ChatGPT can build you a website — and that isn't enough

ChatGPT hands you HTML in 30 seconds. Lovable generates a full site in five minutes. v0 writes React components better than half the junior developers out there.

All of that is true. And all of it, for a business that sells, is still not enough. Not because of the code quality — because of everything the demo never shows. This post is the list of what happens after the "wow," and why it matters to someone who doesn't want a side project, but a page that brings in clients.

One note up front: we use AI every day. The team behind Aurora33 has been using these tools in our workflow since ChatGPT v3. This isn't an "AI doesn't work" post. It's a post about where AI breaks when the goal is to sell, not to test the tool.

The demo is the easy part

What you see in the demo of any "describe your site and the AI builds it" tool is the first 15% of the work: HTML, CSS, a page structure that looks decent, placeholder content generated in seconds.

That's real. That no longer needs a developer. For the landing page of a side project only your brother will ever see, it's perfect.

The problem is what happens after you close the demo tab.

The 12 things the demo didn't do (and that you'll have to solve)

1. Domain and DNS

The AI gave you a site at yourname.lovable.app or chat.openai.com/canvas/.... That isn't your domain. To have yourbusiness.com you need to buy the domain, configure DNS records, point them at the hosting, wait for propagation. If you don't know what an A record or a CNAME is, you're already stuck.

2. The real cost of "free" hosting

The site the AI hosted "free" on its platform rarely stays free once you need it for real. What happens often:

• Traffic limits that show up without warning: the free plan holds 100 visits a month; when your business gains traction, the site goes down or you get pushed onto a plan that charges monthly fees in dollars.
• Features that move behind paid tiers: custom domain, working forms, basic analytics, removing the "made with [tool]" banner — each one usually sits behind an upgrade.
• Technical cost when something stops working: when the site breaks (because the platform updated something, because you changed a component, because you integrated an external service), the only way out is to pay for support, hire someone who understands that specific platform, or learn it yourself.
• Platform risk: several no-code and AI-builder tools have shut down or pivoted in the last 5 years. If your site lives only there, the day they close it goes with you.

For real production you need stable, professionally managed hosting, a properly configured domain of your own, and a setup where the business isn't tied to a single platform that decides how much to charge you next month.

3. SSL / HTTPS

Without HTTPS, browsers flag your site as "Not secure." Google penalizes you in ranking. Some browsers won't even let visitors fill out forms. Setting up HTTPS well (with automatic certificate renewal) is trivial if you know how — and a five-day wall for someone who doesn't.

4. Forms that actually reach your inbox

The form the AI dropped in there is decoration. By default, it sends nothing anywhere. For a visitor's "Contact us" to turn into an email you actually receive, you need integration with a service (Formspree, Resend, your provider's SMTP) and the right configuration. Leave it as-is, and the client thinks they reached you. You never find out. The sale is lost silently.

5. Technical SEO

The AI generated a generic <title> and <meta description>. It did not generate:

• A clean, consistent URL structure.
• An XML sitemap.
• A properly configured robots.txt.
• Structured data (JSON-LD for your business, breadcrumbs, FAQ).
• Open Graph tags so your link looks right on WhatsApp and social.
• Canonical URLs to avoid duplicate content.

Without that, your site doesn't show up in Google. It's not that it ranks low — it doesn't appear. And it's as invisible to the business owner as it is to the AI that generated it.

6. Speed / Core Web Vitals

The AI optimized the code. It didn't optimize the images you uploaded, didn't implement proper lazy loading, didn't prioritize LCP, didn't remove the render-blocking fonts. Your site can weigh 8 MB and take 9 seconds to load while looking beautiful in the demo.

7. Analytics

How many visits does your site get? Where do they come from? What do they do? Which page bounces the most? Without analytics set up (Google Analytics 4, Plausible, Umami, whatever) you're flying blind. The AI doesn't configure it because it doesn't know where you want the data sent.

8. Accessibility

Color contrast that meets WCAG, correct ARIA labels, keyboard navigation, alt text on images, visible focus on interactive elements. The AI generates what looks good on screen. What feels right with a screen reader, almost never.

9. Security

Security headers (CSP, HSTS, X-Frame-Options), anti-spam protection on forms, rate limiting, input validation. Without this, your form fills up with bots a week after launch and your inbox ends up on a blacklist.

10. Maintenance and updates

A new version of dependency X came out. There's a security bug in library Y. Your site stopped rendering well on the latest iPhone. This happens. The AI that generated the site three months ago doesn't notice and doesn't warn you. Someone has to maintain it.

11. Real integrations

WhatsApp Business API. Stripe or Mercado Pago. CRM. A calendar for booking appointments. Mailchimp or Resend for newsletters. Each integration is real configuration work, secure credential handling, error handling for when the external service fails. The AI gives you a button. Making the button do what it says, that you configure.

12. Content that isn't placeholder

The site the AI generated has "elegant Lorem ipsum" in disguise: generic text that sounds fine but says nothing concrete about your business. Replacing it with copy that actually sells takes knowing your customer, your differentiator, and how people buy in your industry. That's known by someone who studied marketing or by the business owner. The AI doesn't.

Each of the 12 items looks small on its own. Together they are the difference between a page that exists and a page a client can find, trust, and buy from.

Why this matters for a business that sells (not for a side project)

For a side project, the AI-generated site is great. For a personal page, a simple portfolio, an experiment — it works and saves hours.

For a business that needs the site to bring in clients, the difference between "I have a site" and "I have a site that converts" is the 12 things above. None is optional. And none is in the demo.

The honest math looks like this:

• Time to build the initial site with AI: 1 day.
• Time to solve the 12 things above if you're not a developer: between 3 weeks and 3 months, depending on how stuck you get.
• Time to solve them if you're a developer but not a web specialist: 1 to 2 weeks.

AI saved you 15% of the work and gave you the false impression it was already done.

What you can do with AI — and what needs human judgment

What AI does well today:

• Generate a first draft of code and design. It speeds up the start. We use it.
• Solve specific, well-scoped technical tasks (write a function, a component, a SQL query).
• Iterate on something that already exists: "make this more compact," "add a loading state," "fix this bug."
• Generate supporting content: product descriptions, image alt text, a first draft of copy.

What still needs human judgment:

• Deciding the architecture (what stack to use, how to structure the pages, when to split into components, what gets hosted where).
• Connecting the site to the real business (what the client wants when they arrive, what action they have to take, how success is measured).
• Knowing what's important and what isn't. The AI generates a critical detail and an irrelevant one with the same confidence. Telling them apart is human work.
• Keeping consistency over time. An AI in separate sessions makes different decisions without knowing it does. A human team builds with judgment that evolves but doesn't contradict itself.

The position

Aurora33 uses AI every day. We integrate it into client workflows (automatic responses, quote generation, lead follow-up). We know what it can do today and we know exactly where it breaks.

What we don't do is sell the demo as the solution. When a client arrives saying "I already tried with ChatGPT but it didn't work for me," what almost always happened is they got stuck on one of the 12 things above and don't know that's the problem — they think they need "another, better AI."

They don't need another AI. They need the other 85% of the work.

---

Binary diagnosis

Want to learn, experiment, build a side project? Use the AI. Use all of it. It'll save you weeks.

Need a page that brings clients to your business in the next 90 days? The AI is just the tool. The rest of the work is still work. The right question isn't "which tool do I use?" — it's "who helps me with the 85% the tool doesn't do?".