I’ll say the quiet part out loud, because you’re going to ask anyway and I’d rather you hear it from me.
Yes, I use AI to build websites. It is a significant reason I can deliver a real custom site in seven days for $399 instead of six weeks for $6,000.
Now here’s the part almost nobody says: AI is terrible at making a website good. It’s excellent at making one fast. Those are not the same thing, and the gap between them is where every bad AI-built site lives.
So here’s exactly how I use it, exactly where I don’t, and what I do to make sure the thing that lands on your domain is better than what a person would have built alone rather than worse.
What AI is genuinely great at
Production speed. Writing out five hundred lines of clean, structured, commented CSS is a mechanical task. I know exactly what I want the result to be. Typing it is just labor. That labor used to be six hours and now it’s forty minutes, and the output is identical because I specified it precisely.
First drafts of everything. A first draft of a services section, a first pass at FAQ questions, ten headline options to react to. I throw out most of it. But reacting to ten bad options is faster and produces better work than staring at a blank page trying to produce one good one.
Being an infinitely patient quality inspector. This is the big one and I’ll get to it in a second, because it’s where the real value is and it’s the part nobody talks about.
Doing the boring, essential work nobody does. Alt text on every image. Schema markup filled out completely. Meta descriptions for every page written individually instead of copy-pasted. This is the stuff that’s genuinely important for how Google reads a site and genuinely tedious, which is why on most sites it’s just missing. Now it isn’t.
What AI is genuinely bad at
Taste. It will produce something competent and completely forgettable every single time, because it’s averaging everything it’s ever seen. Average is the opposite of what you’re paying for. Every decision that makes a site feel like a specific business rather than a generic one is a decision I make.
Knowing what actually matters to your customer. AI will happily write “Welcome to our website. We pride ourselves on quality service and customer satisfaction.” That’s not copy, that’s the sound of nobody. Knowing that a plumber’s headline should be about how fast you’ll get there and a med spa’s should be about how you’ll feel afterward requires knowing something about the business.
Truth. It will invent things. It will confidently produce a review that doesn’t exist, a certification you don’t have, a founding year it made up. This is the single most dangerous thing about AI in this work, and it is exactly why I have a hard rule that no claim, number, review, or credential goes on a live client site unless it came directly from the client.
Restraint. Ask for an impressive homepage and you’ll get a gradient, an animation, a pattern, a texture, three accent colors, and a parallax scroll. All at once. The hardest part of design is deciding what to leave out, and that’s a judgment call.
The part that actually matters: AI as the inspector
Here’s the flip that changed how I work.
Most people think of AI as the thing that writes the code. For me the bigger win is using it as the thing that checks the code, at a level of thoroughness no human would ever sustain.
I have a quality checklist with a hundred point scoring model and about a dozen hard gates. Before anything goes live, the build gets run against all of it, automatically, at four different screen widths, on every page.
Is there horizontal scroll at 360 pixels on page four? Does the contact form actually fire and show a success message? Is the phone number on the site character-for-character identical to the phone number on the Google listing? Does the body text pass contrast against the background? Is there a broken image reference on the gallery page? Does the schema validate? Is there a file path in the code that points at a folder on my laptop and will break the second it deploys?
That’s dozens of checks, on every page, at four widths. A human being will do that carefully the first three times and then start skimming. That’s not a character flaw, that’s just how attention works. A machine does the four hundredth check exactly as carefully as the first.
So the honest picture is this: AI writes the first draft and I make the decisions. Then AI checks my work harder than I would check it myself. That combination produces a better site than I built five years ago with more hours and a bigger budget.
The rules I hold to
These aren’t nice ideas. They’re the actual constraints.
1. Nothing ships without me looking at it. Every page, every breakpoint, with my own eyes. The automated gate catches what’s measurable. I catch what’s ugly.
2. No invented facts. Ever. Reviews come from your actual Google reviews. Years in business, license numbers, certifications, and service areas come from you. If I can’t verify it, it doesn’t go on the page. On demo builds, sample content gets clearly labeled as sample.
3. Design decisions are mine. Layout, type, color, what gets cut. I’ve been doing this for twenty years and that judgment is the product. The AI is a very fast pair of hands.
4. The copy has to sound like a person. I strip out the tells. No “in today’s fast-paced world.” No “we pride ourselves on.” No em dashes everywhere, which is the current giveaway. If a sentence sounds like it was generated, it gets rewritten until it sounds like somebody said it.
5. Speed savings go to you, not into more margin. This is the whole point. The efficiency is why a real custom site costs $399 instead of $4,000. If I pocketed the savings, none of this would matter to you.
What this means for what you pay
Straight math.
A traditional agency build of a five-page local business site runs somewhere between $3,000 and $8,000, and takes four to eight weeks. Most of that is labor: a project manager, a designer, a developer, a copywriter, and a stack of meetings to keep them in sync.
I’m one person who can design, write, and build. AI removes most of the mechanical labor from all three. The meetings don’t exist because there’s nobody to sync with.
What’s left is the part that was always the actual value: knowing what to build, and having the taste to know when it’s right.
That’s $950 for a five-page site delivered in seven days. Not because I’m cutting corners, but because the corners that got cut were meetings and typing.
The honest risk, and how you check for it
There are people right now selling AI-generated websites that are genuinely bad. Generic copy, stock photos, no local relevance, broken forms, no schema, and a load time that would embarrass a 2009 site. The tool made it possible to produce garbage at scale, and some people are doing exactly that.
So here’s how to tell the difference, whoever you hire:
- Does the copy say anything specific? If you could swap in a different company name and it would still read fine, it’s generic. Real copy is unswappable.
- Are the photos actually your business? Stock photography is the loudest signal that nobody cared.
- Does the form work? Submit it yourself. Right now. Check the email arrived.
- Load it on your phone on cellular data. Not on office wifi. If it’s slow there, it’s slow.
- Ask if you own the files. If the answer is complicated, you don’t.
Those five checks take ten minutes and they’ll tell you more than any sales conversation.
The short version
I use AI the way a good contractor uses power tools. It does the cutting and the lifting. It does not decide where the walls go.
If you want to see what that produces, look at the work. That’s the only argument that matters.
