Yes, if people use your product every day to get work done. Nielsen Norman Group tested about 19 AI design tools and found they often get spacing, grouping, hierarchy and contrast wrong. The result looks good in a demo and gets worse with use. Why that happens, and how to get better results if you build with AI anyway.

Why would I need a designer?

It's a fair question now. Founders ask it, but so do product managers, and engineers who can suddenly ship a whole interface on their own. I understand where it comes from. Tools like Lovable, v0 or Figma Make turn a sentence into a clickable screen in about a minute. You get tidy cards, a dashboard, a nice header. It looks like a real product, often better than what the team had before.

The trouble starts when someone tries to use it.

Good from afar

In 2025, Huei-Hsin Wang and Megan Brown at Nielsen Norman Group tested about 19 AI design tools, from v0 and Lovable to ChatGPT and Claude, on real design tasks. They compared the results with work by their own designers and called the write-up Good from afar, but far from good [3]. The tools followed instructions, but they often got spacing, grouping, hierarchy and contrast wrong, and the designs came out looking “flat and interchangeable”.

To be fair, the gap is closing on simple things. Jakob Nielsen wrote about a Google study in which people compared generated interfaces with pages made by expert designers [2]. The designers still won, but only 56% to 43%. Those were simple, one-off pages, though, and Nielsen himself says the results don’t cover complex products.

Most startups are building the complex kind: a product people open every day to get work done. That's where AI design falls short, because the problems don't show up in a demo. It looks finished in a screenshot, and it gets worse the longer someone uses it.

It looks finished in a screenshot, and it gets worse the longer someone uses it.

Why AI can't do it on its own

It designs the average. AI builds what a typical app of that kind looks like, which is why so many AI-made products look alike: the same cards, the same dashboards, the same layout. That's fine for a first draft. It's a guess about a typical user, though, and your users are specific people with a specific job to do.

It adds and never removes. Ask for a settings page and you get every setting you mentioned, all visible at once. Deciding what to leave out is most of design, and AI doesn't do it unless you ask. That's where the overcrowded screens come from. At a company I worked with this year, a big part of my design work was taking things away: simplifying flows, merging features and cutting the ones that were too complicated. The feedback on those changes was good, and no prompt would have suggested them.

Deciding what to leave out is most of design, and AI doesn't do it unless you ask.

It doesn't know who's using it. AI can't know, for example, that your user is filling in an expense claim on their phone in the back of a taxi, with patchy signal and two minutes before they arrive. It designs for a user with no situation at all.

It can't test its own work. It never watches someone get stuck. Accessibility is a good example. Ammar Ahmed and colleagues tested ChatGPT on building web pages that meet accessibility standards (WCAG) and found that its default code often didn’t [1]. It fixed simple problems well when asked; the harder ones needed a person checking and several more rounds.

Why the same tools work better for designers

The Nielsen Norman Group study found one more thing: with detailed prompts, the tools produced designs close to what a human designer would make [3]. Attaching an existing design file worked even better. But writing that detailed prompt means you already know the hierarchy, the flow and the edge cases. That knowledge is the design work, and the tool doesn’t supply it.

Without it, two things go wrong. You can't judge what comes back: if it looks polished, it seems done, while a designer sees the button nobody will find. And you don't know what to ask for. Nobody prompts "what does this screen look like when it's empty?" or "what happens when the name is 40 characters long?" unless they've watched real people run into it.

I design with AI myself now, after ten years in graphic design and several more in user research. It saves me a lot of time on the drawing. It doesn't tell me what to draw though, my experience and studies did.

Writing that detailed prompt means you already know the hierarchy, the flow and the edge cases.

If you're building with AI anyway

Most teams will, and that's fine. A few habits make the result a lot better:

  1. Write down who it's for before you prompt. One user, one situation, the one thing they're trying to get done.

  2. Ask the AI what it would remove. Then remove more.

  3. Watch a few real users try it. Give them a task and say nothing. An hour of that tells you more than another ten prompts.

  4. Check the screens nobody demos: empty states, error messages, long names, and the whole thing on a phone.

  5. Get a designer to review it before you build it properly. A day of review costs less than a redesign after launch.

Use AI to get to a first version quickly. Then spend the time you saved on the questions it can't answer for you: who uses this, and what can go.

Sources

  1. Ahmed, A., Fresco, M., Forsberg, F. & Grotli, H., arXiv (2025). From code to compliance: Assessing ChatGPT's utility in designing an accessible webpage – a case study.

  2. Nielsen, J., Jakob Nielsen on UX (2025). Generative UI from Gemini 3 Pro.

  3. Wang, H.-H. & Brown, M., Nielsen Norman Group (2025). Good from afar, but far from good: AI prototyping in real design contexts.

Edita Dermontaité

Edita Dermontaité

Hands-on CPO and product leader who still builds (with AI)

I’ve often been the first: the first product hire at Leadcamp and Bizzy, where I led product as CPO for five years, and the first user researcher at Lighthouse, where I spent two years building the research function.


For the past year I’ve worked as a fractional CPO and hands-on product leader, helping B2B SaaS teams between traction and scale connect what engineering builds to what brings in revenue.


I started out as a graphic designer, so I still design and test things myself, these days mostly with AI: in Claude and Lovable, or straight in code with Cursor.


More about me →

Working through this at your company?

Book a free 30-minute call, or a coffee in Ghent. We’ll look at what’s slowing your product down and what to do about it. I work with teams fractionally or hands-on as an embedded product leader, 2–5 days a week, depending on your needs and my availability.

Product leadership for B2B SaaS between traction and scale, from a hands-on CPO and product leader who still builds (with AI). I take product decisions off the founder’s shoulders and get product, engineering and sales moving in one direction: revenue.

Edita Dermontaité speaking to an audience

Copyright © 2026 New North BV

Copyright © 2026 New North BV

Made in Belgium with ❤️

Made in Belgium with ❤️