Feature: UX for AI - Marina Rambo Portfolio

Any sufficiently advanced technology is indistinguishable from magic.

-Arthur C. Clark

How do we design AI tools people can trust?

As AI features expand across products, many users are unsure how they work - or if they should rely on them. Without clear UX patterns, these experiences risk becoming confusing, disempowering, or even harmful. Magic is fun, but people don’t trust it.

Approach

Faced with user uncertainty around AI behavior and output, we developed a set of UX guidelines designed to make AI features more understandable, trustworthy, and user-driven. These principles prioritized transparency, user control, and feedback loops, and were applied across both web and mobile platforms.

1. Communicate what the AI is doing.

Users can’t trust what they don’t understand. We focused on making AI behaviors visible and legible - clearly showing when the AI was active, what it was doing, and why it produced a certain result

Use plain language and avoid jargon
Visually indicate when the AI is running or making decisions
Explain Results via modals, tool tips, or in-line explanations

Before and after routes for an AI driven optimizer

2. Build trust with clear system boundaries.

To manage expectations and avoid confusion, we clarified what areas the AI has access to, and what changes it is able to make. Users should know when they are interacting with an AI or a rule based automation, and what limitations may apply.

Show what the AI can and cannot do.
Avoid making the system sound human.
Clarify when users are interacting with AI.

3. Give users controls and options.

AI should assist, not override. We made it easy for users to adjust, edit, undo, or bypass AI suggestions entirely, ensuring they retained agency in every step of the process.
‍

Allow users to set parameters and scope of automation.
Let users opt out, manually override, or undo AI decisions.
Add pause or reset buttons for AI powered flows

4. Offer feedback channels.

Trust deepens when users feel heard. We built simple consistent ways for users to provide feedback on AI output - helping both the user trust and the system improve over time.

Offer the ability to provide feedback at flow completion.
Ensure simple problem reporting exists.
Analyze quantitative and qualitative data to improve flows.

5. Prioritize ethical and accessible design.

We grounded every design decision in equity and inclusion. That means accessible layouts, privacy conscious messaging, and language that reflects a wide range of perspectives and needs.

Use inclusive respectful language.
Communicate data use and privacy clearly.
Meet WCAG 2.1 AA or better.