I Asked an AI to Design a UX Flow & Here’s What It Got Wrong (and What It Got Way Too Right)

HUKHTA PATEL
Jul 20
3 min read

Ever caught yourself staring into blank space with thoughts all over the place as a designer?

Let’s be honest: We’ve all had that moment. You’re staring at a blank screen, a mug of cold coffee in hand, thinking, “What if… I asked ChatGPT to design this flow for me?”

Well, I did. And friends, it was genius, glitchy, and gloriously unhinged.

The Brief I Gave the AI:

Design a basic UX flow for a mental wellness app that helps users track moods, journal thoughts, and connect with AI-powered emotional support. Simple. Soothing. Straightforward. Meaningful

Step 1: The Persona Party

AI Output:

Meet Jamie: a 27-year-old freelance illustrator from Austin who owns three succulents and uses astrology memes to process emotions.

My Thoughts:

Okay, not bad. A bit cliché, but at least the AI understands that millennials love plants and trauma memes.

What It Got Right:

Quick persona generation with enough detail to get started.

What It Missed:

Empathy. Jamie felt like a persona built from Instagram bio tropes, not actual user pain points.

Takeaway: AI can sketch personas, but it’s up to us to dig deeper. Use it as a kickstarter, not a substitute for real research.

Step 2: The Wireframe Wackiness

I asked it to create a low-fidelity wireframe in plain text. Here’s part of what it came back with:

It replied:

[Home Screen]

- Header: “Welcome Back, Jamie 🌞”

- Mood Tracker Button

- Journaling Prompt: “What’s on your mind today?”

- CTA: “Talk to Your AI Companion” [Bottom Nav] - Home | Journal | Insights | Settings

Me, screaming internally:

Why is the sun emoji always the default mood branding? And why does every AI companion sound like a gentle cult leader?

What It Got Right:

A clear, logical layout. AI follows conventions, so the IA wasn’t bad at all.

What It Got Way Too Right:

It was too safe. Too sanitized. No surprise, no delight, no human nuance.

Takeaway: If you’re stuck, AI can help structure your wireframes. But magic lives in the margins, the microcopy, the edge cases, the awkward silences. That’s still our job.

Step 3: The Content Strategy Curveball

I asked ChatGPT: “What should the tone and voice of this app be?”

It replied:

“Supportive, empathetic, non-judgmental, with a sprinkle of positivity.”

SPRINKLE ? Honey, it dumped the whole jar of glitter.

Examples it suggested:

“You’re doing better than you think.”
“Remember, growth isn’t linear.”
“You are valid and worthy.”

Me:

Did I just accidentally build a Pinterest board?

What It Got Right:

Tone matched the domain.

Where It Failed:

No adaptability. No nuance across contexts, same tone whether you missed a meditation or lost your job.

Takeaway: AI can channel generic empathy, but tone without context becomes wallpaper. We need to modulate voice based on user state, timing, and need.

Real-World AI x UX Moments

I’m not the only one playing with the robot intern. A few examples worth mentioning:

Duolingo uses AI to dynamically personalize learning paths , but always with a human in the loop for tone and pacing. (Source: Fast Company, 2024)
Figma’s AI features now auto-generate UI components, but designers are using it more for rapid iteration and less for final design. (Source: Figma Blog, 2023)
Airbnb’s internal AI tools assist with customer messaging, but human agents tweak final copy to fit cultural and emotional contexts. (Source: MIT Tech Review, 2024)

"AI should assist. Designers should decide."

Final Verdict: AI in UX Is Like a New Intern with a Design Degree from YouTube

Helpful? Absolutely.

Perfect? Never.

Quirky? Always.

If you’re a designer navigating this brave new world, treat AI like a brainstorming buddy. Ask it weird questions. Test its limits. Just don’t let it name your buttons or design your empty states without supervision.

Because creativity isn’t just in the doing, it’s in the deciding and that's what makes design soooo intuitive. That, my friends, is still deeply, beautifully human.

TL;DR – What AI Got Right & Wrong

Category	Got Right	Got Wrong
Personas	Fast, structured setup	Lacked real emotion and insight
Wireframes	Logical layout, usable IA	Too sterile, no personality or edge cases
Tone & Content	Empathetic baseline language	Overgeneralized, lacked contextual depth
Overall Role	Speedy assistant for iteration & structure	Not a replacement for research or nuance