UX Design with AI ┃ 1.1 What AI Can and Can't Do in UX Design

AI Fundamentals for the UX Workflow

Using AI in UX design has quietly become a core professional skill.

Which means the question has changed. It's no longer "Should I use AI?" It's "How do I use it well?"

That's what this lesson is for. We're going one level past casual AI use, and building a foundation you can actually work from.

Two kinds of models are reshaping design workflows right now: large language models and image generation models. We'll look at what each one is genuinely good at, where it hits a wall, and how it fits into a real UX process.

First, What Are We Actually Talking About?

Let's define AI in plain terms.

Artificial intelligence(AI) is technology that takes your input and uses data and models to predict, recommend, generate content, or help solve a problem. Put simply: it lets a system handle some of the work a person used to judge or produce by hand.

There are many types of AI, sorted by what they're for and what they output. But in a UX workflow, two show up constantly:

• Large Language Models (LLMs)
• Image Generation Models

Understanding the difference between these two isn't trivia. It determines which tool you reach for, and what you can reasonably expect from it.

Large Language Models(LLM) — The Text Engine

LLMs like ChatGPT and Claude understand and generate text.

That makes them genuinely useful for the text-heavy parts of UX work:

• Drafting user personas
• Writing interface copy
• Summarizing research
• Organizing scattered ideas

If the deliverable is made of words — and in UX, a surprising amount of it is — an LLM can get you to a first draft fast.

Image Generation Models — The Visual Engine

Tools like Midjourney and DALL·E work differently. They turn a text prompt into an image.

That makes them useful when you need to see an idea quickly:

• Exploring early visual concepts
• Building moodboards
• Comparing style directions

But here's the part that trips people up, so let's be precise about it:

Image generation models are not there to produce your finished work. They're there to take the vague direction sitting in your head and make it visible — fast enough that you can compare it against three other directions.

That's the real value. They widen the range of ideas, then help you narrow the direction. Treat them as a thinking tool, not a production tool, and they'll serve you well.

The Rule That Governs Everything Below

Before we get into the seven ways AI shows up in design work, one principle has to be locked in first — because everything else depends on it.

AI now touches almost every stage of the workflow: organizing research, generating ideas, drafting screens, writing copy, exploring visual concepts.

But: AI is not a tool that completes design for you.

It's a tool that produces drafts quickly, lets you compare several possibilities at once, and absorbs the grind of repetitive work.

Which means the designer's job doesn't shrink — it sharpens. You are the one making the final call, and you make it based on user context and product goals. AI can hand you ten options. It cannot tell you which one is right for the person who'll actually use this product.

Hold onto that as we go through what follows.

Seven Ways AI Shows Up in Real Design Work

1. Drafting Layouts

Tools like Uizard and Stitch can generate wireframes or UI screen drafts from a text description, a reference image, or a rough idea.

This is most valuable during early ideation — and the reason is worth understanding. The point isn't to produce a polished, high-fidelity screen on the first try. It's to compare several possible screen structures quickly, before you've committed to any of them.

Example prompt: "Onboarding screens for a fitness app aimed at busy office workers."

In seconds, you can see a rough shape for the welcome screen, the goal-selection screen, the recommendation results screen — and immediately start reacting to what works and what doesn't.

💬 We'll cover how to actually write effective prompts throughout the coming lessons.

2. Drafting User Personas and Stories

LLMs like ChatGPT are effective at producing a first pass at user personas and user stories.

Feed it the user's goals, behaviors, frustrations, and usage context, and it will return a structured persona or a set of user story statements you can build from.

But the quality of what comes back depends entirely on what you put in. To get a draft that's actually close to what you need, your prompt has to be specific and structured. Tell the model:

• Which users you're dealing with
• What product context they're operating in
• What format you want the output in

Vague inputs produce generic personas — the kind that describe nobody in particular. Specific inputs produce something you can actually work with.

3. Preparing User Tests and Organizing Feedback

Testing platforms like Maze now include AI summarization — helping you organize test questions, or quickly cluster response and behavioral data.

That's genuinely useful. But this is also the area where the limits of AI are clearest, so let's be direct about them.

User testing is fundamentally about human observation and interpretation. AI can summarize the results. It can surface recurring problem patterns. What it cannot do is tell you why a user behaved that way, or why they reacted the way they did.

That interpretation is still the designer's job.

Will AI eventually analyze facial expression, tone of voice, and behavioral data to read user reactions more precisely? Possibly. But reading a person's emotion and intent requires reading their context alongside it — and that still calls for a human observing carefully and judging carefully.

4. Drafting Content and Microcopy

LLMs like ChatGPT and Claude are excellent for getting text-based work started:

• Headlines
• Button copy
• Empty state messages
• Error messages
• Onboarding copy

The advantage here is comparison, not completion. For a single button, you can instantly generate a friendly version, a calm version, and a professional version — and see them side by side.

That does two things. It removes the pressure of writing from a blank page. And it gives you concrete material to refine against your brand voice and your user's context — which is where the actual craft happens.

5. Summarizing Research

Anyone who's run UX research knows the pile: interview notes, survey responses, competitive analysis, reference material. Somewhere in there are the insights — but finding them takes hours.

This is one of the highest-leverage uses of an LLM. Feed it multiple sources, and it can surface the repeating patterns and key insights across all of them.

Example prompt: "Here are several interview transcripts. Identify the five user pain points and needs that appear most repeatedly."

The value isn't just speed. It's that you get research organized into a form you can connect to a design decision — which is the whole reason you did the research in the first place.

6. Exploring Visual Concepts

Midjourney, DALL·E, and Stitch turn a text prompt into visual direction — fast.

Example prompts:

• "Moodboard for a calm, trustworthy healthcare app."
• "Main landing page visual for a bright, energetic fitness service."

The point, again, is comparison. You're not commissioning final artwork. You're putting three or four possible directions next to each other so you can see which one feels right — and, just as importantly, which one doesn't.

Widen the range. Then narrow the direction. That's the job.

7. Expanding Feature Ideas

During ideation, LLMs are effective at broadening the possibility space — generating new feature ideas or UX improvement directions you hadn't considered.

One of the most useful techniques: take a single idea and run it through different user situations. How does this feature serve a first-time user? A user in a hurry? A returning user? Switching perspectives like this surfaces possibilities a single viewpoint would miss.

Example prompt: "Based on the reasons busy office workers struggle to exercise consistently, suggest feature ideas a fitness app could offer. For each idea, describe the problem it solves, the key screens involved, and the value the user gains."

Notice the structure of that prompt. It doesn't just ask for ideas — it asks for ideas organized in a way you can evaluate.

💬 How to review AI-generated ideas, and what criteria to use when selecting between them, is something we'll cover systematically in the lessons ahead.

Where This Leaves Us

You now have the map: two model families, seven places they enter your workflow, and one rule that governs all of it.

AI produces drafts. It compares possibilities. It absorbs the repetitive work that used to eat your afternoon.

What it doesn't do is decide. That part — knowing which draft serves the actual human on the other side of the screen — is still yours.

Thank you! 🙌