AI UX | Strategy Exploration
GEMINI: Designing for trust and user control in an AI interface
Overview
Project Summary:
Gemini launched with an advantage most AI tools don't have. People already trusted Google. But brand trust and trust in an AI system's outputs are two different things. That gap is where I started. What would it look like if an AI interface was designed to actively build trust, reduce overreliance, and give users more control over what they're interacting with? This project is my answer to that question.
Project Type: Strategic UX Exploration
My Role: Product Designer
Domain: Human-AI Interaction Design
Tools used: Figma, Miro, Google Gemini, ChatGPT
Focus Areas: Transparency in AI systems, Confidence Signalling, Bias Awareness, User Agency
PROBLEM
Designed for ease, not designed for trust...
AI tools make it effortless to get answers quickly. But effortless isn't the same as reliable. The outputs don't tell you how confident the system is, whether the response might be biased, or where its limitations are. That lack of transparency pulls users in two directions. Some overtrust what they're given, accepting outputs at face value without questioning them. Others don't know when they should trust the system at all. Both are design problems. And both come from the same root cause: an interface that was built for ease, not for informed judgment.
GUIDING PRINCIPLES
What guided the design…
I started by asking myself a simple question: if I were using Gemini and wasn't sure whether to trust what it was telling me, what would actually help? I worked through that from a user perspective first, thinking about what was missing and what could give users more confidence and control.
When I came across the Microsoft Research Human-AI Interaction Guidelines (CHI 2019) I found that my thinking was already aligned with principles established by research. That gave my ideas a stronger foundation and helped me refine them further.
Three priorities emerged from that process: clarity, trust, and user control.
Clarity:
Ensuring users understand what the system is generating and why.
Mapped Guidelines:
Make clear what the system can do
Show contextually relevant information
Explain why the system did what it did
Trust:
Calibrating reliance by communicating confidence, limitations, and verification signals.
Mapped Guidelines:
Make clear how well the system can do it
Mitigate social biases
Show uncertainty where appropriate
User Control:
Allowing users to influence outputs and maintain oversight.
Mapped Guidelines:
Support efficient correction
Allow user control over system behaviour
DESIGN STRATEGY
Turning ideas into design…
Each feature came from a specific gap I identified in how Gemini handles trust, transparency, and user control. Here is what I designed and why.
Confidence Indicator
Principle Alignment: Trust, Clarity
Purpose: Help users gauge how much to rely on a response before acting on it.
Details: Not all AI outputs are equally reliable. This feature surfaces how confident Gemini is in a response, giving users a visible signal to calibrate how much they should rely on what they're reading. It doesn't tell users what to think. It gives them the information to decide for themselves.
Click on the image to enlarge.
Prompt Refinement & Prompt Strength Indicator
Principle Alignment: Clarity, User Control
Purpose: Help users craft stronger prompts so they get more relevant and reliable outputs.
Details: Vague prompts produce vague outputs. This feature helps users craft stronger, more specific prompts before they submit them. A prompt strength indicator shows how well-formed the input is, encouraging more deliberate and structured interaction with the system.
Output Classification (Factual / Creative / Speculative)
Principle Alignment: Clarity, Trust
Purpose: Help users understand what kind of answer they're getting before they act on it.
Details: Not everything Gemini produces is the same kind of answer. Some responses are factual, some are creative, some are speculative. This feature labels each output so users know what they are working with before they act on it. For factual queries it also distinguishes between verified and unverified sources.
Click on the image to enlarge.
Bias Detection Nudge
Principle Alignment: Trust
Purpose: Catch biased or leading prompts before they shape the output.
Details: Sometimes the way a question is framed shapes the answer before the AI even responds. This feature flags potentially biased or leading prompts and suggests neutral alternatives before the output is generated. It catches blind spots early rather than after the fact.
Reasoning Chains & Source Transparency
Principle Alignment: Clarity, Trust
Purpose: Show the thinking and sources behind every response so users can verify what matters.
Details: Trust is easier to build when you can see the thinking behind an answer. This feature shows the step-by-step reasoning Gemini used to arrive at a response, alongside the sources it drew from. Users can follow the logic, verify what matters, and make more informed decisions
Click on the image to enlarge.
User Control Sliders & Output Settings
Principle Alignment: User Control
Purpose: Put control back in the user's hands before a response is generated.
Details: Different users need different things from an AI tool. This feature puts control back in the user's hands, letting them adjust tone, depth, creativity level, and source preferences before generating a response. A bias awareness toggle is also included for users who want an extra layer of critical oversight.
Click on the image to enlarge.
SYSTEM PERSPECTIVE
What changes with this design…
These features don't work in isolation. Together they shift the experience from passively consuming AI outputs to actively and critically engaging with them. That's the difference between a tool that thinks for you and one that thinks with you.
REFLECTION
What I took away…
Interfaces shape behaviour: The way an AI tool is designed doesn't just affect how people use it. It affects how critically they think while using it.
Transparency builds better habits: When users can see confidence levels, reasoning, and sources, they engage more thoughtfully. Visibility changes behaviour.
Control reduces overreliance: Giving users agency over inputs and outputs shifts them from passive consumers to active participants. That's a meaningful difference.
Design has a responsibility here: AI interfaces are making decisions about what people see and trust every day. Designing with that in mind isn't optional, it's the whole point.
The ideal next step would be testing these concepts with real users to see which features actually shift behaviour and which ones add friction.