The AI-generated web,
mapped.

Millions of websites are now designed by machines - same fonts, same gradients, same layouts. We measure how fast the generated look is spreading, and we map what it looks like.

497,865pages detected as AI-looking
31,681websites mapped on the atlas
68design signals in the detector
growth of the AI look in a year

We scan the whole web for machine-made design

Slopymap continuously scans the web - tens of millions of domains, month after month. A transparent detector reads each page's public HTML and CSS and scores it against 68 design signals. Flagged pages are screenshotted, classified, and clustered with their visual siblings - so the same template surfaces as one island of near-identical websites on the atlas.

Why we do it

We believe every design should be unique - it should carry something of the person or the brand behind it. The universally agreed-upon look of AI-generated sites is not helping creativity: it mass-produces slightly different copies of the same page. That sameness costs something real - user engagement, and trust in the brands behind the sites. We map it so people can see it.

Blurred composition derived from a cluster of AI-generated websites

Every generated site belongs to a family

The atlas groups twenty thousand screenshots by pure visual similarity - and the generated ones collapse into a handful of recurring looks. These are four of the biggest islands, straight from the map.

Open the atlas
Mosaic of near-identical purple-gradient websites

Purple Reign

Indigo-to-violet hero gradients, glass cards, one centered CTA. The default SaaS look.

Mosaic of dark websites with warm orange accents

Midnight Embers

Near-black pages lit by warm orange accents - the classic generated dark mode.

Mosaic of corporate blue websites

Boardroom Blue

Trust-me blues, dashboard screenshots and three feature columns that mean business.

Mosaic of pink pastel e-commerce websites

Cotton-Candy Commerce

Pinkish pastel gradients and soft shadows for shops that want to be your friend.

The most AI-looking pages we found

Out of 497,865 pages flagged in a single Common Crawl snapshot, these scored highest - each one tripped thirty-plus detector signals. Click a row to see the full page.

What do we detect?

The same pairing on site after site: Space Grotesk headings, Manrope body, JetBrains Mono numbers.

Em-dashes and middots everywhere: "Not just X — Y" taglines, features separated by dots.

Tiny uppercase labels above every heading: WHY US, HOW IT WORKS, FAQ.

Frosted-glass panels floating on purple-to-pink gradients.

The key word of the headline, clipped to the same purple-pink gradient.

lovable.app and vercel.app domains - with the "Edit with Lovable" badge left in the corner.

··· and 62 more signals
Abstract illustration of the selected detector signal

What we actually do

Open method, run monthly - the same domains, the same signals, humans in the loop.

01

Scan

Every month we read 61 million domains from the newest Common Crawl snapshot.

02

Detect

Sixty-plus HTML and CSS signals score each page - markers of hand-built sites subtract, keeping the verdict conservative.

03

Map

Screenshots are embedded by a vision model and laid out by similarity - that map is the Slop Atlas.

04

Verify

People swipe through screenshot decks to ground-truth the detector - 20,090 labels and counting.

Help us label the data

The detector is only as good as its ground truth. Try it right here - look at the site, go with your first impression. The real deck has 35 cards per round, a two-minute guide first, and no sign-up beyond a name and email.

Screenshot of auctera.io
Screenshot of electree.cz
Screenshot of 1bot.sk

Want to participate?

Join labeling now

See the islands for yourself

Twenty thousand websites on one map. The generated ones are easy to spot - that is the point.