If you are going to vibe code your own website, do it like this
Plenty of people should build their own site, and we would rather you did it well than badly. This is the method we would give a friend: what to settle before you open the tool, how to work, and the checks that separate a site that earns from a site that merely exists.
Before you open anything
Decide four things while it is still cheap.
Every expensive mistake we see started here. The AI will happily build whatever you describe, including the wrong thing, beautifully.
- 01
Who it is for, specifically
Not "customers". One person, with a problem, at a moment. Everything else follows from this, and no AI can decide it for you because it does not know your business.
- 02
The one thing a visitor should do
Call, enquire, book, buy, download. One. A page offering four equal actions gets none of them. Write it on paper before you start.
- 03
What you already have
Logo, photos, testimonials, the words you use on the phone that make people nod. Gather them first. A site built around real assets beats one built around placeholders you intend to replace and never do.
- 04
Where it will live
Domain, hosting, email. Decide this before you build, in accounts in your own name. Retrofitting ownership is the most annoying job in this entire field.
How to work
Six habits that keep it from unravelling.
The difference between a good result and a mess is almost entirely process. None of this requires you to read the code.
- 01
One change at a time
Ask for a single thing, look at the result, then ask for the next. Batched requests produce changes you cannot untangle, and when something breaks you will not know which of the five things caused it.
- 02
Save a version you can go back to
Use the tool's history or version feature, and save a checkpoint whenever the site is in a state you would be happy to keep. The most common disaster is an hour of prompting that makes everything worse with no way back.
- 03
Describe the goal, not the implementation
"Visitors are missing the phone number on mobile" gets you a better answer than "make the phone number bigger". You know your problem; the AI knows more solutions than you do.
- 04
Paste errors in whole
The entire message, not a summary. Error text is the single most useful context you can give, and paraphrasing it throws away the part that identifies the cause.
- 05
Ask it to explain before you accept
"What did that change and what might it break?" A model will tell you, and the answer is often where you notice it has done something you did not want.
- 06
Stop when it starts going in circles
If three attempts have not fixed something, the context is wrong, not the effort. Go back to your last good version and describe the problem differently. Prompting harder into a mess deepens it.
Never skip these
Four checks before you tell anyone the site is live.
If you do nothing else on this page, do these. They take an hour in total and they are the difference between a site that works and a site that looks like it does.
- 01
Send yourself an enquiry from your phone
On mobile data, not wifi, to an address you actually check. Confirm it arrives, and check the spam folder. Then do it again a month later. A form that fails silently is the most expensive bug on the internet.
- 02
Run Lighthouse on mobile
Right-click the page, Inspect, Lighthouse tab, select mobile, run it. Read the accessibility list first and fix every item: they are specific, they are almost always easy, and they are what generated code gets wrong most often.
- 03
Use it with the keyboard only
Tab through the whole site without touching the mouse. You must be able to reach every link, open and close the menu, complete the form, and always see which element you are on.
- 04
Install analytics before launch, not after
Analytics plus a free behaviour tool such as Microsoft Clarity. Launching without them means the most informative weeks of the site's life happen with nobody watching, and you cannot get them back.
Be honest with yourself
When to stop and get help.
Not a sales pitch. These are the specific situations where carrying on alone costs more than stopping.
- You are taking payments or storing anything about a customer beyond a name and an email.
- You are in a regulated field, or a franchisor, insurer or professional body has requirements your site must meet.
- The site is your main source of customers and a week of it being subtly broken would actually hurt.
- You have been going in circles for days and the site is worse than it was on Tuesday.
- You need it to do something specific and real, and every attempt produces something that looks right and does not work.
FAQ
Questions people ask about this
How do I build my own website with AI?
Decide four things first: who it is for, the single action you want a visitor to take, what assets you already have, and which accounts will own the domain and hosting. Then work in small steps, one change at a time, saving a version whenever the site is in a state worth keeping. Before launch, run four checks: submit your own form from a phone on mobile data, run Lighthouse on mobile and fix the accessibility items, navigate the whole site with the keyboard only, and install analytics.
What should I never skip when vibe coding a website?
Four things. Confirming your enquiry form actually delivers, by sending yourself one from a phone. Running Lighthouse on mobile and fixing the accessibility failures. Navigating the site with the keyboard only. Installing analytics before launch rather than after. Each takes minutes, and skipping them is what turns an AI-built site from a bargain into a slow, invisible loss.
Why does my AI-built site keep breaking when I ask for changes?
Almost always because changes are being made in batches on top of code nobody has read, with no saved version to return to. Ask for one change at a time, look at the result before asking for the next, and save a checkpoint whenever the site is in a good state. When three attempts have not fixed something, go back to the last good version rather than prompting harder, because the problem is the context, not the effort.
Is it cheaper to build my own website with AI?
In money, usually yes. In time, often not, and the time is rarely counted. The hidden cost is the checking: the work that makes a site fast, accessible, findable and reliable is invisible, nobody prompts for it, and it is what separates a site that earns from one that just exists. If you will do that work, building your own is a genuinely good decision.
When should I stop and hire someone?
When the site takes payments or holds customer information, when a regulator, insurer or franchisor has requirements it must meet, when the site is your main source of customers and a subtly broken week would really hurt, or when you have spent days going in circles and it is getting worse. Those are the points where continuing alone costs more than stopping.
Start your brief
Start your brief
Tell us what you're launching and what it needs to do. We reply within one business day with a fixed price and the first available Monday.
- Websites from $15,000 ex GST, fixed. The exact figure comes with the reply.
- A reply within one business day, from Michael, not a form robot.
- A thirty-minute call to hear what the site has to do.
- A fixed-price scope, sent as a web page with a version number.
- Kick-off on the next available Monday, held for you. We take four studio builds a month.