Building a business website with AI: five things that have to be true

I’ve been getting a lot of messages about this lately, so I wanted to put my thoughts in one place.

They come from a few different directions. Agencies tell me their clients want to change their site by chatting with it, the same way they chat with ChatGPT. A marketing director at a software company told me they want to build pages faster than their current setup allows. And every so often I get a message from a business owner who built their own site with AI, it’s not turning out right, and they’d like an expert opinion.

So here’s my take. Not the hype, and not the doom. Just what I think is true about building a business website with AI right now, from someone who’s actually doing it.

The short answer

It depends. AI can build your website, and it can build a good-looking one in an afternoon.

But I wouldn’t trust most of my clients’ sites to that approach. Not yet.

That probably sounds strange coming from me, because this site is built that way. I started it on October 17th with Claude Code. It’s an Astro site, kept in GitHub and hosted on Cloudflare, and every page you’re reading was built by talking to an AI. It’s been a good experience. But it’s my site, and it’s an experiment. I went from having no agency site at all to having this one, and some presence is better than none. I’m iterating on it constantly, and I know exactly what I’m watching while I do.

Most business owners don’t have that setup, and that’s what this article is about.

What people actually want from it

Almost everyone who asks me about this wants the same thing: to stop waiting.

They’re on an old site builder where changes take forever. Or they’re stuck in an agency loop where every edit goes into an approval cycle, and the response time is about 24 hours a round. Meanwhile, they spun something up in twelve minutes with ChatGPT and it looked pretty good. That’s a different frame, and I get why it’s frustrating.

Here’s the thing, though. Sometimes the drag-and-drop interface really is the bottleneck. It’s honestly faster to tell a machine a few sentences and have it move things around for you. But most of the time, it’s the process. If your communication and approvals are slow, a faster tool just makes the waiting more obvious. Get the response time down and your current system can get a ton more done.

And there’s another piece. ChatGPT does about 80% good at most things. For most people, that’s better than what they could do on their own, and it’s a heck of a lot faster. But agencies and creative directors have a better eye for how a business should present itself. Quality takes time. AI speeds up some of the work, but you still need effort over time to make something worth reading, visiting and coming back to.

My test: only go ahead if all five are true

I was talking with a friend who runs a web team about this recently, and I tried to put it into a decision tree. I don’t think it’s complicated. It’s more like: only proceed if every one of these is true.

1. Fewer than five people edit the site, and they can follow what the AI is doing.
2. You have a strong brand and design system, written down.
3. You have a technical partner for deployments and cleanup.
4. You have a messaging and copy partner.
5. You’re ready for AI to cost more later.

If even one is missing, wait, or get help first.

Most people I talk to don’t have any of them. What they usually have is a founder who’s technically inclined, or one or two people on the team who want to move faster. That’s a great start, and it’s not the same thing.

Let’s talk through each one.

1. Fewer than five editors, and they can follow along

A recent webinar I watched put this well. The marketing lead at a B2B software company, working with a small team and an outside SEO agency, moved their site off Webflow and onto a coded site she edits by talking to an AI. She’d never written code. She built 18 new pages herself. But she also said it works for two to five people touching the site, and that at fifty it turns into the Wild West.

The “reasonably technical” part matters just as much. You don’t need to write code. You do need to understand roughly what the bot is doing, notice when it’s confidently wrong, and explain clearly where you want it to go. If that sounds like a chore, it will be.

2. A strong brand and design system

AI gloms onto whatever you give it and repeats it everywhere. Give it a clear design system and it’ll follow it. Give it nothing and it falls back on the same “good design” everyone else is getting.

A friend who recently stepped away from the website business said it better than I could: AI design is about as boring as it gets, and you can spot an AI-designed site in less time than it takes to spot AI-written copy. They call it the genericization of the web, and I think they’re right.

Brands with a strong identity have a leg up here. Their rules give the AI a lot of guardrails and keep it moving in the right direction for a while. But even that drifts. Fonts get a hair off. Spacing changes from page to page. The models themselves behave differently from week to week. I wrote more about that drift in Should you rebuild your website with AI?

3. A technical partner for deployments and cleanup

Someone has to own the plumbing. Where the site is hosted, how changes go live, what happens to your forms, tracking and cookie consent, and what to do when a deploy breaks on a Saturday.

That webinar team had a technical CEO and an outside agency in their corner, and I don’t think that was a coincidence. That same friend, who now builds custom apps, put it more bluntly: no vibe-coded app is production-ready. Their team builds the first version with AI, then hands it to human developers before it goes anywhere near customers.

The research backs that up. Georgia Tech’s Vibe Security Radar tracks published security flaws traced back to AI-written code. It found about 18 cases across seven months in the second half of 2025, then 56 in just the first three months of 2026 (Georgia Tech). A simple marketing site with no logins is low risk. Anything that collects customer data needs a developer’s eyes.

4. A messaging and copy partner

This is the one that surprises people, and it’s where I see the most damage.

A law firm in Texas reached out recently. The owner had rebuilt the firm’s site with AI and wanted a second opinion because it wasn’t turning out right. And to be fair, the design was a big step up from what they had before. It looked very professional.

But it was flat. It was one page. It took a law firm with real people and real cases and turned it into a sexy-looking site that didn’t say anything. People hire a lawyer they know, like and trust, and there was no depth there to help anyone see why they should work together. The copy was generic. The only call to action led to a contact form, and you couldn’t tell where it would go or what would happen next.

The technical SEO was maxed out, for what it’s worth. Clean code, fast, well structured. But Google was never going to rank it, because there was no depth or expertise on the page. You can’t prompt your way into experience. That part has to come from you, and someone has to help you get it onto the page. That’s why the four stages of a website that works start with messaging and copy, not design.

5. You’re ready for AI to cost more later

This is the one I’d add after the last few weeks, and I don’t hear anyone else talking about it.

Right now, AI is subsidized. The subscription you pay for doesn’t cover what it costs to run. The numbers are starting to show up in public. Anthropic, the company behind Claude, filed to go public in October. Its filing shows $16.3 billion in revenue for the first half of 2026 and a net loss of $80.7 billion on paper, most of that an accounting charge. Its operations are close to breaking even, and it spent $5.6 billion on computing in a single quarter (Fortune). And OpenAI, the company behind ChatGPT, reportedly brought in about $10 billion in revenue last quarter and lost about $10.9 billion running the business (Forbes, citing internal documents reported by the Wall Street Journal). That’s more than a dollar lost for every dollar earned.

I don’t know what the real price ends up being, and nobody does yet. But when you line up what these companies earn, what they spend and what they’re valued at, the prices have to rise to meet it. So if you build, edit and run your site with AI, plan for that to cost more than it does today. Period. If your budget only works at today’s prices, that’s a reason to wait.

What it actually takes

Here’s what the demos don’t show.

When I build with AI, it’s typically a 200 to 500 prompt journey to take a project from a mess of a starting point to something worth shipping. Each of those messages takes one to five minutes to run. And I’m juggling a lot of projects and sites at once, so I’m reviewing one while another runs.

The problem I see for a lot of business owners right now is that it feels productive. The chat is running, things are moving, there’s a ton of activity on the screen. But they’re sitting there watching a machine run instead of running their business.

The webinar team was honest about this too. Eighteen pages took about 80 hours over two to three weeks, on top of a day job, with help.

And it’s not just building. On my own site I’m constantly watching Google Search Console, page speed, search rankings, how things load, how smooth the animation is, whether the images draw people in, and whether the copy and messaging are landing. I know what I’m looking for because it’s my job. Most owners don’t have the time or the training to watch all of that, and they shouldn’t have to.

How I actually use AI on client sites

I use AI every day. I use it to iterate quickly, to build design systems that scale, and to automate a lot of the building. But there’s always a person in the loop, and that person prevents so many issues along the way.

That’s the real difference. AI isn’t the problem, and it isn’t the solution either. It’s a very fast pair of hands. Somebody still has to know where the site is going, check the work, and say no.

It’s still early

If you only read the headlines, you’d think every business has already rebuilt its site with AI. That’s not what I’m seeing.

I’ve been studying the websites of 250 service businesses around Raleigh and Durham: law firms, dentists, plumbers, accountants, roofers and more. The full report is coming soon, but one finding stood out early. Only a handful of those sites were built with AI site builders. Most small businesses here are still on WordPress, Wix, Squarespace or software from an industry vendor, the same as they were a few years ago.

So if you’re feeling behind, you’re not. Most of your competitors haven’t made the jump either.

That’s good news. It means there’s time to do this in the right order: get clear on your message, write down your brand, and find the people who’ll keep it all lined up. Then pick the tool. The businesses that get those pieces in place first are the ones that’ll actually get the speed AI promises, instead of a faster way to make a mess.

Who it’s actually right for

When I picture the people who should go ahead and build their site with AI, they fall into three groups.

1. You have no budget, and you’ve got the know-how. You don’t want another Wix or Squarespace site, and you’re comfortable spinning up a GitHub repo and deploying it on Cloudflare yourself. That’s basically how this site started.
2. Your team already has the skills. Someone in-house can keep your brand, identity, messaging and the technical side all lined up, week after week.
3. You have partners who handle that for you. You don’t need to do it all yourself, as long as someone owns each of those jobs.

If you’re not in one of those three, you’ll probably get a site that looks great on launch day and slowly stops sounding or looking like you. That’s when the cleanup starts, and cleanup usually costs more than doing it right would have.

Where I land

So if a founder asked me in an elevator whether they should build their site with AI, I’d say “it depends,” and then I’d ask them the five questions above. If the answers are good, I’d outline the process and walk through it with them.

I’d also be honest that migrations are never fast, and it’s going to cost you to switch. But in the right hands, with the right systems and guardrails, you may get more performance out of an AI-built site than you’d get by sticking with the one you have.

I’m not saying don’t. I’m saying don’t hand AI your site and expect it to go well without those roles in place.

If you’re not sure where you land, bring your answers to a free 30-minute call. If you want the full picture first, your site, your systems and what to fix in what order, that’s the Systems Diagnostic.

Hopefully this has been helpful. Let me know what you think.