Vibe-Coded Apps Look the Same: Data From 832 Websites

On this page11 sections

• How We Measured It
• Finding 1: One Design System Under Almost Everything
• Finding 2: Fonts Converge Hardest
• Finding 3: Gradients Are the Tell
• Finding 4: Defaults Ship Unchanged
• The "Default AI Look" Score
• What Did Not Differ
• Why Vibe-Coded Apps Converge
• What to Change First If Your Product Looks Vibe-Coded
• Limitations
• Use This Data

Vibe-coded apps really do look alike, and now there is data on how much. We analysed the homepage design of 409 apps built with Lovable and compared them with 423 websites of Y Combinator startups from the 2025 and 2026 batches. 40% of the Lovable-built apps carried at least three of five default design markers, against 8% of the YC startups, about five times the rate. The same font, the same corner radius, the same gradients and the same component library turn up again and again.

This matters for founders because first impressions are formed from exactly these surface choices, and a product that looks like the default reads as unfinished, whatever it does underneath. Below is what we measured, what we found, what did not differ, and what to change first if your product looks vibe-coded.

Want to know where your own product lands? Our free vibe-coded design checker scores any website on the same five markers.

How We Measured It

We collected every app listed on Made with Lovable, a public showcase of products built with Lovable, and a random sample of 500 companies from the Y Combinator 2025 and 2026 batches using the open yc-oss dataset. On 28 September 2026 we fetched each homepage and its stylesheets once and recorded design signals from the code: the main typeface, whether the site uses shadcn/ui theme tokens and what values they hold, the corner radius, the primary colour, and whether the site uses gradient backgrounds, gradient text or background blur.

409 Lovable-built apps and 423 YC websites could be analysed; the rest were offline, blocked automated requests, or rendered no stylesheet. We report aggregates only, never individual sites, and we treat a difference as meaningful only when it is statistically significant at the 1% level. The aggregate data is available to download.

Finding 1: One Design System Under Almost Everything

86% of the Lovable-built apps use shadcn/ui's theme tokens, the variables that set colours, borders and radius for its components. Among the YC startups the figure is 33%. shadcn/ui is a good component library, which is exactly why every AI coding tool reaches for it; the problem is not the library but that most vibe-coded products never change its defaults.

Finding 2: Fonts Converge Hardest

59% of vibe-coded apps use either Inter or the browser's default font stack as their main typeface. Among YC startups it is 25%. The vibe-coded sample uses 70 distinct main fonts, and the three most common cover 63% of apps; the YC sample of a similar size uses 149, with the top three covering 36%.

Type is the fastest signal a visitor reads, before colour and long before layout. When most products in a category share one typeface, none of them looks like a particular company.

Finding 3: Gradients Are the Tell

78.5% of Lovable-built apps use gradient backgrounds, against 44.7% of YC startups, and 38.9% use purple, violet or indigo gradients specifically, against 14.9%. Half of the vibe-coded apps also use gradient text, the clipped-gradient headline that has become the signature of AI-generated landing pages, against 38% of YC sites.

Finding 4: Defaults Ship Unchanged

A third of vibe-coded apps (33.7%) use a corner radius of exactly 0.5rem, the value the default shadcn/ui theme ships with. Among YC startups it is 10.2%. The radius is one number in one file, and changing it alters the feel of every card, button and input on the page, which makes it the clearest evidence that the defaults are being accepted rather than chosen.

The "Default AI Look" Score

No single marker makes a product look generic; the combination does. We scored every site on five markers: Inter or the default font stack, a 0.5rem radius, gradient text, purple-family gradients, and a primary colour left at the shadcn default or set to purple.

40.1% of the Lovable-built apps carry three or more markers, against 8.3% of the YC startups. At four or more markers the gap is 14.4% against 2.1%.

What Did Not Differ

Two results were not different, and they are worth stating plainly.

• Background blur is everywhere. 81% of vibe-coded apps and 80% of YC sites use backdrop blur, usually on a sticky navigation bar. It is a genuine convention, not a vibe-coding tell.
• Purple is not the primary colour. Among sites that use theme tokens, 16% of vibe-coded apps and 8% of YC sites set a purple primary colour, a difference that is not significant in our sample. The purple shows up in gradients and decoration, not in the brand colour. The most common primary colour in vibe-coded apps is blue (38% against 12%).

Why Vibe-Coded Apps Converge

AI coding tools learn from the same public code: Tailwind tutorials, shadcn/ui examples and starter templates. Asked for "a modern landing page", they return the statistical centre of that material. A 2026 paper on web vibe coding describes the same pattern as design homogenization, noting that generative models tend to reproduce the dominant style conventions of their training data (Shin et al., 2026). Our data puts numbers on it: the tools do not just share components, they share the choices on top of them.

None of this is a reason to avoid vibe coding. It is a reason to treat the first output as a starting point, and to make the few design decisions that the tools cannot make for you.

What to Change First If Your Product Looks Vibe-Coded

In order of impact for the effort involved:

1. Choose a typeface on purpose. Replace Inter or the default stack with a face that suits your market; it is one import and one token, and it changes the whole product.
2. Set your own tokens. Radius, primary colour, borders and spacing live in a single theme file. Decide each one rather than accepting the defaults.
3. Remove decorative gradients and gradient text. Keep colour for meaning: the primary action, state and data.
4. Pick one accent colour tied to your brand, and use it sparingly enough that it means something.
5. Design the states the generator skipped: empty, loading, error and permission states are where vibe-coded products look least finished.

We cover the full process in how to redesign a vibe-coded product without a rewrite, and the wider pattern in why everything designed with AI looks the same. If you would rather have it done, that is exactly what our vibe-coded product redesign service covers: same codebase, a design layer that makes the product look like a company.

Limitations

• The samples are not random draws from all products. Made with Lovable is a self-submitted showcase, and YC startups are funded companies, many of which also use AI coding tools. The comparison is between vibe-coded products and funded startups, not between AI-made and human-made design, and if anything the overlap understates the gap.
• We measured code, not appearance. Signals come from stylesheets and theme tokens, so a site can use Inter and still look distinctive, and layout, imagery and copy were not measured.
• Homepages only, one snapshot. Product screens behind a login may differ, and design trends move quickly.

Use This Data

You are welcome to cite these findings with a link to this page. The aggregate results, including sample sizes and significance tests, are in the downloadable CSV.