What to check before you put an AI-generated site in front of customers
The site looks finished in the preview. Before customers see it, seven things have to be true: working forms, real analytics, HTTPS, uptime, backups, correct metadata and a tested rollback. The production checklist.
An AI-generated site reaches "looks done" long before it reaches "safe to show customers." The gap between those two states is not more design work. It is a short list of things that are invisible in a preview and very visible the moment a real visitor arrives: a contact form that silently drops messages, a page that goes down at 2am with nobody watching, a broken revision with no way back. This is the checklist that closes that gap. Seven checks, most of them a two-minute test, and a launch that stays boring.
1. Forms actually deliver
The single most common failure in an AI-built site is a contact form that looks like it works and does nothing. AI tools frequently generate a form with a nice success message and no working destination behind it, or wire it to a third-party handler the model happened to know about. Either way, a visitor fills it in, sees "thanks, we will be in touch," and their message evaporates.
Test it the only way that counts: submit the form as a stranger would, from a phone if you can, and confirm the message lands in a real inbox with the fields intact. If it does not arrive, the form is decorative. A host built for static sites should provide a forms relay so submissions are delivered without a separate service, which also keeps visitor data on one platform instead of scattering it to a handler you never evaluated.
2. Analytics you can actually read
You cannot tell whether a launch worked if you cannot see whether anyone came. Before launch, confirm the site records at least basic visitor counts and page views. The goal is not a heavy tracking suite bolted onto a two-page site; it is a first-party, privacy-preserving count that tells you the ad in the local paper sent fourteen people to the pricing page and none to the contact form. Check that the numbers appear somewhere you can find them, and that collecting them does not silently drag a cookie-banner obligation onto your site.
3. HTTPS is on, valid, and automatic
Load the live site and look for the padlock. HTTPS is not optional: browsers mark plain HTTP as "not secure," and visitors bounce off warnings. Confirm three things: the certificate is present, it is valid for your actual domain, and it renews itself. A certificate that has to be renewed by hand is a future outage with a date on it. The correct answer is that your host issues and renews the certificate automatically, and you never think about it again. If you are still attaching a domain, the custom domains guide covers how DNS and SSL fit together.
4. Uptime and someone accountable for it
A preview link makes no promise to stay up. A production site needs a host that keeps the site reachable around the clock and a named operator responsible when it is not. Before launch, know the answer to a blunt question: if the site is down on a Sunday, whose problem is that? "Mine, and I have no way to fix it" is not a launch-ready answer. Part of being genuinely online, from the definition of an AI-hosted website, is that a real operator has accepted responsibility for uptime, patching and availability.
5. Backups and a rollback you have actually tested
This is the check people skip and regret. AI-driven editing is fast, which means it is fast to break things too. An enthusiastic revision rewrites the menu page, it looks fine in preview, it goes live, and now the prices are wrong on a page customers are reading. Recovery has to be one action, not an archaeology dig through chat history trying to reconstruct Tuesday's version.
So confirm two things before launch. First, every deploy is captured as a snapshot you can restore. Second, and this is the part people forget, actually perform a rollback once on a test change so you know the button works and you know where it is. A backup you have never restored is a hope, not a safety net.
6. Metadata: title, description, favicon, and social preview
AI tools often leave the boring text fields generic. Before customers see the site, fix the metadata, because it is the first impression in three places customers never think about but always notice.
- The page title. The text in the browser tab and the blue link in search results. "Untitled" or a leftover tool default reads as unfinished.
- The meta description. The grey summary line under the search result. Write it, or a search engine will guess, usually badly.
- The favicon. The little icon in the tab. A missing one shows a blank page symbol that quietly signals "prototype."
- The social preview. Paste your link into a chat app and see what unfurls. A missing or wrong image and title here undercuts every share.
None of this is design work. It is a five-minute pass, and on your own hosting you control all of it, which is precisely what a locked-down preview publish does not let you do.
7. It works on a phone and loads fast
Most visitors will arrive on a phone. Open the live site on an actual phone, not just a resized browser window, and walk through it: does the menu work, is the text readable, do the buttons hit, does the form submit. Then check that the site loads quickly on a normal connection. AI-generated sites are usually static and therefore fast by default, but an oversized hero image or a heavy embedded script can undo that. A slow first load loses visitors before they read a word.
The pattern behind the list
Notice what these seven checks have in common: not one of them is about the code the AI wrote. They are about the layer around the code, the operation of a live site, which is exactly the part an AI cannot generate because it is infrastructure behaviour rather than markup. This is why the choice of where to host is a real decision and not an afterthought. The full arc from prompt to an operated, maintained site is traced in the prompt-to-production lifecycle, and if your site currently lives inside a builder suite, the trade-offs of moving it are in switching from a website builder to AI hosting.
On AI-Hosted
AI-Hosted, powered by VibeDeploy and run by Serso BV on infrastructure in Belgium, is built so that six of these seven checks are already handled by the platform: a forms relay that delivers to your inbox, first-party privacy-preserving analytics, automatic HTTPS, operated uptime, and snapshot-backed rollback on every deploy. The metadata pass is yours to write, because it is your voice, but you control every field of it. Plans are flat at 15, 39 or 129 euro per month including VAT, on the pricing page, with a 14-day free trial and no card, which is enough to run this whole checklist against a real domain before you pay anything. Run the list once, and the launch is the calm non-event it should be.
Put your AI-built site on a real domain
AI-Hosted, powered by VibeDeploy, runs your AI-built website on European infrastructure with a custom domain, automatic SSL, forms and analytics. The 14-day trial needs no card.
Related reading
Custom domains for AI-built websites: how it actually works
A preview link is not an address you own. What a custom domain is, how DNS and automatic SSL turn an AI-built site into a real one, and why the domain is the step that makes a launch count.
AI-hosted websites and GDPR: what a non-lawyer needs to know
A plain, factual guide to GDPR for a small AI-built website: what counts as personal data, the three places a generated site collects it, why where the data sits matters, and the short list you can actually act on. Not legal advice.
What is an AI-hosted website?
A plain definition of AI-hosted websites, five questions that decide whether a site is really online, and the preview-link cases that fail the test.