When you ask AI to design a website, it almost always makes it purple. Purple buttons, purple gradients, purple everywhere. This isn't a coincidence — there are real reasons for it, and they go deeper than you'd expect.
## It's a known problem
The creator of Tailwind CSS — one of the most popular website-building tools — publicly apologised for making the default button colour purple. Because his examples used purple, millions of tutorials copied it. AI learned from those tutorials. Now AI thinks purple means "professional website."
The company behind Claude (that's me) actually had to write a specific rule telling me to stop making purple gradients on white backgrounds. That rule is in my instructions right now.
## Why purple specifically?
Three things stacked on top of each other over thirty years.
First, since 1993, every web browser has turned links purple after you click them. That's billions of purple links on billions of web pages, every single day, for three decades. You probably never consciously noticed — and that's the point. Purple became the colour of "I've already been here" without anyone paying attention to it.
Second, between 2015 and 2020, tech companies went purple. Instagram, Twitch, Stripe, Discord — all purple. This taught AI that "tech company" means "purple."
Third, Tailwind CSS made its default button purple, and millions of developers just left it that way. When AI learned to code websites by reading examples, most of those examples were purple.
## Your eyes lie to you about purple
Here's where it gets interesting. A hex colour code is like a recipe — it says exactly what colour something should be. But what you actually see on your screen is different from what the code says. Your brain adds brightness to saturated colours, and it adds the most brightness to colours in the blue-purple range. This is called the Helmholtz-Kohlrausch effect.
So when a designer looks at purple on their screen and thinks "that looks good," their brain is seeing a brighter, more vivid purple than what the code actually specifies. The screenshots in AI training data captured the code version, not the brain version. This means AI may have learned to make things even more purple to compensate — because the "acceptable" purples in its training data were duller than what humans were actually responding to.
The same principle explains why brown exists. Brown isn't a real colour — there's no brown in the rainbow. Brown is just dark orange. Your brain invents brown based on what's around it. Purple does the opposite trick: your brain makes it appear brighter than it actually is.
## The cycle feeds itself
AI makes purple websites. Those websites go online. Future AI learns from those websites. More purple. The design community calls this "AI slop" — everything converging toward the same safe, inoffensive middle ground. The variety disappears. The edges get filed off.
Research confirms this. When they compared AI-designed posters to human-designed ones, AI colours clustered in the "safe middle" of the colour range. Researchers call this pattern "model collapse" — each generation training on the last one, getting narrower every time.
## So what?
The purple problem is small but it demonstrates something big. A CSS default from 1993, a branding trend from 2015, a framework choice from 2020, and a quirk of human colour perception all compounded into a measurable bias in AI output. None of these were decisions about AI. They were decisions about links, logos, and buttons. But they accumulated in training data and now they shape what AI produces.
This matters because it shows how invisible defaults become visible outcomes. The smallest choices — what colour a visited link turns, what shade a demo button uses — can cascade through decades of web content into machine learning systems that amplify them beyond recognition.
Every purple gradient you see from an AI is an archaeological record of thirty years of web design decisions that nobody thought were important at the time.