The Anatomy of AI Slop

You shipped your MVP. It works. You're proud of it.

Then you see a competitor's app. And your stomach drops.

It looks like yours. Same cards. Same gradient. Same font. Not because anyone copied anyone. Because you both asked an AI to design it — and AI gave you both the same answer.

That's AI slop. Let's take it apart.

What slop actually is

Slop is not bad design. That's the trap. Such a common trap that it actually became word of the year in 2025.

Bad design is easy to spot. Slop is worse: it's average design. Polished, rounded, technically fine. And completely interchangeable. Your product, your competitor's product, a random hackathon demo — one visual language for all of them.

People feel it before they can name it. "It looks... bland?" "Something feels off." What they're sensing is the absence of decisions. Nobody chose this look. A model defaulted to it.

The patterns

Adrian Krebs analyzed HN submissions to show the most common design patterns backed by data. Here are the ones I see most, grouped by what they reveal.

The default palette. Purple-to-blue gradients. Cyan glowing on dark backgrounds. Gradient text on headlines. And the newer reflex: the "tasteful" warm cream background. These aren't choices. They're what the model reaches for when nobody stops it.

The default typography. Inter everywhere. One font for headings, body, buttons, labels. Font sizes so close together there's no hierarchy — just text at slightly different weights. Lately: the oversized italic serif hero, which felt like taste for about three months. Now it's the new uniform.

The default layout. A little uppercase eyebrow label. A huge headline. Three metric numbers. Then a grid of identical cards: rounded icon tile, heading, two lines of copy. Repeat with different section colors. Cards nested inside cards inside cards. The same spacing value between everything, so nothing is grouped and nothing breathes.

The default details. The thick colored border on one side of a rounded card — probably the single most recognizable AI tell. Glassmorphism used as decoration. Buttons that bounce when they appear.

The default words. "Supercharge your workflow." "Enterprise-grade." Sentences chopped by em-dashes — like this — again and again. Copy that lands on a punchy aphorism at the end of every section. Not a feature. A platform. (See? Once is fine. Every section is a tell.)

Why this happens

Here's the part most people miss.

AI doesn't produce slop because it's bad at design. It produces slop because it converges to the average of everything it has seen. No constraints in, average out. Every time.

And prompting harder doesn't fix it. I've watched founders stack design skills, custom system prompts, agents on top of agents — and still get the same landing page. Because none of those layers carry anything specific to their product. "Make it clean and modern" is a wish. The model already thinks the purple gradient is clean and modern.

The real anatomy of slop is simple: generation without decisions. The visual patterns are just symptoms.

How to fix it

Not with better prompts. With constraints that exist before any generation happens.

1. Decide before you generate. Who is this for? What should they feel? What does your brand refuse to look like? If these answers aren't written down, the AI is guessing. Component-shaped guessing.

2. Turn decisions into tokens. Your colors, your type scale, your spacing rhythm — fixed as design tokens, not described in a prompt. Tokens are law. Prompts are suggestions.

3. Give your AI rules, not vibes. A design-system file in the repo. A rules file (CLAUDE.md, Cursor rules) that says: use only these components, these fonts, these radii. Never the side-tab card. Never the gradient text. The model follows constraints beautifully — when they actually exist.

4. Collect references. Five or six products that feel like your brand. Not "make it like Linear" — that's how everything became Linear. References that match your audience, in your corner of the market.

5. Then look at it. With human eyes.The more good taste your eyes have seen before the AI slop started dominating, the better :-) But the final pass is taste. Does this look like us, or like everyone? A machine can flag the purple gradient. It can't tell you who you are.

The honest ending

AI didn't kill design. It killed default design — by making it free and infinite.

What's left is the part that was always the actual job: making decisions. Knowing your audience. Having a point of view and encoding it so consistently that even a machine can follow it.

That part is still human. I'd argue it's more valuable now than ever.

Because when everything can look polished, the only thing left to compete on is looking like you.