Turning Unpredictable AI into a Guided Experience

UX Case Study by Daria Kartseva

UX Design for Generative Content Tools

Creating content with AI should feel exciting - not confusing.
But today, users struggle to understand what AI can do and why the result is sometimes wrong.
This project explores how UI can help people communicate with generative models better - using the example of music creation.

The Problem
Interfaces usually say: “Just write a prompt!”
But users:
don’t know what AI understands,
get ignored by the model,
and can’t control the result.
This leads to frustration and wasted time.
The Users
This project focuses on:
non-expert creators who use AI occasionally,
and beginners who are not familiar with model limits.
They need help understanding what to write - and how to get a good result.
The Goal
Make generative tools:
more predictable,
more guided,
and more satisfying to use.
So people feel in control, not in a lottery.
My Role
Researching interaction with generative AI
Studying model limitations from UI perspective
Creating UI principles & visual examples

Visual examples are based on a music generation interface - but principles apply to any generative AI product.

Give the user a bit of structure

When there is one big input field, users must remember everything:
genre, style, length, mood, voice, language…

This creates heavy cognitive load.

Splitting the input into small, clear groups:

helps users focus step-by-step
reduces mistakes and forgotten details
creates a predictable workflow

It takes more space on screen - but it gives users more confidence and better results.

Structure is a form of assistance.

Show what the model understands

When people use a generative AI tool, they often believe the system can understand anything they type.
But each model has limits.
If a user writes words the model does not support, the system may ignore them - and the user never knows why.

To reduce this frustration, the interface should show what the AI can understand:

a list of supported tags or categories
suggestions while typing

This way, users can choose from visible options instead of trying to guess the right words.
The UI becomes a guide that helps them speak the “language” of the model - with less confusion and better results.

Clear boundaries help creativity,
because users know how to communicate effectively with the system.

Offer flexible input

People create content in different ways.
Some need help - they prefer to click on ready options.
Others know exactly what they want - they want to type freely.

The interface should support both:

Choosing from tags → good for beginners, fast decisions, guided by what the model can actually do
Free text input → good for experts, more control
AI-assisted prompt creation → for users who don’t want to think about wording at all

These input methods work together.
The user can start by picking tags, then add details in text, or give a reference and let AI build the prompt for them.
This flexibility:

reduces stress and guessing
works for different experience levels
keeps creativity open

A good prompt UI is not “one field for everyone”,
but a set of options that adapts to the user.

Be honest about model

A good interface protects users from false expectations before they happen.

limitsInterfaces must set realistic expectations:

which input affects the result
which details are often misinterpreted
what will not work yet

This message should appear in the moment of input,
not hidden in FAQ.

Clear communication builds trust and loyalty.
Honesty is key to a good AI-UX relationship.

Validate and give feedback

Users shouldn’t wait until the result to discover a mistake.

If the model doesn’t understand something, the UI should show it right away - with simple colors, icons, or short hints like:
“This idea may be ignored — try another word?”
“Did you mean electronic?”
“This part will be used only as lyrics.”

Before the user clicks Generate, the system can clearly show: