05/21/2026
Our developer, Amanda gave three AI site builders a real assignment: build an archive of abandoned places, a site meant to log interesting locations on a map so people could find and explore them. Oh, and make it dark, gothic, and spooky. 👻
Amanda documented the good, the bad, and the ugly and presented it to the team. The results were instructive, occasionally frustrating, and ended with a cat photo. As all good research does ⬇️
---
🥇 Lovable.dev— 1st Place
Given the prompt, Lovable actually delivered. It worked conversationally, following directions the way you'd expect from a well-prompted ChatGPT session. It built a functioning database, rendered working maps, and connected to GitHub with minimal effort.
The design needed work, Amanda was clear about that. But for a site meant to log and display locations, the core functionality was there. The limitation worth noting: it only builds React/Vite apps. Ask for WordPress as a CMS and it declines. A real constraint depending on the project.
The honest take: closest to the actual vision. Not polished, but functional in the ways that mattered for this specific brief.
🥈 Webflow— 2nd Place
Webflow let Amanda input brand details upfront, colors, logo, fonts, which is a thoughtful starting point. It generated a style guide and offered component options worth choosing between.
For an abandoned places archive, though, the map was non-negotiable. Webflow didn't attempt one. Amanda asked the AI assistant directly. It returned zero maps.
The limitation: tablet and mobile styles were broken out of the box, and CMS collections weren't set up. The structural work would still need to happen manually.
The honest take: good bones for layout and components, genuinely useful as a starting point, but it couldn't deliver on the most critical piece of this particular project.
🥉 Wix— 3rd Place
Wix asked questions before building, about the site's purpose, its target audience, and then generated a dashboard and site design from those answers. As an approach, that's reasonable.
It produced a map, which put it ahead of Webflow on that front. The map wasn't connected to the listings, which put it back behind where it needed to be. The heading font choices were notable. The backend ran slowly throughout. And without prior familiarity with the platform, the learning curve added friction that the other tools didn't.
The honest take: some components work as a starting point. For a project with specific functional requirements, it didn't clear the bar.
To sum it all up, the tool matters less than what you feed it.
Every site Amanda built looked and performed better when given real brand direction: actual colors, actual fonts, actual design context. Feed these tools a blank slate and they hand one back.
The abandoned places archive had a strong concept. What it didn't have yet was a brand. That gap showed up in every result across every platform. AI builds faster when it has something real to work from. That's where the design thinking still has to happen first.
