Everyone is building the same app: a blue gradient on a dark background

Someone analyzed 210,759 real AI design prompts, and the picture is uncomfortably specific: the default AI product of 2026 is a blue gradient on a dark background, and builders now ask for it "like Linear" more often than "like Apple." This is what I meant when I said AI has no taste — it does not. It averages what it has seen, and now the average is measurable.

Superdesign read the prompts people actually sent to an AI design agent between January and June 2026. Dashboards lead at roughly one in seven projects, ahead of landing pages, pricing, and login screens. Dark mode rose from 27% to 38% of requests in five months. Gradients appear in over 61,000 prompts. Blue is the most-requested color. Put it together and you get the house style of an entire industry.

On this page

• Where the data comes from
• Why everything converges on the same look
• Why builders copy Linear instead of Apple
• What the default aesthetic costs you
• How to opt out of the default

Where the data comes from

The corpus is real product work, not demo reels. Superdesign's analysis of 210,000 real AI design prompts counts 194,000 unique projects, and the leaderboard is unglamorous: dashboards, admin panels, settings screens, the stuff nobody screenshots. Crypto, the category that owns every AI keynote, comes dead last at roughly a twenty-fifth the volume of dashboards. AI design in 2026 is product UI, not art.

Two numbers matter more than the leaderboard. First, 40% of generations are refinements of an earlier draft, not fresh one-shots — the deepest single chain ran 396 generations deep. Second, the median prompt is 806 characters, about 130 words. People who get good results from an AI design agent write a brief, not a one-liner.

Why everything converges on the same look

The convergence is not a mystery. Models produce the statistical median of what they trained on, and the prompt corpus is self-reinforcing: someone generates a screen, screenshots it as inspiration, and feeds it back into the next prompt. The blue gradient on dark is the Ouroboros of design.

It is also the cheapest way to signal "premium." A decade of flat design left everyone craving depth, and gradients are the fastest way to fake it. Designer Fund's AI in Design 2026 report, which surveyed 900 designers across 60+ countries, found 91% now use AI weekly and that designers at early-stage startups ship code at double the rate of their enterprise peers. When everyone can execute, the visual default gets shared too.

Why builders copy Linear instead of Apple

When a prompt names a brand, Linear wins: 229 mentions against Apple's 173, with Stripe, Notion, and Vercel trailing. Those are small numbers, but the ordering is the story. Taste in 2026 is set by the tools builders live in all day, not the consumer giants design schools still teach.

The reason Linear wins is straightforward: its users are the people building the next wave of products. A developer who stares at an interface for eight hours a day quietly adopts it as the implicit definition of "good." The wrinkle is that the median prompt runs 130 words — these builders are copying a feel they cannot fully articulate, which is exactly why the output drifts toward the generic.

What the default aesthetic costs you

The default is not wrong, it is anonymous. When every AI-built product ships the same dark gradient, yours does not look like a competitor — it looks like the same product. Thoughtworks' field guide to designing non-deterministic systems makes the related point well: AI compresses the execution layer, but accessibility, responsive behavior, and design-system coherence still require human judgment. In other words, the blue gradient is cheap precisely because nothing about it is load-bearing.

How to opt out of the default

Opting out is a constraint problem, not a talent problem. Give your agent explicit fences: anti-references that say what the product should not feel like, three personality words, and named reference products with reasons. Write the taste down so it survives sessions — a file beats re-describing it every chat. Then treat generated output as a draft you edit, not a direction you accept.

The parts of your product that feel considered are the parts AI could not have guessed. That is the whole argument for taste as the remaining edge. Everyone converging on the same aesthetic means a distinctive point of view is worth more now, not less. You do not need to out-build anyone. You need to out-decide them — starting by not shipping the blue gradient on dark.