A safety layer for everyday AI health and wellbeing conversations.
NurAi Companion
NurAI is a speculative plug-in concept that helps people use existing AI tools for health and wellbeing questions with more transparency, boundaries and user control.
This project was done as part of Naqiyah Mustafa's Masters Degree in her class Prototyping Services & Interfaces.
Project at a Glance
Module: MMED9029 Prototyping Services and Interfaces
Project type: Service and interface prototyping
My role: UX/service designer, researcher and prototype designer
Tools: Miro, PowerPoint, Wix and AI-supported production
Outcome: Three low-to-medium fidelity prototypes exploring safer AI health and wellbeing interactions
NurAI Companion: a safety layer for everyday AI health and wellbeing conversations.
Project Challenge
The problem began with my own use of AI. I wanted AI to support my health and wellbeing because it was immediate, private, and easy to access. But I began questioning how trustworthy that support really was.
A personal starting point: fast support, comfort and trust.
AI use ranges from practical help to emotionally significant support. A person may ask it to count meal macros, explain confusing health information, talk through stress, or make sense of symptoms. As the personal stakes increase, the interaction often still looks the same.
AI use can move from practical help to high-stakes symptom questions.
The opportunity is to support people without surrendering their judgment. Fast access and simple explanations can be helpful, but the interaction must not encourage over-reliance, false reassurance or the idea that AI can decide what a person should do.
The design requirement: support the user’s decision-making but never replace it.
Context: This is a speculative design project for a Prototyping Services and Interfaces Class. The design context is the public-facing AI platform ecosystem: tools such as ChatGPT, Gemini, and Claude that may be used for sensitive health and wellbeing conversations.
Primary user: Everyday AI users who ask quick health or wellbeing questions at home, especially when they feel unwell, anxious, tired, or unsure whether symptoms are serious.
Scenario: After work, a user comes home exhausted. Their head is pounding, they feel nauseous, their body aches, and they have no appetite. They want quick support, but they are also worried that AI could either make them panic or dismiss something important.
The design challenge is not to make AI diagnose them. It is to help the AI be honest about uncertainty, provide calm general guidance, make safety boundaries visible and leave the user in control of what happens next.
The Proposed Solution: NurAi
NurAI is a cross-platform plug-in and safety layer that could attach to existing AI platforms. It is not a medical diagnosis tool, and it does not replace a doctor. Instead, it makes sensitive AI responses more practical, safe, and transparent at the moment a user needs support.
-
User choice: Safe Mode is offered, not forced.
-
Transparency: The AI shows what it can and cannot know.
-
Boundaries: The response separates general guidance from diagnosis.
-
Evidence: Sources and reasoning are made visible.
-
Control: The user chooses the next step, including when to seek professional support.
NurAI evolved from a standalone health companion into a cross-platform safety layer.
Prototype Strategy
I used three methods to examine the concept from different angles: a user flow and wireflow to test interaction logic; a storyboard to explore the emotional experience; and a speculative transparency label to test what accountable AI guidance could reveal at the moment of use.
Three prototypes test logic, emotional experience and future accountability.
Prototype One: User Flow & Wireflow
Design Question: How should a sensitive AI health interaction work from the user’s first question to their next step?
Problem Addressed: AI can move straight from a symptom description to an answer, giving the user little control over how sensitive health guidance is handled. The flow needed to make consent and boundaries visible before the response becomes influential.
Early flow: the answer was prioritised before user consent.
What changed: My early flow over-prioritised the answer itself. I revised it to introduce a Safe Mode prompt before health-sensitive assistance. The revised interaction offers two possible safety triggers: a user can flag an immediate concern, or NurAI can ask whether the user feels distressed and wants active recommendations.
Choice turns Safe Mode into consent.
Final Prototype: The final flow makes safety visible without taking agency away. The user can choose Safe Mode, continue normally, request more context, or move toward professional support. The response makes relevance, uncertainty, sources, and boundaries visible before the user decides what to do next.
Final flow: safety is visible without removing agency.
Wireflow 1: Safe Mode asks permission before changing the interaction.
Wireflow 2: the response separates support from diagnosis.
What I Learned: NurAI does not make AI ‘correct’. It makes uncertainty, boundaries and user choice visible. The safety interaction should be an invitation, not an automatic takeover.
Prototype 1 learning: the value is in the interaction architecture.
Prototype Two: Storyboard
Design question: How should this experience feel when a user is tired, worried, and looking for a fast answer?
Problem addressed: The user wants reassurance but also honesty. An overly confident AI response can encourage dependence; an alarming one can increase panic. I needed to show a calmer transition from uncertainty to control.
The scenario begins with a common tension: speed versus reassurance.
Early Process: I first used a rough text-driven storyboard to map the emotional journey. It showed an unwell user asking AI for help, receiving an answer, and choosing a next step, but it revealed that the service began guiding before the user had actively consented to the change in interaction.
Rough storyboard: the missing consent moment becomes visible.
What Changed: I moved consent earlier in the sequence and created a low-fidelity line storyboard to test the emotional clarity before polishing the interface. The user now sees Safe Mode, chooses whether to activate it, sees transparent limits and sources, and then chooses their next step.
Low-fidelity storyboard: the story works before the interface is polished.
Final Prototype: The final storyboard follows one person through six moments: they come home unwell, ask AI for help, see the Safe Mode invitation, choose a transparent response, view their options, and regain control. It demonstrates that honest support can be calm without pretending to diagnose the user.
Final storyboard: the journey moves from uncertainty to control.
What I Learned: The goal is not reassurance at any cost. It is honest support that helps a person tolerate uncertainty and decide what support they need next.
Prototype 2 learning: NurAI manages uncertainty instead of hiding it.
Prototype Three: AI Response Transparency Label
Design question: What should AI reveal before a user acts on health-related guidance?
Problem addressed: AI responses can hide their purpose, confidence, uncertainty, source quality and data boundaries. When people are feeling unwell, these invisible details can strongly shape whether they trust or act on the response.
Three speculative directions tested where transparency should live.
Development and decision: I explored a response label, a browser-style notice, and a public awareness artefact. I selected the AI response transparency label because it sits beside the response at the exact point where a user is deciding what to believe or do. It turns transparency into something actionable rather than a warning hidden elsewhere.
Final artefact: accountability sits beside the answer, not below it.
Critical reflection: A transparency label can clarify an AI response, but it can also look too authoritative. A formal-looking label may be mistaken for approval, ‘moderate confidence’ may be misunderstood, source counts do not guarantee quality and users may still form an emotionally dependent relationship with AI.
Transparency must support judgment, not manufacture trust.
What I Learned: The label can clarify an AI response, but it must not look so official that it creates false reassurance. Future work should test whether people understand the label and use it as intended.
Final Reflection
Across the three prototypes, I explored trust as a sequence of visible choices: consent, uncertainty, sources, boundaries, and user-controlled next steps. NurAI is not designed to replace healthcare professionals. It is a way to make everyday AI health and wellbeing conversations safer, more transparent, and less likely to encourage blind reliance.
Trust becomes a sequence of visible choices.
NurAI Companion is a speculative prototype,
not a clinical product or medical advice service.