Vibe Coding Your Own Website Looks Great on Day One. Here's What Breaks by Day 30.

A friend of a friend tells you about Lovable. Or v0. Or Bolt. Or the new AI thing baked into Wix. You spend a Saturday afternoon describing your business in plain English, and by dinner there's a website that looks legitimately good. Hero image. Service pages. Contact form. A little animated banner. You publish it. You feel like a genius.

Thirty days later, the contact form is silently dropping submissions. Two of the service pages have layout breaks on iPhone. Google has indexed exactly one page. The Stripe button that was supposed to take deposits goes to a 404. And the AI that built it for you can't reliably tell you why — because it doesn't actually remember what it built last month.

This is the pattern, and it's not a knock on the tools. Lovable, v0, Bolt, Cursor — they're impressive. They build real things. The tools aren't the problem. They're built for one job, and a small business website is a different job.

What "vibe coding" is actually good at

Vibe coding (telling an AI what you want and letting it build) is good at:

• Prototypes you'll show three people and throw away
• Internal tools where the user is you and you can patch as you go
• Static landing pages with no forms and no checkout
• "What might this look like" mockups before you hire someone

That's a useful tool. For a business owner sketching ideas, it's faster than any designer alive.

Where it falls apart for a real small business website

A small business website is not a prototype. It's the front door to your business, and it has to do a list of unglamorous things flawlessly:

• Receive a form submission and route it to your email and your phone. Not "look like it does." Actually do it, every time, including from a customer using an iPhone with low signal in a parking lot.
• Load in under two seconds on a mid-range phone. Vibe-coded sites routinely ship with multi-megabyte hero videos, blocking JavaScript, and Lighthouse scores in the 40s. Google notices.
• Be readable by Google's crawler and the AI search tools that now cite businesses. Most AI builders ship single-page apps that look great to humans and look like one mostly-empty page to a crawler.
• Survive a tweak. "Change the headline on the homepage" should not break the booking flow on a service page. With vibe-coded sites, it routinely does.
• Comply with whatever rules your industry has. HIPAA for healthcare. State licensing disclosures for trades. Accessibility for any business with employees. Vibe-coded sites are completely unaware of any of this.
• Have a working backend. Email through Resend or SendGrid with your own domain authenticated. SMS through Twilio. Calendar through whatever your business actually uses. The AI builder lets you describe these, but the wiring rarely survives contact with reality.

The 70% problem

Spend serious time with these tools and you'll hit the same wall everyone hits: they get you 70% of the way to a working product. The first 70% feels miraculous. The last 30% is where every project goes to die.

The last 30% is also exactly the part that matters for a customer-facing business website. The form actually working. The booking actually going through. The site actually loading. The 70% — the visual mock-up that made you feel like a genius — is the easy part. It's also the part you could have done in any template-based builder for $20 a month.

Three failure modes we keep seeing

The silently broken form

The AI builds a contact form. It looks great. When a customer submits it, the form shows a success message — but the email never arrives. The owner finds out two months later when a customer asks why nobody called them back.

Why it happens: the AI generated front-end form code that posts to an endpoint that doesn't exist, or to an SMTP setup with the wrong DNS records. The success message is hardcoded.

The mobile that nobody tested

The desktop site is gorgeous. On an iPhone, the navigation overlaps the headline. The "Call Now" button is below the fold. The form is impossible to scroll. Mobile is where 70% of your traffic lives, and the AI builder rendered it in a desktop viewport and never checked.

The SEO catastrophe

The AI generated beautiful animations and a single-page layout. Google indexes one URL. There are no real page titles. The H1 says "Home." Schema markup doesn't exist. After six months, the site ranks for the business name and absolutely nothing else.

What an actual rebuild costs

The honest math, from someone who picks up these projects after they break: starting from a vibe-coded site is sometimes faster than starting from scratch, and sometimes much slower, because you have to figure out what the AI did before you can decide what to keep. If the architecture is bad — and it usually is — the rebuild is most of the work anyway.

The real cost isn't the rebuild. It's the time the broken site spent live. Every week your form is silently dropping submissions is a week of leads buying from your competitor.

So when does vibe coding actually fit?

It fits beautifully for:

• A landing page for a one-time event
• An internal dashboard you'll iterate on every day
• A mockup to show a designer or a developer what you want
• An MVP for a product idea where you'll throw away the code if it works

It doesn't fit for the website that has to bring you customers for the next five years. That's a different job, and the tools that are good at it haven't changed: a real codebase, real testing, real deployment, real ongoing maintenance.