What AI in UX Design Changes, and What It Doesn't

Short answer: AI in UX design makes the slow, mechanical parts of the job faster: sorting research notes, sketching first drafts of screens, writing placeholder copy, catching obvious accessibility misses. It doesn't change the hard part. Someone still has to talk to real users, decide what matters and take responsibility for the result. AI drafts. People decide.

That's our position, and the rest of this article is the detail behind it. It's written for the people who hire design help (founders and product owners, mostly) as much as for designers, because the question buyers really want answered is a blunt one: "if AI can do this, what am I paying a designer for?"

Where AI in UX design actually helps

How AI is changing UX design work is less dramatic than the headlines suggest. A good design process still moves from understanding the problem to testing a solution with real people. (We've written up what a UX design process actually involves if you want the full sequence.) AI hasn't removed any of those stages. It has shortened some of the steps inside them.

Here's where we find it genuinely useful.

Research synthesis

Say you've run eight customer interviews for a booking app. That's hours of transcripts. Before AI tools, a designer would read everything, pull out quotes, cluster them on a board and slowly see patterns form.

Now a transcription tool turns the audio into text, and a language model (ChatGPT or similar) can produce a first-pass grouping of themes in minutes. That's a real saving.

But notice the words first pass. The model will happily tell you "users want more flexibility" because three people used the word "flexible". It won't notice that the one person who hesitated for ten seconds before answering about cancellations was describing the actual reason your bookings drop off. A designer who sat in the interview will. So we use AI to get the mess into rough piles, then a person reads the piles against the raw transcripts and throws half of them out.

Early ideation

Blank-page problems are where AI earns its keep. Ask a model for twenty ways to structure an onboarding flow and you'll get twenty. Most will be ordinary. Two or three might push the team somewhere it wouldn't have gone.

We treat this like a fast, slightly unreliable brainstorming partner. Good for breadth. Bad at knowing which idea suits your users or your budget.

Copy drafts

Designing with "Lorem ipsum" hides problems. A button that says "Submit" fits anywhere; one that says "Book your table for Saturday" might break the layout. AI lets a designer fill screens with realistic draft copy early, so the design gets tested against real-length words.

Draft is the key word again. AI-written microcopy tends to be polite and generic. Error messages in particular need a human pass: "Something went wrong" helps nobody, and a model won't know that your payment provider rejects cards for a specific, explainable reason.

Prototyping

AI UX tools now include generators that turn a text prompt or a rough sketch into a screen layout, and design tools such as Figma have been adding AI features into the editor itself. For a quick clickable prototype to test an idea with five users, this can cut a day's work to an hour.

What you get, though, is the average of every app the model has seen. That's fine for testing whether a concept makes sense. It's a poor starting point for a product that needs to feel like yours, and it tends to ignore the components and rules your team already has. (If you don't have those yet, a design system for startups is where we'd start, before any AI generator.)

Accessibility checks

Automated checkers have existed for years, and AI-assisted ones are getting better at flagging low colour contrast, missing alt text, tiny tap targets and unlabelled form fields. Running them early is cheap and catches the obvious stuff.

They can't tell you whether a screen-reader user can actually complete your checkout, or whether your focus order makes sense. WCAG 2.2, the W3C Recommendation published in October 2023, includes plenty of criteria that need human judgement to assess. Our take on accessibility in UX design goes deeper, but the summary is simple: automated checks are the floor, never the ceiling.

Personalisation

This one is different, because it's AI inside the product rather than AI in the design process. Recommendations, adaptive dashboards, smart defaults, search that understands what people mean.

It can make a product feel noticeably better. It also creates new design problems: users need to understand why they're seeing something and be able to correct it when it's wrong. Designing those moments is its own skill, and we've covered it in designing AI features users trust. If you're planning to build these features, that's work for a design team and an AI development team working together from day one.

Where AI in UX design fails, or gets risky

This is the section most articles rush through. We think it's the most important one.

It can't replace talking to real users

Some tools now offer "synthetic users": AI personas you can interview instead of real people. We don't use them as a substitute for research, and we'd be wary of anyone who does.

A model can only tell you what's plausible based on what it's read. Your users are specific. Imagine a clinic booking app where most patients are booking for an elderly parent, on their phone, during a lunch break. No synthetic persona would tell you that unprompted. Five real conversations would.

Nielsen Norman Group has been making the case for small, frequent usability tests with real people for decades. Nothing about AI changes that.

Generic output

When everyone uses the same tools with similar prompts, products start to look and read alike. The layouts are competent and forgettable. If your product's advantage is that it understands a particular customer better than anyone else, generic design actively works against you.

Hallucinated "insights"

This is the risk we worry about most. Ask a model to summarise survey responses and it may produce a confident finding that isn't supported by the data, or a neat quote that no participant actually said. It reads well. It goes into a slide deck. Someone makes a roadmap decision on it.

Our rule: every insight must trace back to a real source (a transcript line or a recording timestamp). If it can't, it doesn't go in the report.

Bias

Models learn from existing content, and existing content has gaps. Ask for a "typical user" and you'll often get a narrow picture: a certain age and a certain level of tech comfort. If your designer accepts that default, your product quietly fails the people outside it. This matters a lot for businesses selling across very different markets.

Privacy

Pasting interview transcripts or customer data into a public AI tool can mean sending personal information to a third party. Depending on where your users are, that may conflict with GDPR or other data-protection rules, and it may break promises you made to research participants. A careful team strips identifying details before anything goes into a tool with unclear data terms, and asks participants for consent that covers how their recordings will be processed.

Will AI replace UX designers?

No. It will replace some of the tasks junior designers used to spend their time on, and designers who refuse to use it will be slower than those who do.

The core of UX work is judgement: deciding which problem to solve, noticing what users don't say, making trade-offs between business goals and user needs, and owning the consequences. None of that is a generation task. If anything, cheap drafts make the judgement more valuable, because there are now far more options to choose between and far more mediocre ones to reject. (It also helps to be clear on the difference between UI and UX design: AI is much better at producing interface surfaces than at the research and reasoning underneath them.)

What AI UX tools should designers actually use?

Think in categories rather than brands, because the specific products change every few months:

• Transcription and research-synthesis tools for turning interviews into searchable text and rough themes.
• General language models for copy drafts and summarising long documents.
• Layout and prototype generators, including AI features inside mainstream design tools, for quick concept screens.
• Accessibility and usability checkers for automated passes on contrast and page structure.

Pick tools that respect your data, then judge them on one question: does this save time on work we'd do anyway, without lowering the quality bar?

What to ask if you're hiring design help

If you're the one paying, you don't need to know the tools. You do need to know how they're used. A few questions worth asking any design partner:

1. Where in your process do you use AI, and where do you deliberately not?
2. How do you check that research findings are real and not generated?
3. What happens to our customer data if it goes into an AI tool?
4. How many real users will we talk to before we commit to a design?

A good team will have specific answers. A vague "we use AI to work faster" without the second half (how they keep it honest) is a warning sign. If you already have a product and suspect it's drifted, a structured UX audit checklist is a sensible place to start before any redesign.

---

At Amionyx, we use AI where it saves time and keep people on everything that needs judgement. You can see more about how we build or what our UI/UX design work involves. If there's a product you're thinking about, tell us about it and we'll give you an honest view of where AI would help and where it wouldn't.