Make. Through an inductive thematic analysis of participantsâ interviews, prompts, generated interfaces, and reflections, we find that participants reimagined personal health interfaces around three broad directions: connecting health data to provide greater context, extending tracking toward future planning and action, and introducing more personalized forms of visualization and interaction. Participants frequently brought together data that existing interfaces presented separately, creating visualizations and insights that connected health metrics to one another and to everyday circumstances such as weather, schedules, and illness. While these visualizations varied substantially across participants, conversational interfaces tended to converge around familiar chatbot designs. We further find that generative co-design enabled participants to rapidly materialize and refine ideas, including features they had not initially articulated, but that the generated outputs also shaped the subsequent design process. Participants often retained or adapted features introduced by the model, their sense of control depended on how closely these additions aligned with their intentions, and generation latency affected how freely they iterated on the design outputs. This work makes three contributions. First, we characterize an emerging design space for personal health interfaces centered on user-defined relationships across health and contextual data, and identify implications for supporting these relationships in platform-scale health systems. Second, we contribute an empirical account of generative co-design on how generative interfaces redistribute the work of translating participant ideas into functional prototypes. In particular, we identify a potential attribution problem that could complicate how researchers interpret co-designed artifacts when both participants and generative systems contribute to the resulting design. Third, we offer methodological guidance for conducting and interpreting generative co-design, including strategies for eliciting divergent designs and distinguishing participant-specified features from those introduced by the generative system. 2. Related Work 2.1. Personal Health Interfaces for Wellbeing Self-Reflection Personal informatics systems have long sought to support self-awareness by helping people collect, visualize, and reflect on data about their health and everyday behavior (Li et al., 2010; Rapp and Cena, 2016; Epstein et al., 2015). Yet, user-facing interfaces for these systems have often operationalized reflection through conventional dashboards showing metrics, progress indicators, and comparisons against predefined goals, placing much of the work of interpreting what those data mean in context on users themselves (Kim et al., 2016; Kim et al., 2017; Ayobi et al., 2018). Research on reflective informatics has consequently questioned designs that equate successful reflection with monitoring or goal attainment, arguing for systems that better support curiosity, interpretation, and sensemaking (Baumer et al., 2014; Rapp and Tirassa, 2017; Bhat and Kumar, 2020; Zhu et al., 2026b). This distinction is particularly important for wellbeing, where the significance of a data point may depend on experiences that are difficult to capture quantitatively, such as stress, fatigue, motivation, social circumstances, or disruptions to everyday routines (Pantzar and Ruckenstein, 2017; Choe et al., 2014). Qualitative and flexible approaches to self-tracking reveal that people construct meaning through annotations, narratives, comparisons, and other practices that connect data to their lived experiences (Karkar et al., 2016; Morris et al., 2018; Mols et al., 2016). Prior work characterizes reflection as extending from describing what happened to considering alternative explanations and questioning the assumptions through which experiences are understood (Fleck and Fitzpatrick, 2010; Bentvelzen et al., 2022; SlovĂĄk et al., 2017). This human-centered approaches to technology design also raises questions about how interfaces communicate potentially sensitive interpretations (Shen et al., 2022; Poot et al., 2023; van Hierden et al., 2021). For example, systems that demand burdensome tracking, obscure how conclusions are reached, or constrain usersâ control over their data and experiences can undermine engagement and trust (Choe et al., 2014; Rahman and Adaji, 2025). Building on research on reflective personal informatics and compassionate wellbeing technologies (Li et al., 2010; Baumer, 2015; Epstein et al., 2015; van Lotringen et al., 2025), including recent systems that support discrepancy-based reflection and collaborative interpretation of wellbeing data (Bhat et al., 2026; Zhu et al., 2026a), our study treats personal health interfaces not simply as mechanisms for presenting data or encouraging behavioral change, but as spaces for interpretation and emotionally sensitive engagement. We extend this body of work by moving beyond participantsâ evaluation of researcher-designed interfaces and examining the interface possibilities that participants themselves create. In doing so, we surface how individuals could translate their experiences with personal health data into concrete design preferences. 2.2. Participatory and Co-Design Approaches in Digital Health Participatory design positions individuals as experts in their lived experiences and involves them in shaping technologies intended for them (Spinuzzi, 2005; Sanders and Stappers, 2008; Steen, 2013; Harrington et al., 2019). Co-design similarly frames design as a process of joint inquiry and imagination through which participants and designers develop possibilities together (Sanders and Stappers, 2008; Steen, 2013). Complementarily, value-sensitive design makes values implicated in a system an explicit concern, including whose values are represented and how they become embedded in design choices (Friedman et al., 2013; Borning and Muller, 2012; Zhu et al., 2018). Prior work further emphasizes that values are situated in peopleâs lived experiences and should be elicited through engagement rather than assumed in advance (Le Dantec et al., 2009; Shilton, 2013). Within digital health, these approaches commonly move from identifying participantsâ needs to developing and evaluating prototypes (Poot et al., 2023; Lindgren and Kilic, 2025). For example, van Hierden et al. organized co-design as a five-stage process spanning evidence gathering, participatory discovery, prototyping, and pilot evaluation (van Hierden et al., 2021). Prototypes and probes further allow participants to express tacit needs through tangible materials rather than abstract discussion alone (Buchenau and Suri, 2000; Hutchinson et al., 2003). Recent HCI studies have used co-design to explore how shared displays might support family co-regulation (Silva et al., 2024) and how conversational food journals might accommodate personal goals, everyday constraints, and preferences for control (Silva et al., 2025). In mental health, participatory ideation, feature cards, and low-fidelity wireframes have helped clients and therapists articulate opportunities for collaborative goal setting (Oewel et al., 2024). Prior work has also examined how participatory methods can elicit and operationalize the values that should guide digital health technologies (Friedman et al., 2013; Huang et al., 2026). Rooper et al. found that value-elicitation tools should translate patientsâ priorities into concrete actions rather than leave them as abstract preferences (Rooper et al., 2025). Other studies have used design cards, narratives, and prototypes to help participants compare alternatives and articulate health experiences that may be difficult to express directly (Rekkas et al., 2025; Cummings et al., 2022). At a larger scale, participatory approaches have been used to develop health-system principles and requirements around equity, trust, autonomy, and accountability (Piera-JimĂ©nez et al., 2026; Dantas et al., 2026; Vinay and Kowatsch, 2026). However, Duffy et al. found that digital stakeholders often evaluated co-design through user experience and retention, whereas health stakeholders emphasized safety and efficacy (Duffy et al., 2025). These differences also raise questions about whose expertise shapes design decisions. Yoo et al. argue for engaging patients as domain experts throughout mental health AI research (Yoo et al., 2024; Yoo et al., 2026), and participatory design work on diabetes technologies highlights tensions between clinical standards and participant agency (Baseman et al., 2025). Across this work, participants typically articulate their needs and values through discussions, cards, stories, sketches, or researcher-facilitated prototypes. These artifacts make participantsâ priorities more concrete, but their translation into functional interfaces often remains mediated by researchers or designers. Accordingly, these studies demonstrate how participatory activities make situated needs and values concrete through design artifacts. Our work builds on this foundation by examining how participants use generative AI to create and revise interactive personal health interfaces. We focus on how participantsâ values become expressed, negotiated, or constrained as they accept, reject, and modify the systemâs design suggestions. 2.3. Generative AI for Interface Ideation and Prototyping Generative AI is emerging as a tool for design ideation, co-creation, and interface prototyping (Weisz et al., 2024; Li et al., 2024; Moruzzi and Margarido, 2024). Generative UI (GenUI) systems translate natural-language descriptions into outputs ranging from static interface images to executable prototypes (Chen et al., 2025; Chen et al., 2026). Figma Make uses source-code generation to create functional interfaces directly from such prompts. Beyond producing initial drafts, AI-supported design tools can help users explore alternatives, revisit assumptions, and carry revisions across connected design stages (Suh et al., 2025). Recent work on generative and malleable user interfaces further combines prompting with direct manipulation, allowing users to adapt generated outputs as their needs and goals evolve (Cao et al., 2025). These capabilities also introduce challenges in directing and evaluating generation, such as the need for users to articulate their goals, assess whether outputs meet their intentions, and determine how to revise them (Tankelevitch et al., 2024). Prompting strategies shape this work: structured input can surface design requirements but requires additional effort upfront, while presenting multiple alternatives can broaden exploration but complicate comparison (Chen et al., 2026). Continued customization can also become tedious when changes require repeated prompting or exceed the interface structures supported by the system (Cao et al., 2025). Recent HCI work examines how generative AI supports users in developing and refining design intentions. Systems help users make vague ideas concrete, explore structured alternatives, and reuse directions for further exploration (Son et al., 2024; Suh et al., 2024; Choi et al., 2026). These possibilities depend on support for expressing intentions that may change as users encounter and evaluate generated outputs (Kim et al., 2026b). For generated interfaces, maintaining alignment also requires preserving usersâ choices about how information is grouped, prioritized, and described across successive generations (Kim et al., 2026a). Generative AI can accelerate ideation and prototyping while shifting design work toward interpreting and revising generated outputs (Li et al., 2024; Kobiella et al., 2026). However, reliance on generated starting points can constrain exploration, promote design fixation and similarity across designs, and reduce perceived ownership (Wadinambiarachchi et al., 2024; Kobiella et al., 2026). Recently, within health contexts, Ishita et al. built CASEbot which supports people in designing personalized self-experiments, revealing tensions between structured guidance and autonomy (Ishita et al., 2026), and Bhattacharjee and et al. built GUIDE to help generate both mental health intervention content and multimodal interaction structures, extending personalization to how users engage with support (Bhattacharjee and et al., 2026). These recent developments raise questions about how generative interface tools support participation in health technology design. Our work examines how participantsâwith and without design experienceâuse these tools within a co-design process to translate their needs and values into functional interface prototypes. We contribute an empirical account of how participants prompt, interpret, and revise AI-generated interface concepts, and how these interactions shape their agency and ownership over the resulting designs. 3. Study Design and Methods The goal of our research was to unpack how incorporating generative AI into co-design would allow for reimagining personal health interfaces and what design patterns emerge in the process. We conducted this study between June and August 2026, and our study received Institutional Review Board (IRB) approval from our universities. 3.1. Participant Recruitment We began participant recruitment through university-specific online communities on Reddit, focusing in particular on the college subreddits of the universities where the authors hold affiliations. Prior work has noted Redditâs popularity among youth and the college student demographic (Statista, 2021; Saha and De Choudhury, 2017), with documented success recruiting participants from college-specific subreddits (Bhat et al., 2026; Zhu et al., 2026a). We posted our recruitment flyer containing an interest form that included demographic questions, prior experience with smartphone-based health tracking like Google Health or Apple Health, prior experience with design tools like Figma, and contact information for further participation. Our inclusion criteria required that the participants were older than 18 years of age, lived in the US, and currently used wearable or mobile technologies to make sense of their health. Our interest form encouraged interested participants to provide institutional email IDs to allow for filtering out bots, duplicate responses, and participants who did not meet the inclusion criteria. We received 133 responses to our interest form between June and August 2026. We initiated subsequent communications with a subset of these participants to maximize diversity in experiences with health tracking and design, and discovered that several survey respondents were located outside the US and were thus ineligible to participate in our research. We then began snowball sampling from participants who met the inclusion criteria, while ensuring the required range of experiences with design tools. This led to a final set of 14 participants before we arrived at saturation in our data. We compensated our participants with $15 USD after they completed the study. Participant information is available in Table 1. Table 1. Participant Demographics and Health App Usage PatternsTable listing demographics and health app usage for 14 participants (P1 through P14). Columns record sex, race, age bracket, design experience, occupation, app(s) used, tracking frequency, data sources, and metrics tracked. Most participants are aged 18 to 24, identify as Asian, and are students; a smaller number are 25 to 34, employed, Black or African American, or White. Apple Health is used by the large majority of participants, often alongside Whoop or Google Health, mainly via automatic phone detection. Steps, sleep, heart rate, and workouts are the most commonly tracked metrics, while blood pressure, nutrition, and water intake are tracked by only a few participants. ID Sex Race Age Design Exp. Occupation App(s) Used Frequency Data Sources Metrics Tracked P1 Female Asian 18â24 Yes Student Apple Health Weekly Phone (autodetected); Manual Entry Steps; Sleep; Heart Rate P2 Female Asian 18â24 Yes Student Apple Health Few times a week Phone (autodetected) Steps; Menstrual Cycle; Nutrition; Blood Pressure; Weight P3 Male Black or African American 18â24 Somewhat Student Google Health, Apple Health Monthly or lesser Phone (autodetected) Steps; Weight; Heart Rate P4 Male White 18â24 Somewhat Student Google Health Daily Smart Watch; Phone (autodetected) Steps; Sleep; Workout P5 Male Black or African American 25â34 No Employed Google Health Few times a week Phone (autodetected) Sleep; Steps; Nutrition; Workout; Water-intake P6 Male Asian 18â24 Somewhat Employed Apple Health, Whoop Daily Smart Watch; Phone (autodetected) Steps; Sleep; Heart Rate; Nutrition; Weight; Calories; Workout P7 Female Asian 18â24 Yes Employed Apple Health, Whoop Daily Phone (autodetected) Steps; Menstrual Cycle P8 Female Asian 18â24 No Employed Apple Health Few times a week Manual Entry; Phone (autodetected) Menstrual Cycle; Steps P9 Male Asian 18â24 Somewhat Employed Apple Health, Whoop Few times a week Phone (autodetected); Smart Watch Steps; Sleep; Heart Rate; Nutrition; Calories; Workout P10 Female Prefer Not to Say 18â24 Somewhat Student Apple Health Daily Smart Watch; Phone (autodetected) Steps; Sleep; Heart Rate; Menstrual Cycle; Blood Pressure; Weight; Workout P11 Male Asian 18â24 No Student Apple Health Daily Smart Watch Steps;Sleep;Heart Rate;Blood Pressure;Calories;Workout P12 Female Asian 18â24 Yes Student Apple Health Few times a week Phone (autodetected) Steps; Menstrual Cycle; Weight; Calories; Workout P13 Female Asian 18â24 No Student Apple Health Few times a week Smart Watch; Phone (autodetected) Steps; Heart Rate; Calories; Workout P14 Female Asian 18â24 Yes Student Apple Health Daily Smart Watch; Phone (autodetected) Sleep; Menstrual Cycle 3.2. Data Collection and Co-Design Activities Prior to the start of the interviews, we provided the participants with an Entry Survey that began with obtaining their informed consent to participate in our research. This survey asked questions about their familiarity with health apps, their usage frequency, devices used for collecting health data, health metrics they track and if they had prior experience working with Figma. A section of the survey focused on the participants scoring the importance of 9 statements about health apps on a Likert scale ranging from not important to very important. These statements were meant to inspire design ideas such as: having numbers with context, making connections across different metrics, getting suggestions or advice that feels relevant to their life, trusting AI in analyzing health data, having synthesized data with the option of digging deeper, feel that the app understands them and their lifestyle, considering external factors such as workload, injury or schedule in their health data tracking, and being transparent with how metrics were calculated. Their score on this questionnaire were used to asses their preferences and later used as a starting point for the ideation phase in the interview and co-design session. Summary results from this questionnaire can be found in Table 3. We conducted our interviews and co-design activities via Zoom video calls. The sessions lasted 50 to 62 minutes with an average of 53 minutes. All sessions were conducted in English, and they were transcribed automatically by Zoom and were later verified for accuracy by the authors. A four-step process diagram shows the study session flow. Step 1, User Interview, understands the user's health app behavior. Step 2, Ideation, uses sticky notes in FigJam. Step 3, Generative Co-design, uses Figma Make (AI). Step 4, Evaluation, assesses the designs and process. Figure 1. A four-step process diagram showing the study session flowA four-step process diagram shows the study session flow. Step 1, User Interview, understands the user's health app behavior. Step 2, Ideation, uses sticky notes in FigJam. Step 3, Generative Co-design, uses Figma Make (AI). Step 4, Evaluation, assesses the designs and process. The study session was divided into 4 parts: the interview section, FigJam ideation, generative co-design using Figma Make, and evaluations, as depicted in Fig. 1. Interviews After obtaining participantsâ verbal informed consent, we started with initial interview questions that sought to understand their thoughts, concerns, and confusions using their current preferred health apps (Apple Health/Google Health), as well as their experiences with and preferences for setting and tracking health goals. These question oriented them towards the rest of the study activities. Ideation We set up sticky notes with statements from the entry survey the participants rated 3 or higher (mapping the Likert scale to a range of 1 to 5, with 1 = not important and 5 = very important), and were deemed to be important values and potential design features for the participant in the context of a health app. We then asked the participants to brainstorm and ideate ways to operationalize and incorporate these features into their preferred health app. Participants ideated across multiple sticky notes and were also probed to consider how they visualize these ideas into the existing health apps. Examples from P10âs ideation session can be found in Fig. 2. Generative Co-Design The participantsâ ideas were then brought to life in the generative co-design section, where they used Figma Makeâa prompt based AI toolâto generate the interface design. We provided the participants with login credentials for a Figma Make account that they could use for the duration of the study, to ensure that all costs of participation were borne by the research team. Once logged in, they could start with the app UI that they were most familiar with: Apple Health or Google Health. We shared a screenshot of the sticky notes from the ideation section in the chat to allow them to refer back to their ideas along with a short prompting guide that outlined questions to consider while prompting to the AI. We adopted a think-aloud protocol for the co-design section to understand their prompting process. Participants also had the ability to reiterate on the generated designs and were asked questions about their design decisions and asked to evaluate the generated designs. Examples from P10âs generative co-design are presented in Fig. 2. Evaluation The last section of the session focused on a more holistic evaluation of their experience with generative co-design. The participants answered questions about things they would change in the generated UI design, features and designs that came closest to their vision, their thoughts about using AI to co-design familiar interfaces, and their evaluation of individual sections of the study and the overall study design. Lastly, the participants filled out a post completion survey with 6 statements about how generative AI technologies aligned with their values around design, ranked on a Likert scale ranging from strongly disagree to strongly agree. Figure 2. P10âs sticky note ideation and generative co-design outputA composite figure shows participant P10's ideation sticky notes on the left connecting via an arrow to two generated app screens on the right. The sticky notes group nine needs around trusting and understanding AI-driven health insights, including wanting numbers with context, connecting metrics to each other, transparency, and considering external factors like weather or schedule changes. The two app screens show the generative co-design output: a trends screen with a correlation graph and correlation strength indicators, and a summary screen with a sick-day banner that adjusts fitness goals, each generated from a corresponding AI prompt shown below it. 3.3. Data Analysis Following the co-design sessions, all recordings were auto-transcribed by Zoom, and two authors verified all transcripts for accuracy. The transcripts were then anonymized, and the complete datasetâincluding the generated designs, FigJam activity notes, and entry and exit survey responsesâwas used as corpus for analysis. We analyzed the interview data using thematic analysis (Clarke et al., 2015), starting with open coding the transcripts, and iteratively refining them to surface themes that encapsulated insights that could answer our research questions. Two authors led the initial coding, and the rest of the team provided feedback and questions during co-working sessions that informed future rounds of coding. We similarly coded design outputs from the co-design sessions, both identifying emerging themes on design choices and prompting approaches as well as using the generated designs to contextualize participantsâ responses in the interviews. We calculated descriptive statistics from the questionnaires and summarize them in Table 3 and Table 4. 4. Findings 4.1. RQ1: Reimagined Designs of Familiar Personal Health Interfaces Table 2. Summary of what each participant prompted for during the Figma Make co-design session, with high-level tags.Table summarizing what each of the 14 participants (P1 through P14) prompted for during the Figma Make co-design session, listing a prompt summary and corresponding high-level tags for each participant. Prompts commonly requested AI-driven insights or explanations, data visualization improvements, and goal or schedule tracking features, with several participants also requesting menstrual cycle tracking, chatbots, or gamification elements. P# Prompt Summary Tags P1 AI insights layer over Apple Health: metric-relationship explanations, expandable detail toggle, âAsk Health AIâ chatbot, cross-metric trend analysis, personalized recommendations, activity map, AI workout form-coach, calendar with monthly highlights â kept minimal/Apple-consistent. Summary Insights, Calendar view, AI workout coach, AI Chatbot, Metric Explanation â Why, Data Visualization P2 Weekly-schedule page driving goals/tracking; notifications that ask for reasoning on abnormal metrics; more visual (less text) metric explanations; more readable/drillable charts; pattern-discovery section; simplified activity ring. Summary Insights, Reasoning (external factors), Goal tracking, Weekly Schedule, Insights Page, Metric Explanation â How, Data visualization P3 Color-coded, status-driven cards; consolidated overall-health card (donut + %, expandable); 2-per-row grid; spacing/layout fixes; color-coded cycle countdown replacing the existing indicator; header recolor. New Metric â Health score, Color coding, Data Visualization, Goal tracking, UI Improvement P4 Goal-delta indicators (red/green, above/below target) for sleep/calories/steps; step line graph with proper time grouping; dynamic step goal; new sleep page with stage breakdown; both added to home widgets. Color coding, Data Visualization, Sleep tracking by type, Goal tracking P5 New âfree timeâ settings page (separate, to avoid confusion); notification tied to free time; open-ended self-disclosure feedback field. Free-time notification, Onboarding experience â enter info P6 Gym workout tracker â log exercises, weight, reps, save full workout splits. Workout tracker P7 Reorder pinned cards by data availability/goal progress; clearer cycle-tracking visualization (replace progress bar); consolidate redundant data views into trends/highlights; info icons with explanations; chatbot built into nav bar. Goal tracking, Data visualization, AI Chatbot, Menstrual cycle tracking, UI Improvements, Metric Explanation P8 Soft pink background; menstrual-phase display + PMS symptom tracker (review, not log); symptom-based suggestions merged into one checklist; 10,000-step goal; gamified step competition with friends. Menstrual cycle tracking, PMS Suggestions â to do list, Goal tracking, Data visualization, Step count leaderboard, UI Improvements P9 AI chat with full profile context to explain metrics; search moved to top; customizable/reorderable layout; goal-forecasting tool; reorganized bottom nav (Summary, Goals & Projections, More). UI Improvements, AI Chatbot, Goals and Forecasting, Custom layout, Data visualization P10 Cross-metric correlation line graph (metrics vs. time); schedule-exception feature (vacation/sick) adjusting goals and insights. Data Visualization, UI Improvements, External factors P11 Dark theme; âgood morningâ home dashboard; profile with streak count; large graphical summary of steps/calories/workouts/sleep. UI Improvements, Gamification, Data Visualization, Summary card, Goal tracking P12 Weekly (not just same-day) forecast; notes for health events (blood tests, supplement dates); long-term goal log with derived short-term goals; layout rearrangement (drop Sleep Score, promote forecast). Goals and Forecasting, Goal Tracking, Medication tracker, Blood reports archive P13 Personal weekly-schedule section; insights linking schedule to activity/health; weather integration showing effect on daily activity. Weather forecast, Weekly/Daily Schedule, Weather and health data correlation P14 Age/physique-based baseline metrics spectrum; predictive/diagnostic AI combining metrics (e.g., temp + heart rate) with confirm-before-advice notification flow. Health Baseline, Illness Onset, AI Recommendations, AI based Prediction and Diagnosis Participants created interface features spanning 26 themes across the 14 sessions, as summarized in Table 2. Some of the major occurring themes include data visualization (9 participants), goal tracking (7 participants), and general interface improvements (6 participants). In addition, other key design themes consisted of summary insights, metric explanations, conversational features, gamification, and menstrual-cycle tracking. Although participants began with familiar personal health applications, their designs extended these interfaces beyond presenting individual health metrics. We organize the resulting designs around three directions: 1) connecting health data to provide greater context, 2) extending tracking toward future planning and action, and 3) introducing more personal forms of visualization and interaction. 4.1.1. Connecting and Contextualizing Health Data Participants frequently redesigned how personal health data could be organized and interpreted. For instance, several participants introduced new data visualizations, differing in how much information they combined and how they represented relationships among metrics (for example, Fig. 8). P3 created a âComposite Health Scoreâ that summarized multiple health metrics against predefined, personalized thresholds. P4 retained separate metrics and used color-coded visualizations to communicate trends and progress toward their personal goals. P10 created a multi-metric trend overlay line graph along with a metric correlation strength bar reading âSteps to Sleep, Moderateâ correlation. Figure 3. An example of summary card imagined by P1 (top left), P2 (bottom left), P11(right) which shows the importance of visualizing health data within context.Three participant-imagined summary card designs shown together. Top left, from P1: a light-mode "Today's Insight" card with a narrative insight stating the user sleeps 18\% longer on days with 7,000+ steps and that last night's 7h 4m of sleep followed their most active day this week, alongside stat chips for Steps (8,247), Sleep (7h 4m), and Activity (84\%), and an "Ask Health AI" link. Bottom left, from P2: an "Unusual Activity Detected" alert noting today's steps (196) are 93\% below the typical Sunday average (2,994), with quick-reply options: Feeling sick, Rest day, Traveling, Busy day, and Other. Right, from P11: a dark-mode "Today at a Glance" dashboard showing a steps ring (196 steps, 2.0\% of a 10,000-step goal) with an hourly activity bar chart, plus calorie (2,464 kcal total), workout (no workout logged yet), and sleep (7h 24m, with deep, REM, and light stage breakdown) sections. Other designs connected health measurements to events and circumstances in participantsâ lives. For instance, P1 created an insight card explaining that the user slept â18% longer on days with 7,000+ steps,â while displaying the sleep and activity values used to generate that observation. P13 designed a visualization connecting step counts with weather conditions, allowing changes in physical activity to be viewed alongside any external factors that may have influenced them. P14âs interface included a health baseline, recovery deficit, and unusual-activity detection, and asked users whether an unexpected change could be explained by being sick, traveling, resting, or having a busy day. As shown in Fig. 3, these designs treated health data as something that required context and interpretation across different kinds of metrics rather than as a collection of self-explanatory measurements. 4.1.2. Making Health Tracking Forward-Looking and Actionable Figure 4. Forecasting concepts produced during co-design, including an actual-versus-projected calorie goal display (P9 - left) and a weekly step forecast based on prior activity patterns (P12 - right).Two forecasting concepts from co-design. Left: an Active Calories card tracks actual calorie burn against a 400 cal/day goal, with a line chart showing burn fluctuating below the goal line and dropping sharply toward the end. Right: a Weekly Step Forecast card predicts next week will be lighter than this week, estimating about 5,434 steps per day versus this week's average of 6,067, shown in a bar chart comparing daily counts across both weeks. Participants also designed interfaces that connected past measurements with future goals and possible actions. Goal tracking appeared across seven participantsâ designs, making it the second most common feature theme. For example, P9 created a âGoals and Projectionsâ page that compared actual and projected progress, showing how current activity could affect longer-term goals. Another design forecasted the userâs step count for the following week based on patterns from the previous four weeks and presented expected values for each day. As shown in Fig. 4, these interfaces enable users to anticipate their likely progress rather than waiting until a goal had already been met or missed. Other designs incorporated recommendations and planning directly into the health interface. P5 designed notifications that suggested activities during available periods in the userâs daily schedule, and P6 created a workout tracker intended to support exercise planning. Participants also proposed weekly schedules, adaptive workout plans, medication tracking, menstrual-health suggestions, and recommendations responsive to changes in health data. Although these features addressed different aspects of health management, they shared an emphasis on helping users in decision making with personalized and contextualized recommendations. Through these designs, personal health interfaces transformed into tools for planning and action rather than only records of previous behavior. 4.1.3. Personalizing Visual and Interactive Experiences Figure 5. Menstrual cycle redesign from the study against the original card design (top left) - left P7; Right P8Two Menstrual Cycle feature redesigned from the study. Top Left: Original Menstrual Flow and cycle tracking feature visual Bottom Left: P7's redesign, text indicating what day of the cycle user is on, Phase card indicating which phase of the cycle users on with a calendar Right: P8's redesign, Detailed cycle tracking card highlighting no. of days and symptoms like fatigue, mood swings and bloating We found that our participants further re-imagined how people could interact with personal health information. Several participants made broader interface changes, including new layouts, color schemes, navigation structures, and dark-mode designs. These changes were often connected to how participants wanted particular types of health information to be experienced rather than to visual appearance alone. For example, P7 and P8 created different interfaces for menstrual-cycle tracking (see Fig. 5). P7 moved from the familiar circular indicator toward a phase-based display that showed the current phase, the number of days remaining, and its position within a calendar. P8 created a consolidated cycle summary that combined the current phase with predicted symptoms, bodily changes, and suggested actions. Importantly, both participants expanded menstrual tracking beyond recording the date of a cycle, but used different visual structures to communicate phase-specific information. Figure 6. Gamification concepts produced during co-design, including a social step leaderboard (P8 - left) and a personal challenge streak with weekly progress (P11 - right).Two gamification concepts from co-design. Left: a Step Challenge leaderboard toward a 10,000-step daily goal, ranking Sarah (6,843), Jordan (4,210), and Alex (2,994) ahead of the user, who is last with 196 steps and a prompt noting they need 2,798 more steps to pass Alex for third. Right: a personal dashboard greeting "Alex" with a 23-day challenge streak marked as a personal best, and a weekly progress row showing completed days highlighted in purple, missed days in gray, and today (Sunday) highlighted in orange. Conversational and gamified features provided additional ways of engaging with health data. Three participants incorporated AI chatbots through which users could ask questions about their health information or request explanations and recommendations. Other participants created challenges, streaks, leaderboards, and comparisons with friends. For example, P8âs step leaderboard showed the userâs position among friends and indicated how many additional steps were required to move into third place. Another design combined a personal-best indicator with a challenge streak and weekly progress, as illustrated in Fig. 6. These designs positioned health tracking as an interactive process involving inquiry, feedback, and motivation rather than passive viewing. We also observed that across these design directions, similar feature priorities did not necessarily result in similar interfaces. Participants who requested data visualization produced composite scores, separate color-coded metrics, forecasts, and cross-metric overlays, while P7 and P8 represented menstrual-cycle information through distinct visual metaphors. For example, Fig. 8 shows a graphic using color coding within a display of correlation metrics. It highltights the correlation between steps, calories and sleep. That said, the chatbot designs converged around a more uniform structure consisting of message bubbles, suggested questions, and a text-input field. We return to how generative AI contributed to both variation and convergence across these designs in the Discussion section (Section 5). 4.2. RQ2: Opportunities and Limitations of Generative AI-Supported Co-Design Table 3. Entry Survey: Importance Ratings of Design Statements (Scale 1â5, n = 14). Counts and bars show ratings 1â5, with 1 being not important and 5 being very important; bars are scaled to each rowâs own maximum.Table summarizing entry survey ratings on a 1-to-5 importance scale (n=14) for nine design statements, listing mean, median, and standard deviation alongside a bar visualizing how responses were distributed across the five rating levels. Being transparent with how metrics are calculated received the highest and most consistent rating (mean 4.64, lowest standard deviation, responses concentrated at 4-5), while having numbers with context received the lowest and most varied rating (mean 3.21, highest standard deviation). Most statements skew toward higher importance ratings of 4 and 5. Statement Mean Median Std. Dev. Distribution Having numbers with context (e.g., âstep count was low due toâŠâ) 3.21 3 0.97 Connecting the dots across different metrics 4.21 4 0.70 Getting suggestions or advice that feels relevant to you 3.79 4 1.12 Discovering patterns and trends within your health data 4.29 4 0.73 Trusting AI in analyzing your health data 3.57 4 1.09 Having synthesized data with the option of digging deeper 3.57 4 1.02 Feeling that the app understands you and your lifestyle 4.07 4 1.07 Considering external factors such as workload, injury, schedule, etc. 3.86 4 1.17 Being transparent with how metrics are calculated 4.64 5 0.52 Table 4. Exit Survey: Likert Item Ratings (Scale 1â5, n = 14). Counts and bars show ratings 1â5, with 1 being strongly disagree and 5 being strongly agree (Row marked â is reverse-coded: 1 is the favorable end); bars are scaled to each rowâs own maximum.Table summarizing exit survey ratings on a 1-to-5 agreement scale (n=14) for seven Likert statements about the AI co-design tool, listing mean, median, and standard deviation alongside a bar visualizing the distribution of responses across the five rating levels. Participants most strongly agreed that the tool helped them understand what their data means (mean 4.64) and what they could do differently (mean 4.57), and felt in control of the design (mean 4.29) and able to trust the tool (mean 4.43), with narrow, high-agreement distributions. Two statements framed as concerns received low agreement, indicating a favorable outcome: participants disagreed that the AI pushed them toward certain directions (mean 2.50) and, on the reverse-coded item, disagreed that outputs felt generic or template-like (mean 2.36). Statement Mean Median Std. Dev. Distribution The AI helped me express design ideas 4.21 4.5 0.80 The AI pushed me toward certain directions 2.50 3 1.16 I felt genuinely in control of the design 4.29 4 0.47 The outputs felt generic or template-likeâ 2.36 2 1.08 Using this tool helped me understand what my data means 4.64 5 0.63 Using this tool helped me understand what I could do differently 4.57 5 0.65 I would trust a tool like this to help 4.43 4.5 0.51 Participants generally evaluated their experience with generative co-design positively, as shown in Table 4. They agreed that AI helped them express their design ideas (M=4.21M=4.21) and that they remained in control of the resulting designs (M=4.29M=4.29). Participants also reported that the process helped them understand what their health data meant (M=4.64M=4.64) and what they could do differently (M=4.57M=4.57). Conversely, they were less likely to agree that AI pushed them toward particular design directions (M=2.50M=2.50) or that the outputs appeared generic or template-like (M=2.36M=2.36). However, participantsâ accounts revealed that these experiences depended on how their ideas were scaffolded, how the AI interpreted their prompts, and how easily they could revise the generated output. We examine these opportunities and limitations across five aspects of the co-design process. 4.2.1. Scaffolding the Translation from Ideas to Prompts The generative co-design component of our study did not begin directly with prompting. Our participants first identified design priorities through the entry survey and expanded those priorities during the guided ideation activity. The facilitator recorded their ideas on sticky notes and shared an image of those notes when participants began using Figma Make. Several participants found that this scaffolding helped them organize what they wanted to create. P7 explained that the sticky notes âkind of kept me on track,â and P12 said that they âhelped me formulate what I wanted.â Most of the features participants requested from Figma Make could be traced to this earlier ideation. For example, P10âs idea for viewing correlations among multiple health metrics became a request for a multi-metric trend visualization, while P1âs observation about being more active after sleeping for seven or more hours became an example within an insight card. As revealed in Fig. 2, the prompts therefore often served to translate ideas developed during the guided activity into interface features. At the same time, the scaffolding influenced the range of ideas participants considered. P1 described the survey categories and sticky notes as a âbubbleâ that encouraged deeper thinking within the areas presented rather than exploration beyond them. Moreover, not every idea recorded during ideation was included in a prompt. P6 discussed badges, metric explanations, AI-quality indicators, a workout tracker, and adaptive workout planning, but ultimately prompted only for the workout tracker. P13 similarly narrowed a broader interest in relationships among health metrics to a visualization connecting steps and weather, explaining that âadding another prompt would mean taking extra time to build something.â The interfaces participants generated therefore reflected a selective translation of their earlier ideas, shaped partly by the effort and time required to formulate and submit additional prompts. 4.2.2. Translating Loosely Articulated Ideas into Concrete Prototypes Participants used different prompting strategies to translate their ideas into interfaces. P1, P2, and P9 consolidated several requested features into a single detailed prompt, whereas most participants prompted for one feature at a time and revised the interface incrementally. P1 used a second AI tool to help organize their prompt, while other participants included screenshots or online examples to communicate visual references. These strategies allowed participants to provide different combinations of textual, structural, and visual direction. Participants particularly valued how the co-design process helped to interpret ideas that they did not consider fully formed or clearly expressed. P12 stated, âI like how the AI can make sense of what I meant to tell it, because I feel like my prompt was kind of messy.â P8 similarly described their prompt as ânot very coherent,â but found that the generated interface still captured what they intended. Reflecting on this experience, P8 explained that âthere were times where I definitely couldnât have verbalized what I meant, but this was able to catch it and do the job for me.â More detailed prompts could also produce complex features that closely matched participantsâ expectations. P5 requested notifications based on free periods in the userâs daily schedule and found that the resulting interface matched âwhat I was thinking.â P14 similarly stated that Figma Make created âeverything I asked for and moreâ when generating a health-intelligence interface. Participants therefore described the tool as a way to convert an internal idea into something they could see, evaluate, and revise. As P14 summarized, âit was able to bring my vision to life without me having to do a lot.â For these participants, generative AI served as a translation layer between an idea and an interface prototype without requiring them to manually implement the design. 4.2.3. Generative Suggestions Expanded Participantsâ Ideas The generated interfaces did not always remain within the boundaries of participantsâ prompts. In several cases, Figma Make introduced features or design details that participants had not specified but subsequently chose to retain. When P7 rejected the initial progress-bar visualization for menstrual-cycle tracking, the tool proposed four alternative directions. The resulting design introduced phase tracking, which P7 said they âwouldnât have actually thought of,â leading them to conclude that âFigma Make is pretty good with ideation.â Other participants responded similarly to unrequested additions. P14âs interface included recovery-deficit and unusual-activity sections that they described as âvery niceâ and âa lot moreâ than expected. When P8 requested a step leaderboard, the generated interface added a message showing how many more steps were needed to move into third place, which P8 viewed as useful additional motivation. In these cases, participants treated the generated additions as extensions of their ideas. Therefore, generative AI supported ideation not only by rendering requested features, but also by presenting possibilities that participants could evaluate and incorporate into their designs. 4.2.4. Perceived Control Depended on Alignment with Participantsâ Intentions Although participants generally reported feeling in control, their accounts indicated that control depended on whether the AIâs decisions aligned with what they intended. Participants did not necessarily experience an unrequested addition as a loss of control when they considered that addition relevant or useful. However, the same autonomy became a source of friction when Figma Make made assumptions that participants did not endorse. P10 encountered this problem when the AI added a âmove goal multiplierâ while implementing a feature responsive to life events. Rather than having the system select an adjustment on its own, P10 wanted it to ask, âHow would you like the move goal to be adjusted?â P10 generalized this concern by stating that the AI âshould ask me before it does so many changes at once.â P1 similarly observed that an underspecified prompt allowed the system to determine unresolved design details: âif I want a certain thing to look in a certain way [..] otherwise itâl be open-ended for it, and itâl make it how it seems it wants to do it.â Figure 7. Convergent chatbot designs (P1 - left, P7 - center, P9 - right)Three chatbot design concepts for a Health AI assistant. Left, a blue-themed chat opens with an introduction offering to analyze patterns across sleep, activity, heart rate, HRV, and nutrition, followed by suggested prompts like "Why have I been more tired this week?" and "How's my recovery looking?" Middle, a similar light-themed chat shows an example exchange where the user asks "Why are my steps so low today?" and the AI responds that they've logged 196 steps versus a 10,000 goal and daily average of about 2,994, noting a 20-minute walk typically adds about 2,000 steps, with more suggested prompts below. Right, a dark-themed "AI Health Assistant" panel opens by stating it has full context of the user's health profile, activity history, and goals from six months of data, with quick-access chips for step count, active calories, and sleep patterns above the input field. Further, model defaults also limited participantsâ ability to realize some interaction ideas. The three chatbot designs followed a similar template consisting of a header, message bubbles, suggested questions, and a text-entry field. P7 provided a reference image and requested a chatbot that transformed the bottom navigation bar into a text-input field. However, the generated design retained the original template and changed only the navigation label, as shown in Fig. 7. P7 considered the result generic and indicated that they would have changed its branding and interaction. These cases show that participantsâ sense of control depended less on whether the AI made design decisions and more on whether those decisions remained consistent with their intentions. 4.2.5. Generation Latency Constrained Iteration Generation latency further shaped how participants engaged in co-design. Producing or revising an interface took approximately two to ten minutes across the sessions. P10 described three minutes for a design change as âquite a lotâ and noted that waiting beyond five minutes would become increasingly frustrating. Although some participants recognized that interactive interfaces required more time to generate than text, the delays still affected how they structured subsequent prompts. Several participants began combining changes to reduce the number of generation cycles. P10 decided to include multiple changes in the next prompt âso that it only uses one prompt,â while P9 organized several ideas into a single structured request rather than iterating feature by feature. P12 wondered whether the length of their prompt had caused the delay, and P4 asked whether they were âsupposed to one-shot thisâ or could continue updating the design. These responses indicate that participants had to determine an appropriate prompting cadence during the activity rather than finding it immediately apparent. Latency therefore affected more than participantsâ satisfaction with the tool. Because each revision introduced another period of waiting, participants became more selective about which ideas to pursue and increasingly consolidated changes into fewer prompts. This encouraged more front-loaded prompting and reduced the open-ended iteration that generative co-design might otherwise support. 4.2.6. Supported and Unsupported Design Values Finally, there was a marked difference in what design values the participants were able and not able to operationalize during this co-design activity. Across both activities, participants consistently highlighted data interpretabiltiy as a key design value, ranking it #1 most often in their exit survey (as summarized in Table 5). Departing from existing designs that presented different metrics in isolation, several participants sought to design interfaces that allowed for contextualizing data within other metrics. The participants who chose interpretability as their primary goal prompted for features such as insight cards explaining relationships between metrics, menstrual phase and symptom tracking in connection to physical activity changes, one participantâs (P9) request for an AI that âexplains you all the metrics that youâre not aware of,â or a diagnostic AI that âconnects the dots across different health metrics.â (P14) Designing for accountabilityâestablishing goals and tracking progress towards themâsimilarly received sufficient attention in the design outputs. Conversely, privacy was ranked highly by the participants but none of the participants prompted for a privacy related design. This maybe because privacy is a less visually apparent element and manifests itself behind the scenes through account-level settings, sharing permissions, and data-retention rules that may sit outside of the main interface design. This suggests that generative co-design may favor visually expressible values while leaving more abstract values like privacy, trust, and emotional safety outside its scope. Table 5. Borda-count ranking of six design values (n=14n=14). Distribution shows how many participants ranked each value 1st through 6th (most to least important); bars are scaled to each rowâs own maximum.Table showing a Borda-count ranking of six design values based on how 14 participants ranked them from most to least important, with each value's total score (out of a maximum of 84), mean rank, and a bar visualizing how many participants placed it in each rank position (1st through 6th). Interpretability ranked highest overall (score 61, mean rank 2.64), followed by Accountability (57), Privacy (53), Autonomy (51), and Trust (44), with their distributions spread across the middle rank positions. Emotional safety ranked clearly lowest (score 28, mean rank 5.00), with 8 of 14 participants ranking it last. Value Borda score (max 84) Mean rank Distribution Interpretability 61 2.64 Accountability 57 2.93 Privacy 53 3.21 Autonomy 51 3.36 Trust 44 3.86 Emotional safety 28 5.00 Together, these findings show that generative AI made it easier for participants to materialize loosely articulated ideas and explore design possibilities beyond those they initially envisioned. However, the same process also influenced which ideas reached the prototype, how unresolved choices were interpreted, and how freely participants could iterate. The opportunities and limitations of generative co-design therefore emerged from the interaction among the studyâs scaffolding, participantsâ prompting strategies, the AIâs design decisions, what design values it could more easily design for and what it left to the user, and the time required to generate each revision. 5. Discussion This study essentially asks what design patterns emerge when people use generative AI to re-imagine familiar personal health interfaces, and what it affords as a co-design method. In response to RQ1, our findings reveal both convergence and substantial variation in the design outputs that participants produced. Some interfaces, like chatbots, appeared notably stable across participants, while others highlighted data interpretation possibilities that remain poorly supported by existing personal health informatics systems. In response to RQ2, we found that generative co-design enabled participants with varied technical and design backgrounds to rapidly materialize and refine ideas, while simultaneously introducing a new source of constraint: the generated artifact itself. We discuss these findings first as evidence about where personal health interface design has converged and where the design space remains open, then as a methodological tension introduced when generative AI mediates between participant intent and the resulting artifact, and finally as a set of design implications for platforms that mediate personal health data at scale. 5.1. Convergence and Divergence in Generative Co-Design of Personal Health Interfaces The interfaces our participants produced revealed a notable asymmetry between conversational and visualization-based designs. When participants generated conversational interfaces, their designs converged: despite beginning with different experiences and goals, participants frequently arrived at recognizable chatbot forms as demonstrated in Fig. 7. One plausible interpretation is that conversational AI has already stabilized as an interface genre. Chat-based interfaces might take on the designs we commonly associate with chatbots across instant messaging, social media, and generative AI technologies. At the same time, because participants designed these interfaces through a generative system, we cannot attribute this convergence solely to participantsâ existing mental models. The similarities may insteadâor additionallyâreflect conventions encoded in the generative system itself. Thus, convergence through generative co-design can tell us something about where interface conventions have become well established, while also demonstrating how generative systems may reproduce these conventions and constrain the range of alternatives that participants encounter. Participantsâ visualization-based designs followed a different pattern. When participants attempted to relate data streams that their existing tools presented separately, the resulting interfaces varied considerably in their visual form, temporal framing, and the data relationships they mapped. Participants imagined, for example, interfaces that placed weather conditions alongside step counts and weight, or menstrual-cycle stage alongside perceived energy, walking activity, or strength-training performance, and sleep with heart rate, heart rate variability, or steps. In these designs, what is important to note is that the underlying measurements were already available to participants. The goal of their generated designs was representations through which one measurement could become context for interpreting another. This distinction connects to a longstanding challenge in personal informatics (PI). PI systems have traditionally organized support around processes of collection, integration, reflection, and action (Li et al., 2010), while subsequent work has repeatedly shown that making data available does not necessarily make those data interpretable (Choe et al., 2014; Rapp and Tirassa, 2017). Researchers have therefore explored ways of providing context for reflection, including comparison with cohort data (Feustel et al., 2018), flexible and user-defined tracking (Kim et al., 2017; Ayobi et al., 2018), and visualizations intended to facilitate reflection across personal data. More recently, Moore et al. characterize an explicit âpersonal informatics analysis gapâ: despite extensive work on collecting personal data and generating insights from it, personal informatics systems provide comparatively limited support for the flexible analyses through which people might explore questions of their own. Our findings suggest that this gap is not only relevant insofar as better analytical capabilities, but also representational: participants were not merely asking systems to calculate relationships among variables; they were imagining different ways of visualizing those relationships. This distinction is important because a composite visualization necessarily makes choices about interpretation, as extensive visualization research in HCI has found previously (Dörk et al., 2013). Showing menstrual-cycle stage alongside exercise performance and sleep, for example, requires deciding how to align these data temporally, which patterns to surface and emphasize, and whether to simply show how they vary together or offer an initial interpretation of why these patterns might co-occur. Prior visualization work in personal informatics has similarly argued that the design of visual representations shapes opportunities for reflection rather than simply communicating measurements (Cuttone et al., 2014). Likewise, work on contextualizing personal data demonstrates that adding context changes what comparisons people make and how they reason about their own experiences (Feustel et al., 2018). The diversity of representations generated by our participants suggests that these questions continue to remain unanswered and arguably might never fully be answered: the appropriate representation may depend on what relationship a person is attempting to understand. Taken together, we posit that the convergence and divergence in our participantsâ designs is evidence of where conventions in personal health interface design may have matured within HCI. Conversational interfaces showed considerable convergence around an already familiar interaction paradigm. In contrast, participantsâ composite visualizations diverged in both the relationships they constructed and how they chose to represent them. Prior work in personal informatics has demonstrated the value of flexible tracking, contextualization, and analysis (Feustel et al., 2018; Kim et al., 2017; Moore et al., 2022). Our findings extend this work, showing the diversity of relationships that people may wish to construct when given the opportunity to redesign the interfaces through which they encounter their data. This diversity makes it difficult to anticipate in advance which combinations or representations will be meaningful to a particular person. Instead of seeking a universal visualization for integrating personal health data, personal informatics systems can provide greater flexibility for people to construct and explore relationships that matter to them. 5.2. Generative Co-Design as a Research Method The generative co-design process was as informative as the artifacts it produced. Participants were able to transform natural-language descriptions into functional, clickable interfaces within minutes, with relatively little technical overhead. Participatory and co-design traditions have long sought to give people a substantive role in imagining technologies intended for them (Poot et al., 2023; van Hierden et al., 2021), while prior methods have used prototypes and probes to help participants engage with and respond to possible technologies (Buchenau and Suri, 2000; Hutchinson et al., 2003). Yet translating a participantâs idea into an interactive prototype has typically required either design expertise from participants or mediation by researchers and designers. Generative interface tools do not fundamentally change the value of prototyping as a means of articulation; instead, they substantially compress the process through which an articulated idea can become an interactive artifact. The ability to quickly generate designs that enable articulation simultaneously comes with risks. Design research has long documented âdesign fixation,â in which exposure to an example solution causes subsequent designs to reproduce characteristics of that example even when designers are encouraged to avoid doing so (Jansson and Smith, 1991). This concern becomes particularly meaningful when the example is not introduced by a researcher but generated dynamically in response to the participantâs own incomplete description. Once an interface appears on screen, it transforms an open-ended design problem into the more bounded task of responding to an existing artifact. Recent work suggests that generative AI may intensify this problem (Wadinambiarachchi et al., 2024). Wadinambiarachchi et al. found that designers exposed to AI-generated imagery exhibited greater fixation and produced fewer, less varied, and less original ideas than participants in a baseline condition. Their findings further suggest that participants can shift fixation from an initial example toward the AI-generated material. Our observations point toward a related possibility in generative co-design. Participants often refined what the system generatedâchanging elements, requesting additions, or correcting featuresârather than discarding its underlying structure and beginning from a categorically different representation. For generative co-design, however, fixation creates a methodological problem beyond whether participants produce more or less original designs. When a generative system translates a participantâs description into an interface, it inevitably introduces design decisions that the participant did not explicitly request. If participants subsequently accept or modify those decisions, it becomes difficult to determine which aspects of the final artifact reflect participant intent and which originated from the model. Generative AI therefore does not remove the mediation between participant ideas and higher-fidelity prototypes; it relocates some of that mediation from the researcher or designer to the generative system. Artifacts produced through generative co-design should consequently be understood as products of participant intentions, existing interface conventions, model priors, and participantsâ reactions to what the model generates, rather than as direct representations of what participants want. This attribution problem has methodological implications for how generative co-design is conducted and analyzed. Researchers could generate multiple interpretations of an initial prompt before refinement, asking participants to compare what each gets right or wrong rather than immediately iterating on the first output. Researchers could also preserve prompt-generation histories to distinguish what participants explicitly requested from what the model introduced and participants subsequently accepted, rejected, or modified. Such distinctions can help make the modelâs role in shaping the artifact visible and provide a more precise account of how a design emerged. This becomes particularly important when generative co-design is used to elicit user needs: without accounting for the modelâs contribution, researchers and practitioners risk treating AI-mediated artifacts as straightforward evidence of user preference. 5.3. Design Implications for Platform-Scale Health Interfaces Organize views around usersâ questions. Health ecosystems can already aggregate data from phones, wearables, and third-party applications. Our participantsâ designs highlighted the need for interfaces through which these streams could become meaningfully related. Personal health systems could therefore let users choose which streams to view together and how to represent them around questions such as How does my sleep change on days when I exercise later? The diversity of participantsâ designs suggests that these choices should remain open to usersâ own questions and contexts. Let users and third parties create views. Platforms need not anticipate and author every meaningful comparison or representation. They could provide tools for users to create their own views, while allowing trusted third-party applications to build additional views with user permission. This would position platforms as infrastructures for user-directed interpretation, supporting ways of relating data that extend beyond predefined categories and comparisons. Make uncertainty visible. Supporting these comparisons requires distinguishing between showing a relationship and explaining it. People can struggle to interpret self-tracked data (Choe et al., 2014), while reflective-informatics research cautions against reducing complex experiences to prescriptive judgments (Baumer et al., 2014). For example, an interface could juxtapose exercise and menstrual-cycle data without asserting that cycle stage caused a particular performance outcome. Interfaces should make uncertainty and missing context visible and avoid presenting observed patterns as causal or clinical conclusions. Protect privacy and emotional safety. Combining streams can reveal sensitive information that individual metrics may not expose, including inferences about stress or smoking (Saleheen et al., 2016). Users should therefore control which data are combined and which applications can access them. More granular comparisons could also encourage people to continually search for explanations for fluctuations in their health. Prior work has emphasized that people actively construct meaning around self-tracked data (Sharon and Zandbergen, 2017). Interfaces should support this interpretive process while leaving users with authority over how, and whether, patterns become meaningful to them. 5.4. Limitations and Future Directions This work provides an initial exploration of generative co-design and highlights areas for future consideration. First, the one-hour study sessions limited how many ideas participants could translate into prompts and how many cycles of iterative prompting they could pursue, amidst the technical latency constraints of dynamically prompting and receiving an output. The UI design outputs reflected the time constraint rather than the ceiling of what generative co-design can produce. The design outcomes varied with participantâs individual effort and prompting approaches. Future research can adopt longer or repeated sessions to examine how participants explore and refine designs over time, and how prompting experience and support shape this process. Also, participants used a single GenUI tool, Figma Make, whose interaction model and generation latency shaped prompting behavior and may not generalize to other GenUI systems. Future work can compare GenUI tools to examine how different prompting and editing capabilities affect participantsâ exploration, control, and expression of values. In particular, the rapid development and proliferation of these tools further motivate comparisons of design outcomes and participant experiences across systems. Further, since generative co-design introduces model-driven design decisions, it can be difficult to fully attribute elements of the final designs to participant intent versus the modelâs own interpretation of the participantâs prompt. Future work could combine histories of prompts, generated outputs, and revisions with participantsâ explanations of specific design choices. This could help trace how participants accept, revise, or reject model suggestions, and how their intentions evolve through these interactions. Further, incorporating participantsâ own health data (in a consented and ethically sensitive manner) into these interfaces could further reveal how they interpret and respond to its presentation, and whether these experiences change the features and values they prioritize. Broadly, our study captures how people imagine and prototype personal interfaces using generative AI and does not uncover how development and feasibility would unfold in practice. Future work should examine these processes through implementation and use. Longitudinal deployment would provide insight into how people build, use, and revise generative co-designed systems. Finally, our participant pool mainly consisted of college students and young adults, and our study suffers from representativeness and self-selection bias. Future work can engage more diverse populations with varying levels of digital familiarity, literacy, and accessibility needs to examine whether generative co-design as a method is an equitable approach to ideation and prototyping for a diverse group of people. 6. Conclusion In this study, we explored generative co-design with 14 participants who reimagined familiar personal health interfaces using Figma Make. Participants envisioned interfaces that connected health metrics with everyday contexts, supported planning and action, and introduced new metrics and personalized visualizations and interactions. Generative AI helped participants develop loosely articulated ideas and explore new possibilities, while session scaffolding, model assumptions and defaults, and generation latency constrained exploration and revision. We also found a gap between participantsâ stated values and their expression in interfaces: interpretability and accountability were more readily represented than privacy, trust, and emotional safety. These findings showed how generative AI shaped both participation and design outcomes, highlighting the need to preserve participantsâ agency over how their ideas and values were interpreted and represented. They also pointed to further research through longitudinal deployments with diverse populations and GenUI systems. References Ayobi et al. (2018) A. Ayobi, T. Sonne, P. Marshall, and A. L. Cox Flexible and mindful self-tracking: design implications from paper bullet journals. In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems, CHI â18, New York, NY, USA, p. 1â14. External Links: ISBN 9781450356206, Link, Document Cited by: §2.1, §5.1. Baseman et al. (2025) C. M. Baseman, S. M. Dembure, N. Swinger, M. T. Bondu, B. DiSalvo, and R. I. Arriaga Clinical standards and proximate futures: participatory design futuring of diabetes technologies with an under-resourced community. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 9 (4), p. 1â30. Cited by: §2.2. Baumer (2015) E. P. S. Baumer Reflective informatics: conceptual dimensions for designing technologies of reflection. In Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems, CHI â15, New York, NY, USA, p. 585â594. External Links: Document Cited by: §1, §2.1. Baumer et al. (2014) E. P.S. Baumer, V. Khovanskaya, M. Matthews, L. Reynolds, V. Schwanda Sosik, and G. Gay Reviewing reflection: on the use of reflection in interactive system design. In Proceedings of the 2014 Conference on Designing Interactive Systems, DIS â14, New York, NY, USA, p. 93â102. External Links: ISBN 9781450329026, Link, Document Cited by: §2.1, §5.3. Bentvelzen et al. (2022) M. Bentvelzen, P. W. WoĆșniak, P. S. Herbes, E. Stefanidi, and J. Niess Revisiting reflection in hci: four design resources for technologies that support reflection. Proceedings of the ACM on interactive, mobile, wearable and ubiquitous technologies 6 (1), p. 1â27. Cited by: §1, §2.1. Bhat et al. (2026) K. S. Bhat, J. M. Shi, W. Song, D. W. Yoo, and K. Saha "In my defense, only three hours on instagram": designing toward digital self-awareness and wellbeing. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems, CHI â26, New York, NY, USA. External Links: ISBN 9798400722783, Link, Document Cited by: §1, §2.1, §3.1. Bhat and Kumar (2020) K. S. Bhat and N. Kumar Sociocultural dimensions of tracking health and taking care. Proc. ACM Hum.-Comput. Interact. 4 (CSCW2). External Links: Link, Document Cited by: §2.1. Bhattacharjee and et al. (2026) A. Bhattacharjee and et al. Generative experiences for digital mental health interventions: evidence from a randomized study. External Links: 2604.07558 Cited by: §2.3. Borning and Muller (2012) A. Borning and M. Muller Next steps for value sensitive design. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, CHI â12, New York, NY, USA, p. 1125â1134. External Links: ISBN 9781450310154, Link, Document Cited by: §2.2. Buchenau and Suri (2000) M. Buchenau and J. F. Suri Experience prototyping. In Proceedings of the 3rd Conference on Designing Interactive Systems: Processes, Practices, Methods, and Techniques, New York, NY, USA, p. 424â433. External Links: Document Cited by: §2.2, §5.2. Cao et al. (2025) Y. Cao, P. Jiang, and H. Xia Generative and malleable user interfaces with generative and evolving task-driven data model. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, CHI â25, New York, NY, USA. External Links: ISBN 9798400713941, Link, Document Cited by: §2.3. Cecchinato et al. (2019) M. E. Cecchinato, A. L. Cox, J. Vincent, C. Elsden, c. Mastrodonato, A. Skatova, W. Moncur, D. Harrison, and J. Bird Designing for digital wellbeing: a research & practice agenda. In Extended Abstracts of the 2019 CHI Conference on Human Factors in Computing Systems, CHI EA â19, New York, NY, USA, p. W17:1âW17:8. External Links: Document Cited by: §1. Chen et al. (2025) X. â. Chen, T. Knearem, and Y. Li The genui study: exploring the design of generative ui tools to support ux practitioners and beyond. In Proceedings of the 2025 ACM Designing Interactive Systems Conference, DIS â25, New York, NY, USA, p. 1179â1196. External Links: ISBN 9798400714856, Link, Document Cited by: §2.3. Chen et al. (2026) X. â. Chen, S. D. Petridis, T. Deng, H. Bari, R. Du, and Y. Li Rethinking the ui of genui: a tale of two designs. External Links: 2606.13843, Link Cited by: §2.3. Choe et al. (2014) E. K. Choe, N. B. Lee, B. Lee, W. Pratt, and J. A. Kientz Understanding quantified-selfersâ practices in collecting and exploring personal data. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, CHI â14, New York, NY, USA, p. 1143â1152. External Links: Document Cited by: §1, §2.1, §2.1, §5.1, §5.3. Choi et al. (2026) D. Choi, K. Son, J. Yu, H. Jung, and J. Kim IdeaBlocks: expressing and reusing divergent intents for graphic design exploration using generative AI. In Proceedings of the 2026 Designing Interactive Systems Conference, DIS â26, New York, NY, USA, p. 621â642. External Links: Document Cited by: §2.3. Clarke et al. (2015) V. Clarke, V. Braun, and N. Hayfield Thematic analysis. Qualitative psychology: A practical guide to research methods 222 (2015), p. 248. Cited by: §3.3. Consolvo et al. (2006) S. Consolvo, K. Everitt, I. Smith, and J. A. Landay Design requirements for technologies that encourage physical activity. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, CHI â06, New York, NY, USA, p. 457â466. External Links: Document Cited by: §1. Cummings et al. (2022) M. Cummings, J. Bradley, and G. Teal Patient co-design of digital health storytelling tools for multimorbidity: a phenomenological study. Health Expectations 25, p. 3073â3084. External Links: Document Cited by: §2.2. Cuttone et al. (2014) A. Cuttone, M. K. Petersen, and J. E. Larsen Four data visualization heuristics to facilitate reflection in personal informatics. In Universal Access in Human-Computer Interaction: Universal Access to Information and Knowledge, Lecture Notes in Computer Science, Vol. 8514, Cham, Switzerland, p. 541â552. External Links: Document Cited by: §5.1. Dantas et al. (2026) C. Dantas, M. Cabrita, M. Bobowicz, H. Op den Akker, X. Rafael-Palou, T. Hakkarainen, I. Haavisto, F. Strand, L. Marti-Bonmati, E. Divjak, G. Ivanac, S. Joshi, R. Osuala, S. Charalambous, A. Tsirikoglou, G. Ribas, O. DĂaz, A. S. Carvalho, and E. Costa Integrating participatory social innovation into requirements engineering for ai health care solutions: case study. Journal of Medical Internet Research 28, p. e87984. External Links: Document Cited by: §2.2. Dörk et al. (2013) M. Dörk, P. Feng, C. Collins, and S. Carpendale Critical infovis: exploring the politics of visualization. In CHIâ13 Extended Abstracts on Human Factors in Computing Systems, p. 2189â2198. Cited by: §5.1. Duffy et al. (2025) A. Duffy, N. Boroumandzad, A. L. Sherman, G. Christie, I. Riadi, and S. Moreno Examining challenges to co-design digital health interventions with end users: systematic review. Journal of Medical Internet Research 27, p. e50178. External Links: Document Cited by: §2.2. Epstein et al. (2015) D. A. Epstein, A. Ping, J. Fogarty, and S. A. Munson A lived informatics model of personal informatics. In Proceedings of the 2015 ACM International Joint Conference on Pervasive and Ubiquitous Computing, UbiComp â15, New York, NY, USA, p. 731â742. External Links: Document Cited by: §1, §2.1, §2.1. Feustel et al. (2018) C. Feustel, S. Aggarwal, B. Lee, and L. Wilcox People like me: designing for reflection on aggregate cohort data in personal informatics systems. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 2 (3), p. 1â21. Cited by: §5.1, §5.1, §5.1. Fleck and Fitzpatrick (2010) R. Fleck and G. Fitzpatrick Reflecting on reflection: framing a design landscape. In Proceedings of the 22nd Conference of the Computer-Human Interaction Special Interest Group of Australia on Computer-Human Interaction, OZCHI â10, New York, NY, USA, p. 216â223. External Links: Document Cited by: §1, §2.1. Friedman et al. (2013) B. Friedman, P. H. Kahn, A. Borning, and A. Huldtgren Value sensitive design and information systems. In Early Engagement and New Technologies: Opening Up the Laboratory, N. Doorn, D. Schuurbiers, I. van de Poel, and M. E. Gorman (Eds.), Philosophy of Engineering and Technology, Vol. 16, p. 55â95. External Links: Document Cited by: §1, §2.2, §2.2. Harrington et al. (2019) C. Harrington, S. Erete, and A. M. Piper Deconstructing community-based collaborative design: towards more equitable participatory design engagements. Proceedings of the ACM on human-computer interaction 3 (CSCW), p. 1â25. Cited by: §1, §2.2. Huang et al. (2026) M. Huang, V. J. Rodriguez, K. Saha, and T. August Designing beyond language: sociotechnical barriers in ai health technologies for limited english proficiency. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems, p. 1â18. Cited by: §2.2. Hutchinson et al. (2003) H. Hutchinson, W. Mackay, B. Westerlund, B. B. Bederson, A. Druin, C. Plaisant, M. Beaudouin-Lafon, S. Conversy, H. Evans, H. Hansen, et al. Technology probes: inspiring design for and with families. In Proceedings of the SIGCHI conference on Human factors in computing systems, p. 17â24. Cited by: §2.2, §5.2. Ishita et al. (2026) S. Z. Ishita, S. Kaliappan, M. Rashik, D. A. Epstein, and R. Karkar CASEbot: a conversational agent for structuring and personalizing the design of self-experiments in personal health. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems, p. 1â26. Cited by: §2.3. Jansson and Smith (1991) D. G. Jansson and S. M. Smith Design fixation. Design studies 12 (1), p. 3â11. Cited by: §5.2. Karkar et al. (2016) R. Karkar, J. Zia, R. Vilardaga, S. R. Mishra, J. Fogarty, S. A. Munson, and J. A. Kientz A framework for self-experimentation in personalized health. Journal of the American Medical Informatics Association 23 (3), p. 440â448. Cited by: §2.1. Kim et al. (2026a) E. Kim, D. Choi, B. Min, H. Yi, Y. Jiang, and J. Kim Maru: information architecture as a shared language for generating aligned and persistent user interfaces. arXiv preprint arXiv:2608.25565. Cited by: §2.3. Kim et al. (2026b) Y. Kim, K. Son, S. Kim, B. Chin, and J. Kim IntentFlow: investigating fluid dynamics of intent communication in generative ai. In Proceedings of the 2026 Designing Interactive Systems Conference, p. 3804â3837. Cited by: §2.3. Kim et al. (2016) Y. Kim, J. H. Jeon, E. K. Choe, B. Lee, K. Kim, and J. Seo TimeAware: leveraging framing effects to enhance personal productivity. In Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems, p. 272â283. Cited by: §2.1. Kim et al. (2017) Y. Kim, J. H. Jeon, B. Lee, E. K. Choe, and J. Seo OmniTrack: a flexible self-tracking approach leveraging semi-automated tracking. Proceedings of the ACM on interactive, mobile, wearable and ubiquitous technologies 1 (3), p. 1â28. Cited by: §2.1, §5.1, §5.1. Kobiella et al. (2026) C. Kobiella, L. Schneider, A. Schmidt, and N. TerzimehiÄ Co-constructed or constrained? how ai collaboration tools reshape ui design practice in a time-boxed design challenge. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems, CHI â26, New York, NY, USA. External Links: ISBN 9798400722783, Link, Document Cited by: §1, §2.3. Le Dantec et al. (2009) C. A. Le Dantec, E. S. Poole, and S. P. Wyche Values as lived experience: evolving value sensitive design in support of value discovery. In Proceedings of the SIGCHI conference on human factors in computing systems, p. 1141â1150. Cited by: §2.2. Li et al. (2010) I. Li, A. Dey, and J. Forlizzi A stage-based model of personal informatics systems. In Proceedings of the SIGCHI conference on human factors in computing systems, p. 557â566. Cited by: §1, §2.1, §2.1, §5.1. Li et al. (2024) J. Li, H. Cao, L. Lin, Y. Hou, R. Zhu, and A. El Ali User experience design professionalsâ perceptions of generative artificial intelligence. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems, CHI â24, New York, NY, USA. External Links: ISBN 9798400703300, Link, Document Cited by: §1, §2.3, §2.3. Lindgren and Kilic (2025) H. Lindgren and K. Kilic Participatory design and evaluation of digital coaching for improving healthâthe star multicomponent lifestyle intervention. Frontiers in Digital Health 7, p. 1600535. External Links: Document Cited by: §2.2. Mols et al. (2016) I. Mols, E. Van den Hoven, and B. Eggen Informing design for reflection: an overview of current everyday practices. In Proceedings of the 9th Nordic Conference on Human-Computer Interaction, p. 1â10. Cited by: §2.1. Moore et al. (2022) J. Moore, P. Goffin, J. Wiese, and M. Meyer Exploring the personal informatics analysis gap: âthereâs a lot of baconâ. IEEE Transactions on Visualization and Computer Graphics 28 (1), p. 96â106. External Links: Document Cited by: §5.1, §5.1. Morris et al. (2018) R. R. Morris, K. Kouddous, R. Kshirsagar, and S. M. Schueller Towards an artificially empathic conversational agent for mental health applications: system design and user perceptions. Journal of medical Internet research 20 (6), p. e10148. Cited by: §2.1. Moruzzi and Margarido (2024) C. Moruzzi and S. Margarido A user-centered framework for human-ai co-creativity. In Extended Abstracts of the CHI Conference on Human Factors in Computing Systems, p. 1â9. Cited by: §2.3. Oewel et al. (2024) B. Oewel, N. Azizan, P. A. Arean, and E. Agapie Technologyâs role in fostering therapist-client collaboration and engagement with goals. Proceedings of the ACM on Human-Computer Interaction 8 (CSCW2), p. 1â28. Cited by: §2.2. Pantzar and Ruckenstein (2017) M. Pantzar and M. Ruckenstein Living the metrics: self-tracking and situated objectivity. Digital health 3, p. 2055207617712590. Cited by: §2.1. Patel et al. (2015) M. S. Patel, D. A. Asch, and K. G. Volpp Wearable devices as facilitators, not drivers, of health behavior change. Jama 313 (5), p. 459â460. Cited by: §1. Piera-JimĂ©nez et al. (2026) J. Piera-JimĂ©nez, N. Vilarasau Creus, A. MaymĂł Costa, X. Michelena, A. Climent Fageda, A. FarrĂ©, L. Herczeg, L. Parameswaran, G. Carot-Sans, and L. Valle Cocreating principles for digital health equity: cross-sectional, qualitative study for participatory human-centered design in catalonia. Journal of Medical Internet Research 28, p. e84129. External Links: Document Cited by: §2.2. Piwek et al. (2016) L. Piwek, D. A. Ellis, S. Andrews, and A. Joinson The rise of consumer health wearables: promises and barriers. PLoS medicine 13 (2), p. e1001953. Cited by: §1. Poot et al. (2023) C. C. Poot, E. Meijer, A. Bruil, M. Venema, N. J. Vegt, N. Donkel, V. van Noort, N. H. Chavannes, and A. A. Roest How to use participatory design to develop an ehealth intervention to reduce preprocedural stress and anxiety among children visiting the hospital: the hospital hero app multi-study and pilot report. Frontiers in pediatrics 11, p. 1132639. Cited by: §1, §2.1, §2.2, §5.2. Rahman and Adaji (2025) P. Rahman and I. Adaji Assessing trustworthiness in persuasive prototypes. In Proceedings of the 13th International Workshop on Behavior Change Support Systems, CEUR Workshop Proceedings, Limassol, Cyprus, Cited by: §2.1. Rapp and Cena (2016) A. Rapp and F. Cena Personal informatics for everyday life: how users without prior self-tracking experience engage with personal data. International Journal of Human-Computer Studies 94, p. 1â17. Cited by: §2.1. Rapp and Tirassa (2017) A. Rapp and M. Tirassa Know thyself: a theory of the self for personal informatics. HumanâComputer Interaction 32 (5-6), p. 335â380. Cited by: §2.1, §5.1. Rekkas et al. (2025) P. Rekkas, K. Verbert, and V. Vanden Abeele Design and evaluation of chatblend: a framework and card deck for the human-centered design of mental health chatbots in blended care. In Proceedings of the 8th ACM Conference on Conversational User Interfaces (CUI â25), External Links: Document Cited by: §2.2. Rooksby et al. (2014) J. Rooksby, M. Rost, A. Morrison, and M. Chalmers Personal tracking as lived informatics. In Proceedings of the SIGCHI conference on human factors in computing systems, p. 1163â1172. Cited by: §1. Rooper et al. (2025) I. R. Rooper, W. W. Liem, M. Burla, J. Gordon, L. M. Baez, R. Kornfield, and A. B. L. Berry Designing values elicitation technologies for mental health and chronic care integration: user-centered design approach. JMIR Formative Research 9, p. e68419. External Links: Document Cited by: §2.2. Saha and De Choudhury (2017) K. Saha and M. De Choudhury Modeling stress with social media around incidents of gun violence on college campuses. Proceedings of the ACM on Human-Computer Interaction 1 (CSCW), p. 1â27. Cited by: §3.1. Saleheen et al. (2016) N. Saleheen, S. Chakraborty, N. Ali, M. M. Rahman, S. M. Hossain, R. Bari, E. Buder, M. Srivastava, and S. Kumar MSieve: differential behavioral privacy in time series of mobile sensor data. In Proceedings of the 2016 ACM International Joint Conference on Pervasive and Ubiquitous Computing, p. 706â717. Cited by: §5.3. Sanders and Stappers (2008) E. B. Sanders and P. J. Stappers Co-creation and the new landscapes of design. Co-design 4 (1), p. 5â18. Cited by: §1, §2.2. Sharon and Zandbergen (2017) T. Sharon and D. Zandbergen From data fetishism to quantifying selves: self-tracking practices and the other values of data. New Media & Society 19 (11), p. 1695â1709. External Links: Document Cited by: §5.3. Shen et al. (2022) J. Shen, L. Iandoli, and M. Aguirre-Urreta Human-centered design for individual and social well-being: editorial preface. AIS Transactions on Human-Computer Interaction 14 (4), p. 446â460. Cited by: §2.1. Shilton (2013) K. Shilton Values levers: building ethics into design. Science, Technology, & Human Values 38 (3), p. 374â397. Cited by: §2.2. Silva et al. (2024) L. M. Silva, F. L. Cibrian, C. Bonang, A. Bhattacharya, A. Min, E. M. Monteiro, J. A. Beltran, S. Schuck, K. D. Lakes, G. R. Hayes, et al. Co-designing situated displays for family co-regulation with adhd children. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems, p. 1â19. Cited by: §2.2. Silva et al. (2025) L. M. Silva, X. Lu, E. X. Liang, and D. A. Epstein Foody talk: exploring opportunities for conversational food journaling. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, p. 1â19. Cited by: §2.2. SlovĂĄk et al. (2017) P. SlovĂĄk, C. Frauenberger, and G. Fitzpatrick Reflective practicum: a framework of sensitising concepts to design for transformative reflection. In Proceedings of the 2017 chi conference on human factors in computing systems, p. 2696â2707. Cited by: §2.1. Son et al. (2024) K. Son, D. Choi, T. S. Kim, Y. Kim, and J. Kim Genquery: supporting expressive visual search with generative models. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems, p. 1â19. Cited by: §2.3. Spinuzzi (2005) C. Spinuzzi The methodology of participatory design. Technical communication 52 (2), p. 163â174. Cited by: §1, §2.2. Statista (2021) Statista Percentage of u.s. adults who use reddit as of february 2021, by age group. External Links: Link Cited by: §3.1. Steen (2013) M. Steen Co-design as a process of joint inquiry and imagination. Design issues 29 (2), p. 16â28. Cited by: §1, §2.2. Suh et al. (2024) S. Suh, M. Chen, B. Min, T. J. Li, and H. Xia Luminate: structured generation and exploration of design space with large language models for human-ai co-creation. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems, p. 1â26. Cited by: §2.3. Suh et al. (2025) S. Suh, M. Lai, K. Pu, S. P. Dow, and T. Grossman StoryEnsemble: enabling dynamic exploration & iteration in the design process with ai and forward-backward propagation. In Proceedings of the 38th Annual ACM Symposium on User Interface Software and Technology, p. 1â36. Cited by: §2.3. Tankelevitch et al. (2024) L. Tankelevitch, V. Kewenig, A. Simkute, A. E. Scott, A. Sarkar, A. Sellen, and S. Rintel The metacognitive demands and opportunities of generative ai. In Proceedings of the 2024 CHI conference on human factors in computing systems, p. 1â24. Cited by: §2.3. van Hierden et al. (2021) Y. van Hierden, T. Dietrich, and S. Rundle-Thiele Designing an ehealth well-being program: a participatory design approach. International Journal of Environmental Research and Public Health 18 (14), p. 7250. Cited by: §1, §2.1, §2.2, §5.2. van Lotringen et al. (2025) C. M. van Lotringen, A. Zirnheld, S. M. Kelders, G. J. Westerhof, and M. L. Noordzij Compassion as a guiding framework for the implementation of digital mental health interventions: an interview study with clients and professionals. Plos one 20 (10), p. e0320710. Cited by: §2.1. Vinay and Kowatsch (2026) R. Vinay and T. Kowatsch An ethical approach to co-designing scalable digital health interventions for vulnerable populations. In Proceedings of the 19th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2026), Volume 1: BIOSIGNALS, p. 683â690. External Links: Document Cited by: §2.2. Wadinambiarachchi et al. (2024) S. Wadinambiarachchi, R. M. Kelly, S. Pareek, Q. Zhou, and E. Velloso The effects of generative ai on design fixation and divergent thinking. In Proceedings of the 2024 CHI conference on human factors in computing systems, p. 1â18. Cited by: §1, §2.3, §5.2. Weisz et al. (2024) J. D. Weisz, J. He, M. Muller, G. Hoefer, R. Miles, and W. Geyer Design principles for generative ai applications. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems, p. 1â22. Cited by: §2.3. Yoo et al. (2026) D. W. Yoo, J. M. Shi, V. J. Rodriguez, and K. Saha Ai chatbots for mental health self-management: lived experienceâcentered qualitative study. JMIR Mental Health 13, p. e78288. Cited by: §2.2. Yoo et al. (2024) D. W. Yoo, H. Woo, S. R. Pendse, N. Y. Lu, M. L. Birnbaum, G. D. Abowd, and M. De Choudhury Missed opportunities for human-centered ai research: understanding stakeholder collaboration in mental health ai research. Proceedings of the ACM on Human-Computer Interaction 8 (CSCW1), p. 1â24. Cited by: §2.2. Zhu et al. (2018) H. Zhu, B. Yu, A. Halfaker, and L. Terveen Value-sensitive algorithm design: method, case study, and lessons. Proceedings of the ACM on Human-Computer Interaction 2 (CSCW), p. 1â23. Cited by: §2.2. Zhu et al. (2026a) S. Zhu, W. Song, J. M. Shi, D. W. Yoo, K. S. Bhat, and K. Saha Designing kriya: an ai companion for wellbeing self-reflection. arXiv preprint arXiv:2601.14589. Cited by: §1, §2.1, §3.1. Zhu et al. (2026b) S. Zhu, H. Zhang, J. D. Chi, S. Nepal, and K. Saha Causal stories from sensor traces: auditing epistemic overreach in llm-generated personal sensing explanations. arXiv preprint arXiv:2605.08590. Cited by: §2.1. Appendix A Appendix Figure 8. Visuals of how different participants prompted for interface design that correlated different health metrics (P10 - left, P2 - center, P2 - top right, P13 - bottom rightFour interface concepts for surfacing relationships between health metrics. Left, from P10: three correlation cards pairing metrics with a strength label and coefficient, plus a plain-language explanation: Steps and Calories show a strong positive link (+0.99), while Steps and Sleep, and Sleep and Calories, each show a moderate negative link (-0.65 and -0.62). Center, from P2: a dark-mode "Sleep & Activity Correlation" card stating that on nights with under 7 hours of sleep, next-day steps drop by an average of 45\%, shown as an overlaid line graph of sleep hours and steps across the week, with a second card below noting Sundays average 67\% fewer steps than other days. Top right, from P2: a "Weekly Schedule" card showing goals adapted per day, with Sunday selected and stats for estimated calories, steps, and tasks, plus a note that the move goal was adjusted from the schedule by 500 cal. Bottom right, from P13: a card titled "Perfect weather for outdoor activity," showing a small bar chart of daily step counts and text explaining that clear skies and 81°F fall within the user's ideal 68â82°F range, historically corresponding to 4,200+ steps on similar days.