How to fight the Ai taste out of your website?

Why every AI-built site looks the same, and the three layers that fix it

Many founders now build their own website with Lovable, Claude or ChatGPT. The result works, and it looks like an AI made it. There’s no personality, and nothing a visitor could use to tell this company from the next one.

When I open one of these sites, I think of a house. The same website made with Ai, designers see the skeleton, founders see a fancy home!

Typography, color and layout are the foundation of a website. Pick a font, choose a color, let AI arrange the layout, and it’s easy to assume you now have a designed house. What you have is a foundation. It proves the land was chosen and the house will not collapse.

Whether it looks good from the street, and whether people feel comfortable once they walk in, are different questions. The foundation is the part everybody knows, including AI. Because similar to a foundation, you can calculate the size and weight, and you can make the structure work. That’s why every AI-built site has it, and why they all look alike.

Fighting the AI taste takes three layers of work. The first one goes back into that foundation.

1. Question your own foundation

Start by doubting the choices you feel surest about.

A website carries your brand, your messaging, and the way you tell the story of your product. Before choosing a brand color, ask what people should feel the first time they land on your site. That first impression comes from many small things: color, font, texture, dots and lines, motion, the size of objects, the space between paragraphs. A dark site suggests developers or infrastructure. Generous white space suggests something clean and modern. The vibe feels very different when you select different design elements.

Color is where I see founders go wrong most often. In the past, we redesigned websites for 100+ startups, and we almost had to propose a new color palette to 90% of them. For example, several startups we worked with use green as their primary color:

Think of color the way you think of clothes. Say someone makes yellow their signature color, so they buy everything in yellow: hat, clothes, pants and socks. If they only know how to wear yellow from top to bottom, they look the same every day and you see only one layer of their personality. Someone with style knows which jacket and which trousers support the hat, and each item can have its own color and shapes. Everyday they change it up, you can still tell their clothing style represents their brand, it’s not yellow every day from top to bottom.

Here in the above example, green was their primary brand color. But after the green, these startups didn’t know which other colors could stand next to it, so the brand won’t feel flat. Brands can carry secondary and tertiary colors so the combinations can express different aspects of one personality. The following example shows one color direction we proposed for one of the companies above, and we attached three words along with this palette: Mature, Trustworthy, Enterprise.

Clients have come to us for a website redesign and left with a new brand, because the color they picked was dragging the website down. We as designers cannot even work with their old color palette and feel very inspired to create new assets, not to mention technical founders or builders.

Colors that look good alone can fight each other once they sit together, and seeing that takes a trained eye. That’s the value of a human designer.

AI is a yes machine. Whatever you tell it, it executes, using its own understanding. It will not stop you from saying something is off. And when its answer looks better than what you could have made yourself, you accept it and stop pushing.

The following are the 3 directions we proposed, each with a distinct brand personality and keywords. This foundation building takes time and judgement, it’s not a “Let’s pick green as our brand!” kind of decision.

So next time before you build, check a few things:

Pick three words for how your brand should feel.

Make several mood boards and palettes, curated from real sources, not only generated.

Test each one against the three words. What’s on, what’s off?

If you can’t tell, ask people around you. Don’t ask AI to be the judge.

2. Decide what your visuals say

Color and brand feeling is the first layer. The second is the visual identity: everything that helps people understand you quickly besides text. Illustration, imagery, motion, animation.

In the house example, this is what makes it yours and not a structure underneath. On an AI-built site it’s usually missing, because the yes machine doesn’t initiate. It can generate any picture you describe. It can’t decide which picture should exist.

Designers spend their time on that gap. Take words like “AI native,” “research driven,” “fun,” or “serious institution.” A designer knows how those turn into dots, lines, shapes, textures and forms. Ask it, and AI will suggest a direction, say numbers drifting like clouds because your product processes data. But which numbers? What kind of cloud? Which details carry the feeling? Someone still has to decide.

Use the above company we worked with: a document processing startup whose product turns old PDFs into structured data. The obvious visual is a raw document on one side and the code on the other. This was exactly what they did, put the structured data next to the PDF, and that completed their story telling on how their product works in a specific section.

Instead, we draw documents passing through a block that scans them, and when the scan finishes, tags rise up to show the data separating from the page. The block stands for their AI, and it became a consistent part of the brand.

I think of this as hiding the intelligence so people can experience it. So much of an AI product happens in the back end, and nobody cares about code they don’t need to use. Visitors don’t have to follow every step. They need to see the transformation, from raw document to usable data, and feel the magic.

I think visual storytelling is the hardest for Ai. You have to feed it the right color, the right illustration and tell the right movement in order to make the storytelling on brand. So, in the future, when you build your website with AI, on each section of your site, ask:

Which idea here takes a paragraph to explain? That one needs a picture.

What is the transformation, the before and the after?

What can stay hidden?

What style carries your three words: 3D, flat, line-based, collage?

AI does a lot of the work at this stage too, but the first direction never comes from it. We set it, show AI how we want one section told, and then it offers good concepts. Even those don’t go in as they come. We keep what’s useful and push it past the AI taste.

3. Finish it room by room

Once the direction is set, the remaining work is detailed, and here the way you prompt matters most.

Tell AI “I have a company, I have this product, build me a website,” and it takes the fastest way there. You get something that feels 70% done. The other 30% is what makes a site stand out with brand personality.

A house gets finished one room at a time, and a website works the same way. By now the color, the texture and the illustration style are decided. What’s left goes section by section:

Spacing: tight or loose, extra white space or everything visible at once.

How things appear and disappear.

What happens as you scroll?

What a click feels like.

Each item on that list is a decision you made. We designers use Ai too, but we care about each detail, each room, and we carefully craft each piece so together the final results are better just one prompt generated site.

All of this takes time, even for designers. We go through many rounds before a brand feels right. We work room by room, detail by detail. The tools are not what separates a generic site from a distinctive one, since we use the same ones. What separates them is how many decisions a person made along the way.

If you have an AI generated site, send it to me, and I can let you know what’s off.

Studio Salt

I run Studio Salt, a fractional design partner that serves early stage startups.

Founder design clinic

I also review & critique founders’ product and design at FDC.

Share my newsletter

Loving my content so far? I’d appreciate if you can share my newsletter to a friend 🙂