We rebuilt this site's homepage twice. The first attempt changed the palette, swapped the typeface and stripped the decoration, and it looked exactly the same. Not similar. The same.
The second attempt worked, and the difference tells you what the problem actually is.
Count the tells on your own site
Open the console on your homepage and run this:
const c = s => document.querySelectorAll(s).length;
console.table({
pills: c('[class*="rounded-full"]'),
hoverLifts: c('[class*="hover:-translate-y"]'),
glassRings: c('[class*="ring-white/"]'),
blurGlows: c('[class*="blur-2xl"], [class*="blur-3xl"]'),
});
Ours returned 28 pills, 16 hover lifts, 12 glass rings and 2 blur glows. Plus six decorative sparkle icons, a gradient-clipped phrase in the headline, and three floating white chips over the hero photograph.
Every one of those is a defensible individual choice. Together they are a fingerprint, because language models reach for the same defaults that every other language model reaches for. The result is not bad design. It is average design, arrived at very quickly.
The first fix failed, and that is the useful part
We removed the sparkles, the glows and the floating chips. Pills went from 28 to 6, hover lifts from 16 to 4. We moved the ground from navy to a true neutral near-black, so the brand blue read as a chosen accent rather than the whole room. We replaced Plus Jakarta Sans, which, paired with Inter, is the default typography of this entire genre, with a grotesque that has actual character.
Measurably different. Visually, the owner's verdict was: "I don't see any major difference."
He was right. We had changed the paint and left the architecture. The hero was still text-left, image-right. The services were still four identical cards in a row. Every section was still a centred container at the same width with the same vertical padding. That grammar is the tell, not the ornament on top of it.
What actually changed it
Three structural moves, none of which are about styling:
The hero stopped being two columns. The photograph became the page: full-bleed, viewport height, headline set over it behind a directional scrim that is dark where the type sits and clears where the image is interesting. The stats moved out onto their own rule beneath.
The card grid became alternating rows. Four full-width bands, each with a large ghosted index numeral, the image bleeding to the outer edge and flipping sides row to row. No cards, no borders, no equal-height grid.
One section breaks the container entirely. A statement band at 48px that escapes the max-width every other section respects.
The page is now roughly 30% taller and reads as deliberate. Same copy, same colours, same brand.
Why subtraction alone does not work
Removing decoration makes a page cleaner. It does not make it distinctive, because the thing being recognised was never the decoration.
A four-card feature grid under a centred heading is the shape of a SaaS marketing page. Readers have seen it thousands of times. Strip every sparkle and gradient from it and you still have a four-card feature grid: you have made a familiar thing tidier.
Layout carries more identity than surface treatment, which is unintuitive because surface treatment is what we notice consciously. The centred container repeated eight times is doing more to make your site feel generic than the gradient on your button.
If you only do three things
- Make the hero not be two columns. Full-bleed image, or type-only at large size. This is the single highest-impact change available.
- Break the container once. One full-bleed band anywhere on the page disrupts the rhythm that makes generated layouts feel interchangeable.
- Change the display typeface. Inter and Plus Jakarta Sans are the default pairing of this genre. Almost any characterful alternative reads as a decision rather than a default.
Deleting decoration is worth doing and it is not sufficient. Do it after the structure is right, or you will do what we did and spend a round discovering your site looks identical.
The honest limitation
None of this makes a site good. It makes it yours. A distinctive layout with weak copy and invented statistics is still a weak site, and we have written elsewhere about why unverifiable numbers cost more trust than they buy.
There is also a real cost to breaking conventions. Familiar patterns are familiar because they work: visitors know where to look. A fully custom layout that nobody can navigate is worse than a template that converts. The goal is not novelty; it is not being immediately recognisable as something assembled from defaults.
Start with the hero. If a visitor's first screen looks like every other first screen they saw this week, nothing further down the page will fix it.