Technical Summary: Developing an AI-Powered UX Research Point of View for Digital Health in a Regulatory Context

Problem Statement

Digital health interventions hold significant potential for improving health outcomes among marginalised populations, yet their implementation in vulnerable, legally restrictive contexts presents unprecedented User Experience Research (UXR) challenges. In Nigeria, the Same-Sex Marriage (Prohibition) Act 2014 criminalises same-sex relationships, creating a "triple stigma" (HIV-related, sexual minority, and legal criminalisation) for Men who have Sex with Men (MSM) and transgender individuals living with HIV/AIDS. This legal environment suppresses care-seeking behaviours and renders traditional UXR methods—such as focus groups or the collection of personally identifiable information—potentially dangerous for participants.

Existing UXR frameworks, such as the UXR Point of View (PoV) Playbook, have primarily been applied in commercial or clinical contexts with accessible user populations. They lack explicit adaptations for populations where research participation itself carries life-threatening consequences. Furthermore, there is a gap in methodologies that leverage recent advancements in Generative AI (GenAI) to enhance research rigour while simultaneously maintaining participant safety through privacy-preserving and risk-assessment capabilities.

Methodology

This study introduces a Generative AI-augmented UXR methodology grounded in the UXR PoV Playbook, specifically adapted for the Nigerian MSM and transgender HIV care context. The research employs a four-stage process, integrating empirical data from two primary sources:

1. A co-design study involving clinicians, NGO staff, and care providers in Nigeria.
2. A systematic review of digital health interventions for HIV/AIDS in West Africa.

These empirical sources are synthesised alongside five theoretical frameworks: Extended UTAUT, Minority Stress Theory, WHOQOL, Privacy Calculus, and XAI Trust Theory. GenAI is utilised as a co-analytic tool under strict researcher-led oversight to process evidence, generate hypotheses, and articulate stakeholder narratives.

The four-stage process is defined as follows:

1. GenAI-Supported Evidence Synthesis and Hypothesis Generation: GenAI extracts recurring patterns from the evidence corpus (psychosocial, behavioural, and design dimensions) and clusters them to propose testable hypotheses linking UX variables to adherence and Quality of Life (QoL) outcomes.
2. Foundational Planning and Stakeholder Roadmapping: GenAI contextualises how target populations manage stigma and maps the stakeholder landscape, identifying tensions between individual privacy and institutional reporting requirements.
3. Insight Generation and Play Card Development: Using the "Building Blocks" model (Foundation, Data Collection, Insight Generation, PoV), GenAI iteratively connects theoretical frameworks and empirical patterns to produce actionable design artefacts.
4. PoV Narrative Construction: GenAI supports the creation of role-specific Point of View narratives for diverse stakeholders (users, clinicians, developers, NGOs), ensuring tone consistency and clinical accuracy.

Key Results

The methodology yielded a structured progression of findings across the four stages:

1. Theoretical Synthesis and Hypotheses

A three-layer analytical model emerged: Psychosocial → Behavioural → Design. This model illustrates how internal states (e.g., stigma internalisation, fear of disclosure) produce observable engagement patterns (e.g., concealment-driven interaction) that must be addressed by specific design decisions. Ten theory-informed hypotheses were generated, including:

• H1 & H2: Perceived privacy control positively predicts emotional comfort, which mediates the intention to use the platform.
• H3: Lower perceived interface complexity increases sustained engagement among high-stigma users.
• H9: Neutral notification framing reduces anticipatory anxiety.
• H10: Optional mood check-in features improve emotional awareness without increasing disengagement.

2. Stakeholder Mapping

The study identified eight distinct stakeholder groups, highlighting two critical tensions:

• Privacy vs. Reporting: The conflict between the need for anonymised user data and the institutional requirement for outcome reporting.
• Safety vs. Feasibility: The tension between clinicians' desire for functional completeness and users' need for minimal cognitive friction and maximum discretion.
  Crucially, the research distinguishes between MSM users (who primarily navigate identity concealment) and transgender users (who face compounded discrimination regarding gender identity and HIV status), necessitating differentiated design responses.

3. UXR Play Cards

Ten actionable UXR Play Cards were developed. These serve as "boundary objects" bridging research evidence and interface design. Each card operationalises a theoretical principle into a concrete design artefact, featuring:

• Front Face: Thematic title, issue type, and best practice.
• Back Face: Theoretical anchor, empirical evidence, testable hypotheses, and specific design plays (e.g., "Privacy Architecture," "Stigma Sensitive" design).
  Examples of design implications include "Privacy-by-architecture," "Stealth mode UI," "One-task-per-screen," and "User-defined phrasing" for notifications.

4. Point of View (PoV) Narratives

Population-specific PoV statements were refined to guide design strategy:

• MSM PoV: Emphasises that psychological safety and user-controlled privacy are stronger determinants of engagement than feature breadth.
• Transgender PoV: Highlights the need for identity-affirming, confidentiality-preserving interfaces that coordinate ART adherence with gender-related health needs.
• Developer PoV: Asserts that privacy-by-architecture and low cognitive load are non-negotiable technical requirements, not optional enhancements, in marginalised contexts.

Significance and Claims

The paper claims to provide a replicable, stigma-aware, and privacy-centred framework for the responsible integration of GenAI into UXR practice. Its core contributions include:

1. Methodological Adaptation: It demonstrates how to adapt the UXR PoV Playbook for populations where research participation is dangerous, moving beyond commercial contexts to address structural criminalisation.
2. Human-in-the-Loop Governance: The study positions GenAI as a "cognitive collaborator" that accelerates synthesis and hypothesis generation, while explicitly retaining human researchers as the ultimate authority for interpretive validation. This is critical because GenAI training data often skews toward majority-world, legally permissive contexts, risking the production of dangerous recommendations (e.g., suggesting transparency features that could expose users).
3. Differentiated Design: The framework resists the homogenisation of "key populations," offering distinct design guidance for MSM and transgender users based on their unique psychosocial profiles.
4. Theoretical Integration: It successfully operationalises multiple theoretical frameworks simultaneously within single design artefacts (the Play Cards), capturing the complex interdependencies of trust, stigma, and infrastructure.

The authors note that while the framework advances human-centred digital health design, the Play Cards and PoV narratives require future empirical validation through participatory co-design workshops and field pilots with end-users in Nigeria. The work serves as a foundational step toward formalising capability engineering for digital health in marginalised contexts.