Why AI-generated websites all look the same

You've probably noticed it. Big centred headline, a purple-to-blue gradient, rounded cards, a floating screenshot of a dashboard, three neat feature columns underneath. Different companies, different tools, same site. This isn't a mystery or a conspiracy — there are two well-understood causes, and knowing them tells you exactly which parts of an AI-built site you need to take control of.

First, let's describe the thing precisely

The house style goes roughly like this: an oversized hero headline with a short subheading under it, a gradient background running from purple to blue, buttons with softly rounded corners, a mockup floating at a slight angle, then evenly spaced feature sections made of rounded cards with a small icon at the top of each. Scroll further and you get a testimonial row, a pricing table with the middle plan highlighted, and an FAQ accordion.

That's not one tool's output. That's the output of most of them. And people notice — the same way they've started noticing the rhythm of AI-written copy, which we covered in what people actually complain about with AI writing tools. The complaint is real. The interesting bit is why it happens.

Cause one: everyone is building from the same box of Lego

Modern websites are rarely drawn from scratch. They're assembled from component libraries — pre-built buttons, cards, form inputs, navigation bars, spacing scales — so a developer doesn't have to decide, for the thousandth time, how much padding goes inside a button.

Two of these have become close to universal: Tailwind CSS, a system of utility classes with a fixed set of sizes, spacings and colours, and shadcn/ui, a set of ready-made components built on top of it. They're popular for good reasons. They're well made, they're accessible, they behave properly on phones, and they save enormous amounts of time.

Here's the consequence. Many different AI site generators — built by different companies, running on different models — reach for the same building blocks, because those are the blocks that appear most often in the code they learned from and the ones most likely to produce something that works first time. So the buttons are the same shape. The cards have the same corner radius. The gap between sections is the same number of pixels. The colour palette comes off the same default swatch. Change the generator and you change almost nothing, because the parts are shared.

This is a design-system effect, not a flaw in any single product. A thousand kitchens fitted from the same supplier's catalogue will look related, however different the people who ordered them are.

Cause two: a machine that averages produces the average

The deeper cause sits above the visual layer, in structure and content.

When a generator decides what sections your homepage needs, in what order, with what headings, it's drawing on patterns across an enormous number of existing websites. Hero, then social proof, then three benefits, then how it works, then pricing, then FAQ — because that is the most common arrangement. Ask for a headline and you get a headline shaped like the most common headline. Ask for a benefit and you get the most commonly stated benefit for that kind of business.

So the output is the statistical average of websites. And an average, by definition, is the thing nothing is distinct from. Every individual decision is defensible. Collectively, they produce a site that could belong to any of four hundred other firms in your trade.

This is also why the copy reads flat. "Quality workmanship you can trust." "Committed to excellence since 2010." Nobody would write those sentences about their own business unprompted — they're the arithmetic mean of what plumbers' websites say.

Which decisions come from averages, and which should come from you

Look at the right-hand column. Almost none of it is design. It's information about your business that a generator cannot invent, so it fills the gap with the average instead. That's the whole mechanism in one line: the sameness is what turns up wherever your specifics are missing.

Does it actually cost you anything?

Sometimes yes, often less than people assume, and it depends heavily on how the visitor arrived.

If someone has already been given your name by a neighbour and is checking you exist, a tidy generic site is fine. It loads, it's got your number on it, it doesn't look abandoned. Job done. The generic look costs you nothing there.

It costs you when the visitor is comparing. Three tabs open, three local firms, deciding who gets the call. If all three sites say "quality workmanship you can trust", the only differentiator left is price or whoever answers first.

What actually breaks the pattern

Not a bespoke design. Not hiring an agency for £4,000. The sites that read as genuinely different are the ones where real decisions drove the content. Concretely:

• Say the specific thing. Not "emergency call-outs available" but "we answer the phone at 2am, and if you're within 20 minutes of Leeds city centre we'll usually be there inside the hour."
• Put numbers in. A price range, a call-out fee, a typical lead time. Most competitors won't, which is exactly why it works.
• Use your own photographs. A phone photo of a job you finished last Tuesday beats any stock image, even a blurry one. Especially a slightly imperfect one — it's evidence, and it can't be generated.
• Name your limits. "We don't do new builds." "We're not taking on work past Doncaster." Saying what you don't do makes everything else you claim more believable.
• Write one paragraph in your own voice and leave it un-polished. The bit where you explain why you left your old firm, or why you only take three jobs a week. It will sit oddly against the tidy sections around it. Good.
• Cut sections you have nothing to put in. An empty testimonial row with three invented quotes is worse than no testimonial row. If you haven't got reviews yet, delete the section.

None of this requires design skill. It requires you to answer questions only you can answer, and then insist the site reflects those answers rather than the default. That's also the argument for keeping a human in the loop at the point of sign-off — the generator produces the scaffolding fast, you supply the facts that make it yours. We've written up how that process works on an AI-built website you still sign off.

Where this argument doesn't apply

Three honest caveats, because "looks the same" gets overused as a criticism.

Conventions exist for a reason. Navigation at the top, phone number top-right, contact form at the bottom. People know how to use those. Moving them to be different makes your site worse, not distinctive. The sameness worth fixing is in the content, not the furniture.

Some businesses genuinely don't need a distinctive site. If nearly all your work comes from one contractor, or a trade counter, or repeat customers who ring you directly, your website is a business card. Make it load fast, put the number in large type, and spend your time elsewhere. Missed calls will be costing you far more than your homepage is — there's honest arithmetic on that in what missed calls really cost a UK trades business.

Different isn't automatically better. A site with an unusual layout, custom fonts and a slow-loading animated header can perform worse than the generic one it replaced. If you're going to change something, change the sentences before you change the styling.

The tidy-but-identical look is what you get when a tool makes every decision on your behalf. It's the honest visual record of a website where nobody supplied any specifics. Supply them — your prices, your patch, your photos, the jobs you turn down — and the same underlying components stop reading as a template, because the thing people are actually reading is no longer the average of anything.