Personalizing Personal Health Interfaces: Co-Design with Generative AI
Karthik S. Bhat, Vidhi Shah, Vedika Agnihotri, Dong Whi Yoo, Koustuv Saha
cs.HC · cs.AI · cs.CY
Quick verdict
Co-design study (n=14) shows generative AI helps users prototype health interfaces but injects model defaults; rich qualitative insight, low generalizability.
Summary
This paper reports an IRB-approved qualitative co-design study in which 14 US participants — mostly college students aged 18–24, recruited via college subreddits and snowballing — reimagined Apple Health and Google Health interfaces by prompting Figma Make, a generative UI tool, during roughly one-hour think-aloud Zoom sessions combining interviews, sticky-note ideation and iterative prototype generation. Thematic analysis of transcripts, prompts and generated designs (26 feature themes, Table 2) identifies three design directions: connecting health data for context, extending tracking toward planning and action, and personalizing visualization and interaction. Conversational features converged on standard chatbot layouts while visualization designs varied widely. Participants valued how AI materialized loosely articulated ideas, but model defaults, unrequested additions and 2–10 minute generation latency shaped iteration, and perceived control depended on alignment with their intentions (exit-survey control mean 4.29/5). A values gap emerged: interpretability and accountability were readily operationalized, whereas privacy, trust and emotional safety, despite high stated rankings, rarely entered prompts. Key limitations are a single proprietary tool, one-off sessions, a homogeneous self-selected sample, no comparison condition, and the attribution problem the authors themselves foreground.
Review complete
Assessment at a glance
5
risk flags identified
● high 0
● medium 2
● low 3
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Participants' redesigned health interfaces clustered into three directions: contextual connection of health data, forward-looking planning and action, and personalized visualization/interaction.
- Conversational AI features converged on standard chatbot layouts while data-visualization designs diverged widely, suggesting generative systems may reproduce and stabilize interface conventions.
- Generative co-design lets non-designers materialize loosely articulated ideas, but model defaults, unrequested additions, and 2–10 minute generation latency shape iteration and perceived agency.
- The method operationalizes interpretability and accountability far more readily than privacy, trust, and emotional safety, which rarely entered prompts despite high stated importance.
- Generative co-design creates an attribution problem: resulting artifacts conflate participant intent with model-introduced design decisions, complicating their use as evidence of user preference.
Strengths
Genuinely novel empirical site
Moves participatory design beyond researcher-mediated translation: participants themselves generate functional prototypes via Figma Make (Sections 1, 5.2), an early empirical account of GenUI used by non-designers in a co-design process.
Explicit methodological contribution
Section 5.2 articulates the attribution problem — model-introduced decisions conflate participant intent with model priors — and proposes concrete mitigations (multiple first-pass generations, preserving prompt-generation histories) directly usable by future studies.
Concrete, checkable design corpus
Triangulated values-gap finding
Candid limitations treatment
Weaknesses
No comparison condition
Thin analytic reliability reporting
Undocumented, non-reproducible generative substrate
Homogeneous self-selected sample
Quantitative tables invite over-reading
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
The paper (arXiv:2609.15046, v1, 14 Sep 2026) reports an IRB-approved qualitative co-design study in which 14 participants redesigned Apple Health or Google Health using Figma Make, a prompt-based generative UI tool. Two research questions structure the work: what design patterns emerge (RQ1) and what opportunities and limitations generative AI introduces into co-design (RQ2). Claimed contributions are a design-space characterization for personal health interfaces, an empirical account of how generative tools redistribute the translation of participant ideas into prototypes — including an 'attribution problem' — and methodological guidance for future generative co-design. The scope is appropriately modest: an exploratory account, not a system evaluation or causal test.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 5 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is generative co-design in this paper?
Instead of sketching ideas for researchers to implement later, participants themselves prompt a generative UI tool (Figma Make) to create working interface prototypes. In this study, 14 people redesigned Apple Health or Google Health this way in roughly one-hour, think-aloud Zoom sessions.
What is the attribution problem in AI-assisted design research?
When a generative model turns a rough description into a prototype, it adds design decisions the participant never requested. If the participant keeps or tweaks them, researchers cannot tell which features reflect user intent versus model defaults. The paper proposes generating multiple first drafts and preserving prompt histories to make the model's role visible.
What did participants want changed in their health apps?
Mostly contextual connections across metrics (e.g., sleep vs. steps, weather vs. activity), forward-looking features like goal forecasts and planning, and personalized visuals — while every chatbot feature converged on the same standard chat layout of message bubbles, suggested questions and a text field.
What were the study's main limitations?
Single one-hour sessions with one proprietary tool (Figma Make), a homogeneous self-selected sample of mostly 18–24-year-old college students, no comparison condition, and an inherent difficulty separating participant intent from the AI's contributions to each design.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures 1–8 are referenced but only captions were extracted; judgments about the generated visual designs rest on textual descriptions, not the images.
- Verbatim participant prompts are not included in the extracted text; RQ2 claims about prompting behavior rest on the authors' characterizations and the Table 2 summaries.
- This is arXiv v1 (14 Sep 2026), apparently a preprint; peer-review status and any venue-specific revisions are unknown.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.