How AI-made do YC startup homepages look?
We checked the homepages of 819 companies from Y Combinator's four 2026 batches with the same free check this site runs. It looks for design habits that make a site look made by AI. 705 loaded and were scored. These are the totals. No company is named or ranked.
- 87.4%show at least one AI-look pattern
- 78.2average score out of 100
- 37.2%use one of the overused fonts
- 12.6%score a clean 100
The habits that cost points
Each pattern found takes points off a score of 100, 8 for most of them. The average homepage shows 2.7 of the 31 patterns. The bars show the share of the 705 homepages that loaded with each one.
Five changes that would lift the most scores
The five most common patterns, each with the fix the free check gives for it.
Overused font
Replace Inter, Roboto, Geist, Fraunces, Plus Jakarta Sans or Space Grotesk with a less common face that suits the brand, and pair it with one body face.
Nested cards
Flatten cards inside cards. Keep the outer section plain and use spacing, headings and dividers to group the items.
Glowing shadow accents
Remove zero-offset coloured box-shadow and text-shadow glows. Use a soft neutral shadow with a vertical offset for elevation.
Kicker / eyebrow label above heading
Delete the small tracked label sitting above each heading. If the words matter, fold them into the heading or the first sentence.
Auto-scrolling marquee
Replace the auto-scrolling marquee with a static row, or scroll it with the page and stop it when not visible. Respect prefers-reduced-motion.
The same few fonts
37.2% of the 705 homepages that loaded (262) use a face the check counts as overused as their main font. Inter alone is on 124.
Gradient text and glows
Two of the easiest habits to spot. Gradient text fades from one colour to another, and a glow is a coloured light around a box or a word. A homepage counts once, wherever on the page it appears.
- 9.9%gradient text
- 28.9%glowing shadows
- 33.8%one or both
How the scores spread
Share of the 705 homepages that loaded in each band.
By batch
Every batch has more than 150 homepages with a score. The averages sit within 4 points of each other, too close to read a trend into.
| Batch | In the list | Scored | Average | Median |
|---|---|---|---|---|
| Winter 2026 | 191 | 159 | 78.9 | 84 |
| Spring 2026 | 188 | 161 | 77.2 | 76 |
| Summer 2026 | 228 | 203 | 80 | 84 |
| Fall 2026 | 212 | 182 | 76.4 | 76 |
How we checked
The list
Every active company with a website in YC's Winter, Spring, Summer and Fall 2026 batches, from a public mirror of the YC directory taken on 6 November 2026. 830 were listed, 11 were left out because they were inactive, acquired or had no website, and 819 were checked.
The check
An automated test for 31 visual patterns that make sites look generated, run on each homepage at desktop and mobile width with the open-source Impeccable detector. It uses the same detector, rules and scoring as the free check on this site.
How the sites were visited
We read the robots.txt of each listed site first and skipped any that refused us. Each site was scanned once, and a second time only after a timeout or a failure on our side, never two at once. Every visit named this report and linked to it in its user agent. One difference from the free check: this browser identified itself as Chrome, where the free check's says it is headless.
What was not scored
114 homepages are counted here and left out of every share: 109 did not finish loading in time, including on a second try, 2 blocked automated visits, 1 refused us in robots.txt, 1 could not be reached and 1 was too large to check.
What a score means
A low score means the patterns are on the page. It does not mean the site was made with AI, or that the product or the team behind it is any less good. Scores are from one date, and a homepage checked today may score differently.
Questions about this report
Does a low score mean a site was made with AI?
Which YC companies were checked?
Can I see one company's score?
What counts as an overused font?
How is the score worked out?
Can I quote these numbers?
Check your own homepage
The same test, free, at desktop and mobile width, with what to change.