Should You Let AI Build Your Website? Here’s the Test

Back in July, someone asked me, half poking the bear, “Can’t you just describe what you want to an AI and have it spit out a site?”

Two months later, the question is sharper: “Should I just do that instead of hiring anyone?”

Fair question. I’m constitutionally suspicious of new build platforms. If you’d handed me a shiny app builder in 2002, I’d have asked who owns the export button. I’m a grumpy old web guy with twenty-plus years of scar tissue, so take that prior into account.

Yes, sometimes. If you need a five-page site for a side project, a launch page, or something you could rebuild over a weekend without crying, an AI builder is a fine call. I’d tell you so. The question was never whether AI can build the site. It can. The question is who owns it and who maintains it in month six, when the tool changes the deal.

Before you type a prompt, ask three questions.

Can I leave with my site?

Framer has no native HTML export or self-hosting. Wix Harmony, released in January, is a fully hosted AI website builder with no code export.

Webflow lets you export static code on paid plans, but its CMS, forms, and ecommerce stay on Webflow’s hosting. You can take the brochure. You can’t necessarily take the machinery.

Lovable gets real credit here. It syncs actual React and TypeScript code to GitHub on every plan. Pro is $25 a month for 100 credits, and generated apps can use a Supabase backend. But the AI loop, hosting, credits, and backend still live inside that ecosystem.

Portable code is not the same as a portable business.

Could a smart egg move it? Probably. Could you move it at 9 p.m. when checkout stops working? That’s the test.

Who sets the price next year?

According to the pricing trackers, Squarespace moved its Basic plan from roughly $16 to $19 a month this year. Wix increases across markets have landed around 15 to 30 percent over 2025 and 2026. Webflow restructured its plans again in May, folding CMS and Business into a $25-a-month annual Premium tier.

Those figures are approximate. The pattern matters more.

When you rent the building, the landlord sets the rent.

That doesn’t make these companies villains. Software and AI cost money to run, and pricing changes. But your cheap website is a subscription whose future price you don’t control. Once the site becomes central to your business, leaving may cost more than staying.

Who’s responsible when it leaks?

CVE-2025-48757 was disclosed on May 29, 2025. A scan of 1,645 public Lovable apps found 170 with database tables readable by unauthenticated visitors because the generated apps lacked Supabase row-level security. The exposed information included emails and other personal data, payment information, and tokens.

The tool built what was requested. Nobody asked about the lock on the door because the person prompting didn’t know there was a door.

That’s not just a Lovable problem. It’s an every-builder problem.

AI is good at producing the thing you described. It can’t reliably catch every requirement you didn’t know existed. Security, accessibility, backups, privacy, deliverability, and recovery plans are the boring stuff that keeps grinding after launch day.

Won’t Google punish an AI-built site?

No.

Google’s spam policies target scaled, low-value content, not how a site was produced. Google doesn’t care whether a human, an AI, or a caffeinated raccoon assembled your HTML.

Google doesn’t punish AI sites. It doesn’t reward generic ones either.

The designs coming out of these tools are fine. Often good. They’re also kinda generic. Nothing necessarily signals that a real decision was made about your customers, positioning, or why someone should trust you instead of the next tab.

The number nobody has yet

Everybody in my world has a story about the vibe-coded project abandoned in month three. Snyk published one such account, but it’s an anecdote, not a statistic.

I’d love to cite a credible abandonment rate. There isn’t one yet.

If somebody tells you “80% of AI sites need a rebuild,” they’re making that shit up. Right now, we have stories, warning signs, and very new software.

It isn’t AI or a developer

The pitch presents two choices: use an AI builder or hire a developer.

That’s a false choice.

WordPress runs about 40.3% of all websites. It’s open source, it can live on your server, and you control the files and database. WordPress 6.9 added an Abilities API and an official MCP adapter. In plain English, plugin features are becoming things an AI agent can read and operate.

I build with AI every day, including Claude Code. I’m using it on a platform my client and I own. That shapes how we build: build with AI on something you own instead of renting the AI’s house.

Ownership doesn’t eliminate the month-six question of who maintains the site. It does mean a pricing change or product decision can’t trap your business behind somebody else’s door.

When I’d tell you not to hire me

Is the site temporary? Are you testing an idea? Is your budget under a couple thousand dollars? Are you comfortable fixing it at 9 p.m.?

Use the builder. Ship it. Do the thing.

If it turns into the business, come back when you need something more fully baked.

These tools are great, and they’re getting better. The question isn’t “Can AI build it?” It’s “When this thing is my business, who holds the keys?”

Build the thing. Just know whose door it’s behind.