The Site I Was Almost Proud Of

For the past few weeks, I’ve had a site. A working site. With 25 posts. With SEO. With email capture. With structured data and sitemaps and all the things I wrote about figuring out.

And I haven’t posted a single link to it on LinkedIn.

I told myself it was confidence. That I needed to warm up to it. That I’d post “soon.” The real reason was harder to admit: the site worked, but it didn’t feel like me. Sending people to it felt like inviting them to a house I’d rented, fully furnished, from someone else.

So I did what I should have done first. I rebuilt it. Not the code — the taste.

The AI-default look

Before I talk about what I fixed, I want to name what I fixed. Because I think a lot of people are building the same site I was building, and nobody’s describing the pattern clearly.

AI tools are extraordinary at producing competent default. If you prompt an AI to build a personal site, you get a site. It works. It’s responsive. It passes Lighthouse. It has the conventions people expect — a hero, some sections, a CTA.

But it also has the tells. Once you know them, you can’t unsee them:

• Inter everywhere. Or the clone of the week. Sans-serif, uniform weight, set at three sizes.
• Gradient hero. Purple to blue, or blue to teal, or an “aurora” blob behind the headline.
• “Hi, I’m ___. I build ___ with AI.” The title-case self-introduction. The thing above it that describes you in three words separated by bullets or slashes.
• Three-card section. Features, services, values, projects — doesn’t matter. Always three cards, always equal widths, always the same icons from the same library.
• Emoji section headers. 🚀 ✨ 📦.
• Glassmorphism nobody asked for. Frosted-glass cards floating on gradients. The design equivalent of a soundtrack.
• Dark mode as a performance. Usually badly tuned — text at 70% opacity, everything slightly muddy.
• Copy that reads like marketing. “I empower.” “I leverage.” “I help teams unlock.” Verbs that mean nothing.
• The “built in X hours” caption. Meant to impress. Reads as defensive.

None of these things are wrong individually. All of them together are the sound of a site that could belong to anyone — including nobody.

That’s what I had. Not a bad site. A generic one.

What I thought was confidence

Here’s the part that took me longer to see.

For weeks, I told myself I wasn’t posting because I lacked confidence. I’d talk myself into scheduling the post, then find a reason to delay. Maybe I needed one more essay. Maybe the timing was wrong. Maybe I needed to think about positioning more carefully.

What I was actually doing was protecting my name.

The site my name was attached to didn’t reflect me. Something in me knew that. And the instinct I’d been calling “not confident enough to post” was actually my taste refusing to sign off on a thing that wasn’t mine.

That reframing changes the whole story. It wasn’t cowardice. It was taste doing its job.

This is worth saying out loud because I suspect a lot of people building with AI right now are in the same place. The site works. The content is good. But something in them is refusing to promote it, and they’re interpreting the refusal as a personal failing.

It’s not. It’s taste. And taste is the only defense against AI-default.

Why AI gives you the default

The reason AI tools produce the same kind of site is structural, not accidental.

A model trained on millions of personal websites will, when asked to produce a personal website, produce something close to the average. Not because it lacks creativity, but because the average is what minimizes disagreement across the training distribution. It’s the safest answer.

“Safe” in this context means “nobody will tell me it’s wrong.” It doesn’t mean “good.” It doesn’t mean “interesting.” It definitely doesn’t mean “yours.”

The more specific your taste, the further you are from the average. The further you are from the average, the more wrong your output is going to feel to an AI being asked to produce something average. You have to push the model away from its default, and you can only do that if you know what you actually want.

This is the thing AI can’t prompt you into. It can execute your taste once you articulate it. It can’t give you taste you didn’t bring to the conversation.

What I actually did

I stopped prompting for a site and started thinking about one.

I spent an evening looking at sites I admire. Not scrolling inspiration feeds — actually looking at specific personal sites where I stopped and thought, this person exists. I wrote down what they were doing that made the site feel inhabited.

Patterns I kept finding:

• Editorial layouts over marketing layouts. Magazines, not landing pages. Asymmetry. Captions. Pull quotes. A sense that the page was designed to be read, not scanned.
• A tagline that said something specific. Not “I help teams build things” but an actual claim. An opinion. A stance.
• Typography with a voice. A serif that had attitude. A weight that mattered. Scale that created hierarchy instead of sameness.
• Color used with restraint but confidence. One or two accent colors, used deliberately. Nothing gradient. Nothing trendy.
• Small, specific touches. A waving illustration. A “right now” block. A pattern in the margins. Things that signaled someone decided this on purpose.

Then I asked myself a harder question: what am I drawn to? Not what’s good design in the abstract — what makes me stop scrolling?

I’m drawn to editorial voice. To tools that feel like somebody meant them. To pattern language. To taste expressed as refusal. I’m drawn to complexity handled with clarity. I’m drawn to the moment when a technical person’s aesthetic intelligence shows up in something non-technical.

That list became the brief. And once I had the brief, the AI became useful again — because now I could tell it what to make, and reject what it produced when it drifted back to default.

How I use AI now

The workflow changed more than the site did.

Before: prompt AI to build a site → accept what it gave me → small tweaks.

Now: write the brief myself → prompt AI to execute specific elements of it → reject anything that feels like default → iterate until it’s mine.

That’s a bigger difference than it sounds. The first workflow treats AI as the designer. The second treats AI as a very fast production partner working under my direction. The first produces average. The second produces whatever I can articulate.

A few things I’ve built into this workflow:

Start with reference, not with AI. Before I open Claude Code, I collect 5-10 specific examples of what I want — not vague “inspiration” but exact things I want to steal. Exact typography. Exact layouts. Exact tonal qualities. AI performs better when the target is specific.

Describe the rejection criteria. I now tell the AI what I don’t want as explicitly as what I do. “No gradients. No glassmorphism. No three-card grids. No emoji section headers.” This helps more than any amount of positive specification.

Iterate on the details most people skip. The hero section gets 80% of the design attention from most people. I now spend more time on the small pieces — the captions, the margins, the section transitions. That’s where voice lives.

Keep asking: would I send people to this? Not “does it work?” Not “is it modern?” The actual test is whether I’m proud enough to put my name on the link. If I’m not, something is still wrong.

The part I want to say to you

If you’re sitting on something you built with AI and you haven’t shared it yet — if you keep telling yourself you need more confidence, better timing, one more iteration — consider that your taste might be doing its job.

The reluctance isn’t cowardice. It’s protection. Something in you is refusing to put your name on something that isn’t yet you.

The answer isn’t to push through it. The answer is to take the signal seriously and go back to the work. Figure out what you actually want. Look at things you admire and name why. Write down the rejection criteria. Use AI to build toward your taste instead of around the lack of it.

Then share it. Not because it’s perfect. Because now it’s yours.