The honest question anyone weighing up building their own site with AI has is this: do I get a real website, or an ugly page I'll have to rebuild with someone in three months?
I rebuilt my business's website this way. I own a children's entertainment business in shopping centres, I'm not a developer or a designer, and I didn't write a line of code.
The short answer is that you get a real website, as long as you understand one thing: the AI builds, but it doesn't decide. The decisions that make a site sell are still yours, and there are few of them. This post is about the exact border between the two.
What AI delivers today when you give it proper context
Worth starting with what's changed, because a lot of people still judge the idea on an impression from two years ago.
Today, with your business context in hand, an AI that executes builds the whole site: page structure, copy, layout, mobile version, contact form, and publishes it on your domain. You ask in plain English, look at the result, ask for a change, look again.
What comes out close to finished on the first pass:
- The section structure, in the order that makes sense for selling.
- The copy, in your voice, if you handed over the business context first.
- The look, consistent with your identity, or a proposed identity if you don't have one yet.
- The behaviour on a phone, which is where most of your customers will see it.
- The technical part nobody sees: title and description for search, speed, optimised images.
What usually takes two or three rounds of adjustment: the copy at the top (the first sentence is always the hardest), the choice of proof, and the hierarchy of what appears first.
One detail worth its weight in gold that almost nobody asks for the first time: ask for the proposal before the build. In other words, ask the AI to show you the structure and the copy for each section for you to approve, and only then tell it to build. Correcting copy in a proposal takes minutes. Correcting a finished site born of a misunderstanding takes hours.
The decision that comes before the site: capture leads or sell online
This is the first decision that's yours alone, and it shapes everything.
Will your site sell start to finish, with the customer paying right there? Or will it capture the details of anyone interested, so you can close later by message or phone?
These aren't variations of the same site. They're different sites.
Full online selling calls for a shop window, product descriptions, checkout and payment trust signals. Lead capture calls for a short form, a messaging button and a fast response from your side.
Mixing the two dilutes both. A page with two goals converts worse than two pages with one goal each.
The practical rule I use: a fixed-price product with standardised delivery sells online directly. A service that needs a conversation, a quote or trust captures the enquiry and closes in a chat. You can change later, but starting right saves rework.
Decide this before you open the AI. It's an owner's question, not a tool's.
A structure that converts, and why you should ask section by section
A site that sells isn't creativity. It's a known sequence that answers the visitor's questions in the order they appear in their head.
Ask for the site in sections, in this order:
- The top. One sentence saying what you do and who for, plus an action button. The person decides in seconds whether they're in the right place.
- What you offer, as benefits. A benefit is what the customer takes away, not a feature of the product. "300 degree oven" is a feature. "Pizza ready in 90 seconds" is a benefit.
- Proof. A real testimonial, a real photo, a real number. It's what makes the rest believable.
- Objection handling. The reasons the person doesn't buy, answered before they ask.
- FAQs. The real questions, the ones that land in your messages every week.
- Footer with contact, address and social links.
And the action button repeated down the page, always with the same wording and going to the same place.
Asking section by section has a practical advantage. When something comes out wrong, you fix one section, not the whole site. And it's far easier to say what you want done differently.
Adjusting in plain English: how to fix things without knowing a line of code
This is the part people don't believe until they do it.
You look at the page and ask, in your own words: "make the top more direct", "that text is far too long, cut it in half", "swap that colour for the blue in my logo", "put the messaging button here as well", "that photo looks odd on the phone".
The AI applies it and shows you again. You look. You ask for something else.
Two tips that save a lot of time:
Ask for one thing at a time when it's detail. A list of twelve tweaks at once tends to come back with nine done and three interpreted in a way you didn't want.
Describe the problem, not the solution. "This section isn't convincing me" works better than "increase the font size", because the AI sometimes sees the cause better than you do. If its solution doesn't work, then you send the specific one.
You don't need to learn technical vocabulary. You need to be able to say what's bothering you.
The test that separates a good site from a pretty site: open it on your phone
There's no simpler and more ignored criterion.
The vast majority of a small business's visits arrive by phone: from the ad, from Instagram, from the link you sent in a chat. A site can be beautiful on a monitor and broken on a phone, and you'd never know if you only looked on the computer. The reverse almost never happens.
So the rule is: once published, open it on your phone. Not in a simulator, on the actual handset.
What to check in two minutes:
- Can you read everything without squinting and without zooming?
- Does the main button appear without much scrolling?
- Does the form work? Fill it in and send it for real. Did it arrive?
- Does the messaging button open the chat with the message already written?
- Does the page load fast on mobile data, off your home wi-fi?
If it's good on the phone, it's approved. That's the measure.
Where to publish with no monthly fee, and what that means for cost
This is one of the parts that most surprises people coming from the agency model.
A small business site doesn't need paid hosting. There are publishing platforms with free plans robust enough for the traffic volume of a local shop, a clinic or a service provider. It's the same infrastructure that serves big sites, it's just that a small business's usage fits comfortably inside what they offer for nothing.
In practice that means your fixed site cost tends towards zero, and the only compulsory annual cost becomes the domain, which is the site's address (that yourbusiness.com).
To put a size on that: Cloudflare Pages, which is what I use, has a free plan that, according to Cloudflare's own documentation in July 2026, allows 500 deployments a month, up to 20,000 files per site and 100 custom domains, and serves the site's static files with no request limit. A local business with hundreds of visits a day doesn't come close to that.
Two honest caveats. This applies to a brochure site and campaign pages: an online shop with stock and high volume is another conversation. And free plans have limits, which change. Check the current limits of whichever platform you choose, on the day you choose it.
The point that matters: the monthly fee you pay today is almost never the real cost of the site. It's the cost of someone holding the site for you.
What AI won't decide for you: offer, price, proof and promise
This section is the most important in the post.
The AI writes, builds and publishes. It doesn't know four things, and none of them is technical:
The offer. What exactly you're selling, with what included. That comes from your business, your margin, your operation.
The price. It can suggest formats. It doesn't know your cost, or how much your customer can bear, or what the competitor down the road charges.
The proof. Testimonial, photo, number, real case. It can organise whatever you have. It can't invent it, and if it invents it you have a serious problem. Fake proof destroys trust faster than no proof at all.
The promise. What you guarantee the customer. That's a business commitment, not page copy. If the page promises a response within an hour and your operation delivers in a day, the site isn't helping you, it's burning you.
The site is the shop window of a commercial decision. If the decision is wrong, a well-built site just helps more people discover the mistake faster.
When a site built this way isn't enough
I won't pretend it fits everything.
This route is excellent for a brochure site, a services page, campaign pages and a blog. It solves what the overwhelming majority of small businesses need.
It isn't the shortest route when:
- You need an online shop with stock, product variants, calculated shipping and several payment methods. There are ready-made platforms for that, and fighting them isn't worth it.
- Your business depends on custom software, with customer logins, a members' area and complicated rules. That's software, not a website.
- You have a large team publishing content daily, with an approval workflow. There a traditional content management system may make more sense.
- There's a specific compliance requirement in your sector. In that case, find someone who knows the sector.
If you're in one of those cases, an AI-built site can still be your shop window, with the complicated part running elsewhere. But it would be dishonest to say it solves everything.
Where to actually start
If you've got this far convinced, there's a temptation to avoid: starting with the whole site.
A site has many sections, many decisions and takes a full day to get right. That's too much for a first experience.
Start with one page, with one goal. A campaign page, with one offer and one form. It's faster to build, far easier to judge (either the enquiry arrived or it didn't) and uses exactly the same skills. If you enjoy the process on that, the site is the same thing at larger scale.
And before you put days into a site, it's worth knowing what a customer costs you today. The free Cost Per Customer X-Ray, which I send by email, is a prompt you paste into a free AI: a few questions about your business, about five minutes, and you have the figure. It costs nothing, and it also shows you in miniature what working with an AI actually feels like.
Read next: How to give the AI your business context before asking for anything and How to build your ad's landing page without coding.