Can ChatGPT Build Your Website? I Tested It Myself

Can ChatGPT build your website? About half of it, and it’s the half most people underestimate. Ask nicely and the assistant will plan a site, write its copy and even generate a static page of code. Then it hands you a folder of files that has no way to get onto the internet by itself. For this experiment I split the job into the words half and the publishing half, ran the words half on a real brief, and then looked at what can finish the publishing half when an assistant can’t.

The brief and the ground rules

I used the same brief as our weekend field test, where five chat-first AI builders worked from it under the same rules. A general assistant should get the same fair fight as the specialists. The job: a fictional dog-grooming salon that needs a home page, priced services, a before-and-after gallery, an about page and a contact page, with warm copy, for customers who book by phone.

The rules:

• One afternoon of work with the assistant, plus a publish attempt that runs until it honestly ends.
• No new subscriptions unless the publish step demanded one. It did, and that’s the point of the test.
• Everything written down as it happened: what the assistant produced, what the stalls cost, and how the bills split.
• Paid like a customer throughout.

Round 1: planning and copy, the best work of the afternoon

The plan came back after one request: five sections, a reason for each one, and a sensible copy approach for a local service business. The copy came page by page and stayed on brief. It was warm without sounding like an ad, and it kept the same voice across pages, which is where a tired copywriter usually slips.

What I got:

• a five-section plan listing the images each section needed, with draft alt text;
• full copy for every page, including service descriptions that understood grooming without getting sentimental about it;
• two alternative headings per section when I asked;
• revisions that did what I asked: making the services page less stiff took one instruction;
• a meta description for the site when requested.

This is the prompt that did most of the work, exactly as I ran it:

You are the web person for a small dog-grooming salon.
Plan five sections: home, services with prices, gallery, about, contact.
Tone: warm, plain-spoken, no corporate filler.
For each section: the copy, two alternative headings, and the images it needs with alt text.
Before writing, ask me three questions about the business if the plan has gaps.

Round 2: the code, scanned

When I asked directly, ChatGPT produced a single static HTML page for the salon. The markup was separate from the styling, nothing unusual, and it opened in a browser from my desktop without trouble. Then I asked it to check its own file for broken tags and accessibility mistakes and explain each fix. It reported back, applied the changes, and the second version was cleaner.

That’s as far as it goes: a readable brochure page that looks fine on a desktop. A static page is closer to a flyer than a business site, and its advantages disappear once the brief’s other requirements come into play.

Round 3: the wall, in four places

Hosting. A file on a desktop isn’t a website. ChatGPT has no hosting, no domain step and no publish button, so its code lives wherever you paste it. Going live means dealing with registrars and hosting, the chores the brief said to skip and the ones that make non-technical owners give up within a few days. An AI builder app treats this as a basic feature: you chat, it renders the site, and publishing and hosting are included.

Images. The copy called for photos, and the assistant gave me descriptions and stock-search terms instead of image files. A good image brief is worth having, but ChatGPT couldn’t compete on the image half of this test at all. A chat builder like Lufibot generates copy and images together, so the gallery actually exists instead of being described.

Mobile behavior. Calling a page “responsive” isn’t the same as seeing it on a phone. Without a hosted URL I had nothing to open on a phone, so the most important factor for a business whose customers book by phone went unchecked. Builder apps show the phone version where you work, and you can’t sign off on a phone-first site from a folder on your desktop.

Updates. As a small test inside the test, I changed a price on the menu. The only ways to do it were regenerating the code, retyping it, or editing it by eye. There was no dashboard, no form field and no menu structure to edit. Salons change their menus weekly, and this is where the gap stops being an opinion and becomes a chore.

The half the assistant could not reach, a chat builder did

To finish the brief, I gave the same salon outline to Lufibot’s chat on a phone. Its questions collected what the salon needed, the answers turned into a rendered site, and publishing and hosting are part of the app instead of a separate errand. The phone preview is the actual phone in your hand. Copy and images arrive together, and changing the menu means a chat message or a form field instead of editing code.

Pricing as I track it: a free tier to try the flow, $12/mo billed monthly, or $9/mo billed yearly. In the field test (five builders, one salon brief, one weekend, stop at publish) it scored 4.7, my field-test score, largely because it closes exactly the gap where ChatGPT got stuck in this experiment. The flow, question by question, is in our Lufibot first look.

The workflow I would actually keep

• ChatGPT for planning and copy. The site plan, the voice, per-page copy, revisions and image briefs. This thinking half was its best work in the experiment.
• A chat builder for publishing. It turns the plan into a real site with hosting, a publish button, a phone preview you can trust, and editing in place after launch.
• The handoff is easy. Lufibot’s interface is a conversation, so the assistant’s plan goes into the builder as more conversation. I used the same sentences for the salon brief in both tools and nothing got lost.

The tally, qualitative

Job	ChatGPT alone	Chat builder app
Plan and page copy	Its best work; worth keeping	Decent, faster, less controllable
Images	Descriptions and search terms only	Generated together with the site
Hosting and publish	No path from file to URL	Included in the app
Mobile check	Nothing honest to open on a phone	Real preview where you work
Weekly edits	Regenerate or edit code	Chat answer or a form field

Cost note

The words half ran on the ChatGPT plan I already pay for on my personal card, so planning and copy added no new subscription. The costs come in the publish half: a builder plan (Lufibot’s tiers are above) plus a domain if you connect your own name. Nothing surprising there. That’s how this category normally works, and it’s a good reason to do the words first, so the money you spend on publishing goes to finishing the site rather than drafting it.

FAQ

Can ChatGPT create a live website for free?

Not on its own. A chat gives you text and code, not hosting. Going live takes a publishing tool: a builder plan and, if you want your own address, a domain. The words half runs on a plan you may already have.

Is ChatGPT good at website copy?

It was the strongest part of the experiment: the plan, per-page copy, revisions that followed instructions, and alt text. If you use it for nothing else, use it for the words.

Can ChatGPT replace an AI website builder?

No, they do different jobs. The assistant plans and writes, and the builder renders, hosts, publishes and updates. Trying to use the assistant for the builder’s job is how I ran into the four walls above.

What should I do first if I want a website this weekend?

Plan and write the copy in ChatGPT first, one section at a time. Then hand the plan to a chat builder and publish from there with the phone preview on. Getting the order right matters more than which tools you pick.

Conclusion

The split is clear, and it took the whole experiment to see it. ChatGPT is an excellent planner and copywriter and a terrible publisher. Hosting, images, real mobile behavior and weekly edits are where it hits a wall, and no amount of prompting turns a folder of files into a live URL. A purpose-built chat tool is an ordinary planner and an excellent publisher: it works through the same kind of conversation and ends with a site hosted and published from the same phone. Use the assistant for the thinking and the builder for turning it into an actual website. A single tool that handles both halves on a phone does exist, and that’s what my 4.7 was scoring. Now the words half has a method too.