You paste a company description into a generator and get a website. It looks pretty good. You paste a description of a different company and get another website. It looks... the same. Different photos, same layout. Third company: there it is again.
This isn’t a coincidence, and you’re not imagining it. There are concrete reasons why generated websites look like siblings from the same litter. They’re worth knowing, especially if you’re commissioning a website and want to sense whether you’re getting thoughtless generator output.
The model learned from the same internet
A large language model is, roughly speaking, a statistical summary of a huge number of web pages that have existed online. When you ask it for a “modern business website”, it gives you the average of those statistics. And the average of a million business websites is: a gradient, a centred headline, three columns with icons, and a strip of client logos.
What you’re seeing isn’t the model’s bad taste. It’s the internet’s taste in concentrated form. The model has no reason to stray from the average, because the average is safe and close to what people have clicked on before.
Six signs that give a generator away
After going through hundreds of these sites, we know this list by heart:
- A navy-to-purple gradient behind the hero, often with blurry blobs of light.
- Three icons in a row with short captions: “Fast. Secure. Modern.”
- Generic claims that fit any business: “We are a leader in our industry.”
- A symmetrical layout where every section has the same number of columns and identical spacing.
- A stock-photo vibe: smiling teams at desks, handshakes.
- Zero specifics: no prices, no place names, no opening hours, no people.
The last point is the most important, and it’s easy to check. A real company website is full of awkward, inconvenient details: an actual price, the name of the person who answers the phone, a photo of the workshop taken on a phone. A generated website is smooth, because smoothness is statistically safe.
Why this is a problem, not just a matter of looks
A visitor who lands on your website decides in a split second whether they’re dealing with a real business or yet another artificial showroom. Research on website trust keeps saying the same thing: specifics build credibility, vague claims eat it away. A website that could belong to anyone belongs to no one.
There’s another side to it, too. Google and other search engines already assess whether content is machine-generated and cautiously reduce its visibility. A website that looks like a thousand others from the same month signals to a search engine that it deserves a more cautious approach.
How to get out of the trap
The recipe is short, and generators won’t like it: real content, a real brand, deliberate asymmetry.
Real content. Facts from the business: prices, hours, past projects, people, address. You can’t generate these, because they don’t exist anywhere on the internet. At Webomat, the agent works only with content from the client’s own website and is forbidden from adding anything of its own. No facts means no entry, not fiction.
A real brand. Colours and character drawn from the logo, the industry and the company’s own photos, not from a default theme. A nursery school and a sporting-firearms manufacturer can’t get the same template in different colours.
A design audit. Before any proposal is shown to a client, it goes through an automated check against 27 traits typical of generated websites. Above a threshold, it goes back for rework. That list of traits is, in practice, the list from this article, written into code.
Where exactly this look comes from
It helps to understand the mechanism, because then it’s easier to avoid. The model has seen millions of websites and learned what tends to appear together. A big headline, a gradient and a button in the middle show up together so often that, for the model, they are the one “correct” homepage. Three columns with icons are another such pairing. When you ask for a “nice business website”, the model spits out the closest average of everything it has seen.
It’s a bit like the restaurants around train stations: nobody designed them on purpose, yet they all look the same, because every element solves the same problem in the same tired way.
The second cause, forgive us for saying it: lazy prompts. “Make a modern website” doesn’t contain a single piece of information about your business. The model has nothing to extract character from, so it fills the gap with generic stuff. The third: tools that sell ready-made themes with AI bolted on inherit the look of those themes. They refresh the copy, not the visual skeleton.
How we test for the “generator look”
At Webomat, every proposal goes through an automated trait audit before the client sees it. The list has 27 items and keeps growing, because every new generator fashion adds entries. A few examples from the list, so you know what we count:
- a two-colour gradient on the first screen,
- three icons in an even row,
- round avatars of smiling people,
- a headline made of three sentences separated by full stops,
- a “Get started” button that doesn’t say what’s getting started,
- an overall symmetrical layout, section after section,
- a “5,000 happy customers” counter,
- an “Our mission” section full of platitudes.
Above the points threshold, the proposal goes back for rework, not to the client. The result: our websites differ from one another, because they start from different brands, and they differ from generator websites, because they pass through this filter.
Does “looks like AI” really hurt
In short: yes, in two ways.
The first is with customers. Someone who has seen five websites from the same generator this month will recognise the sixth in half a second. They won’t think “oh, artificial intelligence”. They’ll think “a business that couldn’t be bothered to invest”. Whether that’s fair is another question. That’s how snap judgements work, and you can’t switch them off.
The second is in Google. The search engine has been fighting mass-produced content for years and has a whole infrastructure for it. A website that looks and sounds like a thousand others from the same period gets a weaker start. Not because it’s bad, but because nothing makes it stand out.
A recipe for a website that doesn’t look generated
Specifics from our own practice, starting with what works best:
- Real photos. Your office, your workshop, your team. One genuine photo defuses the entire stock-photo vibe.
- Numbers and names. “We serve 12,000 lunches a year”, not “thousands of happy customers”.
- Asymmetry. A section where the photo is wider than the text. A gallery off to the left, not in an even row. The human eye will spot machine symmetry.
- Brand voice. One sentence no generator would ever write, because it requires your knowledge: about the customer who called at three in the morning, about the project that went wrong.
- Industry details. A price list by the hour? Directions with a note like “park round the back, next to the supermarket”? You can’t generate that, and that’s exactly why it works.
The shortest version of the recipe: anything that isn’t on the internet but is in your business, put it on your website.
Website generators are good. But let them decide everything and you get the average. And the average, let’s say it again, doesn’t sell to anyone.