??% of yc 2026 homepages are slop.
yc writes every startup a $500,000 cheque. 78% of their homepages still trip the slop scanner.
and it is not only yc 2026. some very big names trip it too, including two of the biggest website builders. the stamp is the scanner's score out of 100. none of these are in the 78%.
ugly is not slop. same scanner, same day, also outside the yc 2026 set.
find your startup
every yc 2026 company and 150 from winter 2022, with the score and the tells. we print the receipt and stamp it with one of six levels.
| level | score | yc 2026 | what it means |
|---|
all 748, by tier
every yc 2026 company we scanned, as its logo, stacked by slop score. hover for the name, click for the receipt.
same 27 checks, four groups. yc's winter 2022 batch is the last one that launched before chatgpt, and 68% of those trip a tier too. so the template problem is older than the chatbot. it just got 10 points worse. the builders are the tools that sell you the template. the control group is big tech and sites obviously made by a person who cared, including craigslist.
the tells
what trips, and how often. the top one is not purple. it is the little label above every heading in capital letters. the grey number is the same tell on the winter 2022 batch, so you can see what four years did.
the font
the scanner reads the computed font of the first headline. of the 705 homepages where it could, 25% set it in inter, geist or space grotesk, the three faces it calls "ai default".
they sell you the template. their own homepage scores 42.
every ai website builder we could think of, scanned on its own homepage. a few score zero because their homepage is an app shell with almost no marketing page to judge, so read a zero here as "nothing to see", not "handcrafted".
| site | score | tier | what fired |
|---|
the control group
sites nobody would call vibe coded. some of them are ugly. ugly is not slop. craigslist and berkshire hathaway score 3. tailwindcss.com, the site that ships the defaults everyone copies, scores heavy. so does shopify. the scanner has no idea who you are.
| site | score | tier | what fired |
|---|
two homepages from the builders and the control group. tap the one the scanner scored higher. keyboard: 1 or 2. the scanner does not blink.
batch by batch
four batches in one year. same checks.
by industry
| industry | sites | mean | heavy |
|---|
hall of craft
the hall of craft. zero tells. we checked twice. named, because this is a compliment.
| company | batch | score | grade |
|---|
the heavy end
the twelve heaviest yc 2026 homepages, names left out of this list. they are in the search box and the csv, because the data is public, but a leaderboard of shame is not the point. if one is yours, dm me for the fix prompt.
| batch | score | grade | what fired |
|---|
we scanned all 748 of them. 27 design tells, one headless browser, no model judging anything. only rules you can argue with.
the 78% covers yc companies only: every company listed in the winter, spring, summer and fall 2026 batches of yc's public directory, pulled through the yc-oss api on 23 oct 2026. that is 779 companies. 31 were blocked, broken or had too little page to judge, which leaves 748 scored: 187 winter, 186 spring, 221 summer, 154 fall. 6 of them are now marked inactive or acquired and are still counted. the fall 2026 batch only started this month, so some of its pages may be early: without it the figure is 77% of 594. separately, for comparison only: 150 random active companies from winter 2022, 39 ai website builders and 41 control sites. those are never mixed into the 78%. 1009 urls were attempted in total.
scanner: slop-detect 0.8 (MIT, by Ravindra Kumar), definitions 2026.09, run locally in headless chromium with the design and copy axes. it checks computed styles and dom structure for 27 patterns with fixed weights (ai-default fonts +8, filled indigo ctas +8, gradient text +6, eyebrow pill +5, and so on), sums them, and bins the total into clean, mild and heavy. its author reports 5 to 10 percent false positives against a hand-labelled set.
what the score is not: proof that a site was built with ai, or that nobody designed it. it is a count of defaults. a page can be generic by hand. a page can be built in lovable and score 4. the whole pattern list and weights are public in the scanner repo. the $500,000 is yc's standard deal: $125,000 for 7% plus a $375,000 uncapped safe.