Is Vibe Coding Polyester? Testing 4 AI Website Builders

There's a word doing the rounds with Gen Z right now: polyester. It started in video-editing circles to describe a specific style of edit, flashy, frantic, cut together fast enough that the seams show. It's since spread to mean anything that looks impressive for about three seconds before you notice it's cheap. Mass-produced. Trying hard and showing it.

I've been testing out AI website builders recently on a real, live project, not a demo, not a toy brief, an actual charity site I was already committed to delivering. And "polyester" is the word that kept coming to mind.

The test

Outside Thinkers is a London-based charity supporting care experienced people into and within the workplace, founded by Linda Davidson. I was building their site for real, so instead of running a synthetic benchmark I used the actual project brief, the actual proposal, the actual brand guidelines, navy, cream, red and teal, Nunito Sans throughout, and a site structure map I'd already built out myself. I collated that source material into whatever format each tool actually accepted, PDF, Word, PowerPoint, plain text, and fed the same core prompt into four different tools: Google Stitch, Bolt.new, Lovable, and Duda Vibe. Same three-journey site structure, Care-Experienced Talent, Partner & Pledge, Get Involved/Donate, every time, same strapline, same brand rules, same instruction to prioritise AEO and search visibility over visual polish. Only the platform, and the file format it needed, changed.

Then, separately, I built the site the way I've built Duda sites since around 2015: structured editor, AI Copilot for the scaffolding, then my own hands on the design.

What came back

Google Stitch struggled to cope with documents. It only takes text prompts and images, no PDF, no Word, no PowerPoint, so it never actually saw the brief, only whatever I could manually compress into a prompt box. Weakest of the four, and structurally so, not just aesthetically.

Bolt.new didn't produce anything. Same brief, same brand guide, same logo, five minutes in and there was nothing usable to show for it. On paper Bolt's document support is stronger than Stitch's, it takes PDF and Word directly through the chatbox rather than being limited to images. That didn't translate into a working site. It just failed.

Lovable was the strongest of the three pure AI tools, and it wasn't close. Full document support, including the PowerPoint proposal as-is, and a working first pass in around six minutes. If you're going to hand a brief to an AI and walk away, this is where I'd point you.

Duda Vibe produced the best output of the four, by a clear margin. Closest to the actual brand, closest to the actual structure I'd briefed. It also burned through more than 150 AI credits doing it, my entire monthly agency allowance, gone on a single test pass.

Bolt failed in five minutes. Lovable succeeded in six. Stitch and Vibe both took somewhere in the 10 to 30 minute range.

At a glance

• Google Stitch : no document support, weakest structurally, roughly 10 to 30 minutes
• Bolt.new : takes documents on paper, produced nothing, five minutes, failed outright
• Lovable : full document support, best of the three AI-only tools, around six minutes
• Duda Vibe : best output of the four, 150+ AI credits, an entire month's agency allowance in one pass
• Duda Editor 2.0 + Copilot : what shipped, structural and branding pass saved roughly three to four hours against building the same scaffold by hand

Three things stood out while I was prompting. First, explaining to a machine something I could just move by hand in an editor, a whole paragraph of prompt to describe what would be a five-second drag. Second, the time lost coaxing a result into shape, asking for one thing and getting back something else entirely. Third, straightforward misreadings of the brief, a prompt taken too literally, or a detail that mattered dropped altogether.

Why I still built it by hand

Here's the part that surprised me least: prompting alone, even into the best of these tools, felt like generating polyester. Fast, flashy, structurally present, but you can see the seams the moment you look closely. Spacing that's almost right. Copy that's almost the right tone. A brand palette applied close enough that you'd only catch the drift if you'd written the brand guidelines yourself.

I've been building on Duda since roughly 2015. I'm faster by hand than I am prompting, because I already know exactly where I want a section to sit and what three words the CTA button should say. So for the actual Outside Thinkers handoff, I used Duda's Editor 2.0 with the internal Copilot doing the structural first pass, then finished it myself. Getting the structure, basic theme and branding in place that way saved me roughly three to four hours against doing the same scaffold manually. My agency plan gives me 150 AI credits a month, the same monthly pool the Vibe test alone used up entirely, and building the whole real site with Editor 2.0 and Copilot together left credits spare. Fully editable, on-brand without correction, and ready to hand to a charity team who need to update it themselves without touching code.

That's not a knock on vibe coding. It's a genuinely useful way to get from nothing to something in twenty minutes. But "something" and "a site you can hand to a charity board and a funder" are different bars, and right now, for a professional handoff, the gap between them is still a human designer's time.

Where this landed

Duda has kindly agreed to sponsor Outside Thinkers' hosting and security for two years, a genuinely useful outcome for the charity, and part of why this project was worth doing properly rather than fast. The finished, fully editable site is heading to handoff shortly. I'll drop the link in here once it's live.

The takeaway

If you're an agency deciding between prompting a site into existence and building it properly, the honest answer is: it depends what you're optimising for. For a quick internal tool or a throwaway landing page, vibe coding is genuinely useful, Lovable especially. For anything a client has to live with, edit themselves, and put their name to, hand-finished still wins. Not because the AI is bad. Because polyester photographs fine. It just doesn't wear well.