What makes a website feel designed, not generated

Put a page a designer spent a fortnight on next to one an AI site builder produced in four minutes, and most people can pick which is which straight away. Ask them how they knew, and they struggle.

That gap is worth paying attention to. Everyone can feel the difference between a site that was designed and one that was assembled. Most people can't name it. And now that a passable page takes minutes to generate, that difference is doing more and more of the work of deciding who looks credible.

This post is our attempt to name it.

The feeling has a cause

What people call taste is mostly a long run of small decisions that all lean the same way. Which typeface, and how few sizes of it. How much space sits between things, and whether that spacing follows a rule. How many colours there are, and whether each one has a job.

None of these is visible on its own. Nobody leaves a site thinking "the spacing was inconsistent". They leave thinking it felt cheap, or rushed, or slightly off. Then they apply that feeling to the business.

That's the commercial point. Visitors read the surface as evidence about the substance. If the site looks like nobody made decisions about it, people assume nobody made careful decisions about the product either. They're often right, because the same lack of attention rarely stops at the homepage.

What generated design looks like

"AI slop" has become the shorthand for this, but the tools aren't really the cause. A rushed human with a template produces exactly the same thing. What the tools changed is that the default output now costs nothing, so the defaults are everywhere, and people have learned to spot them.

These are the tells we see most often.

Typography

The default font of whatever tool built it, or the same two or three fonts every generator reaches for.
Lots of weights and sizes, with no clear ladder from most important to least.
Headings that are big but don't actually organise the page.

Layout

A centred hero, then three feature cards with an icon each, then a logo wall, then three testimonials, then a call to action. Every section the same width and the same height.
Everything symmetrical, so nothing is more important than anything else.

Colour

A purple to blue gradient, or a soft glow behind the hero, doing no work beyond looking modern.
A dozen near-identical greys that crept in one component at a time.

Copy

Sentences that sound confident and say nothing. "Unlock your potential." "Built for the way you work."
Everything in threes. Three adjectives, three benefits, three columns.
Copy that could describe any company in the category, with the name swapped.

Detail

Six slightly different button styles.
Spacing that's nearly consistent, which looks worse than spacing that's obviously varied on purpose.
Icons from three different sets.

What considered design does instead

Considered design is mostly subtraction. Where generated design adds another gradient, another weight, another column, a designer keeps removing things until only the decisions that earn their place are left.

In practice that looks like:

One typeface, or two with clearly separate jobs, used at a small fixed set of sizes.
A real hierarchy. The most important thing on the page is clearly the most prominent, and your eye knows where to go next.
A short palette. A handful of colours, each with a role, all with enough contrast to read.
Spacing from a system, so the rhythm holds from the top of the page to the bottom.
Copy that could only be about this business, written by someone who knows it. Names, numbers, places, the specific thing you do that others don't.

That last one matters as much as the visual points. We've seen beautifully designed pages undone by a headline that could belong to anyone.

Count things instead of arguing about taste

Design feedback goes nowhere when it's about feelings. "Make it pop" and "it needs to feel more premium" can't be acted on, and they can't be checked afterwards either.

So when we review a site, we count. Counts are hard to argue with, and the number itself is usually the finding.

Greys. Inspect the page and list every grey in use. More than five or six usually means nobody chose them.
Button styles. Count every distinct button treatment. Most sites need two or three.
Type sizes. List every font size on the page. A clear system rarely needs more than six or seven.
Calls to action above the fold. If there are three, none of them is the main one.
Contrast. Check body text against its background. WCAG's AA level asks for a ratio of at least 4.5 to 1 for normal text. Light grey on white usually fails, and it's one of the most common tells of a page styled for screenshots rather than reading.
Copy swap test. Replace your company name in the hero with a competitor's. If the sentence still works, it isn't saying anything about you.

"Fourteen greys, and the body text fails contrast" is a finding someone can fix this afternoon. "Elevate the visual language" is not.

Five fixes, roughly in order

You don't need a redesign for most of this. These are in the order we'd tackle them, cheapest first.

Rewrite the generic copy. Start with the hero and the first section. Swap every sentence that would survive the copy swap test for one with a specific fact in it. This usually does more than any visual change.
Collapse the greys and the buttons. Pick the smallest set that still works, and replace everything else. On a site built with design tokens this takes an hour or two. On one that wasn't, it's how you find out.
Fix the type scale. Choose one family and a fixed set of sizes. Remove the extra weights.
Delete decoration that carries no meaning. Gradients, glows, floating shapes and stock illustrations, unless they tell the visitor something. Restraint reads as confidence.
Choose the one thing each page is for. Make that the most prominent element, and demote everything competing with it.

The cheap fix and the real fix

Cheap fix: do steps one to three directly in your current site builder. You'll get most of the benefit in a week.

Real fix: set up a small design system, even a basic one. Named colours, a type scale, a spacing scale, a few components. Then new pages inherit the decisions instead of drifting from them. This is the difference between fixing the site once and keeping it fixed.

When it is worth a redesign

If the counts come back clean and the copy is specific, your site probably doesn't have a taste problem, whatever it looks like next to a competitor's latest launch.

A redesign makes sense when the structure can't hold what you now need to say. A new offer, a new audience, or a site so tangled that every fix breaks something else. It doesn't make sense just because the site looks a few years old. Looking dated costs you much less than looking careless.

Common questions

Isn't taste just subjective?

Less than it feels. The failures are remarkably consistent: the same fonts, the same grids, the same gradients, the same empty sentences. Once you can name them, spotting them stops being a matter of opinion. Taste comes in when you're choosing between good options, and that's a much smaller argument.

Does AI-generated design always look like this?

No. The tools default to these patterns because defaults are effortless, but someone with good judgement can use the same tools and remove the tells. The problem is skipping the decisions, not the tool that made them easy to skip. We use AI throughout our own work. We just don't let it make the calls.

Does any of this affect conversion?

Indirectly, and a lot. A visitor who doesn't trust the surface is less likely to read far enough to be convinced. Craft buys you the benefit of the doubt, and specific copy gives them a reason to stay.

Where should I start if I only have a day?

The copy swap test on your homepage hero, then the grey and button counts. Those three checks will tell you whether you have a design problem, a writing problem or neither.

If you'd like a second pair of eyes on your site, book a 30-minute call.