Can Claude Build a Website? Yes — Here's How to Do It

What Claude can actually do

Three tools, three ceilings. Most articles on this topic blur them together, which is why their answers are vague.

This guide uses the first one, because it's the one you can actually use today with no setup. Claude is free to start, with paid plans from $20/month.

If you're after the second one — building sites on your machine with Claude Code, for clients — that's a different job, and we've written it up separately.

What it costs, up front

Because this is what you actually came to find out.

About that free tier

Plan on paying for the month you build. The free tier is very unlikely to get you to the finish line.

Here's why, specifically. Asking Claude a question is cheap. Making it write out an entire website — then rewrite the entire website because you want the button bigger — is one of the most expensive things you can ask it to do. Every round regenerates the whole page from scratch.

Step 1 tells you to iterate three or four times, because that's what takes the output from generic to actually yours. Round one is never the one you ship. So the exact behaviour that makes this work is the behaviour that empties a free session fastest.

What it looks like when you hit it: you're two or three rounds in, the site is nearly right, and Claude stops and tells you when it'll reset — usually a few hours out. You can't pay to skip it. There are no top-ups on claude.ai; consumer plans are flat subscriptions, not pay-per-token. (Pay-per-token is the Claude API, a different product for developers.)

So your options are come back tomorrow, or pay $20 and finish today.

Step 1 — Prompt it

Time: 5 minutes

Open Claude and describe the business, not the website. The instinct is to specify layouts and colours. Don't. Claude is better at design than most people are at describing design, and worse at guessing what your business actually does.

A prompt that works:

"Build me a one-page website for my hair salon in Bangsar. We do cuts, colour, and treatments. Show our services with prices, our opening hours, photos, and a way for people to book. Our vibe is calm and minimal, not flashy."

What comes back: a complete, styled, working page — usually in under a minute.

Then iterate. This is the part people skip. Talk to it like a designer you've hired:

• "Make the booking button more prominent."
• "The hero text is too long. Cut it in half."
• "Add our address and a Google Maps link."
• "Show me a version with a warmer colour palette."

Three or four rounds gets you somewhere genuinely good.

Heads up: this is the step that empties the free tier. Every round regenerates the whole page, which is expensive — there's a good chance Claude pauses you two or three rounds in, right as the site is getting close, and names a reset time a few hours out. See the cost section: it's wait until tomorrow, or $20 to finish today.

Step 2 — Understand what you're holding

Time: 1 minute, and it's the most important minute here

What Claude gave you is a design, not a website.

It's real code and it really works — but it's one file with your prices, your hours, and your colours welded directly into the markup. Nothing separates the words from the layout. That has two consequences, and everything below follows from them:

• It'll go live in five minutes. It's a static file. Any host will take it.
• You're the only person who can ever change it. There's no "edit text" button, and there never will be, because there's nothing underneath it to edit. Changing a price means going back to Claude, asking it to regenerate the page, downloading it again and re-uploading it — or opening the HTML yourself. Your receptionist can't. Your VA can't. The person you hire next year inherits a file, not a system.

Step 3 — The fast way: host the file yourself

Time: 5 minutes. Free.

If this site is a one-off — a launch page, an event, something you'll bin in a month — stop here. This is genuinely the right call and you don't need anything else.

1. In the Artifact panel, copy or download the code
2. Make a folder. Save the file inside it as index.html — that exact name, it matters
3. Go to Netlify Drop (app.netlify.com/drop)
4. Drag the folder on

Live URL in about ten seconds. It'll look like random-words-12345.netlify.app — ugly, but real. Free, permanently, for a static file like this. Vercel, Cloudflare Pages, and GitHub Pages all do the same job; Netlify Drop is the only one needing no account to try.

If Claude gave you separate CSS or JS files, put them in the same folder with the names Claude used. Ask it: "give me this as separate files and tell me exactly what to name each one."

Step 4 — The way that lasts: move it somewhere you can run it

Time: 30 minutes. Free.

Here's the part most people work out too late.

The design Claude made you doesn't transfer. The two things worth keeping do.

That sounds like a downgrade. Stay with me, because it's the opposite — and it's the single most useful thing in this guide.

Why the design can't come with you

Any platform that lets you edit your own site works by keeping the content separate from the code. Your headline lives in one place, the layout in another, your colours in a third. That separation is what puts an "edit" button next to your prices.

Claude's output has no separation. The words are inside the markup. That's not a flaw — it's what a single HTML file is. But it means no editor on earth can give you a text box for a headline that's fused to a <div>.

So you can't import the file. What you can do is take the parts that were always the valuable bit.

What Claude is actually best at

Not the layout. Your words and your look.

Go back to the same chat — it already knows your business — and ask for both, cleanly:

Prompt 1 — your copy:

"Using everything you know about my business from this conversation, write me the finished website copy as plain text. I need: a hero headline (8 words max), a subheadline (20 words max), an about paragraph (60 words), a one-line description for each service, and a closing call to action. Label each one. No HTML, no code — just the words."

Prompt 2 — your brand:

"Suggest a colour palette and font pairing for this business. Give me a primary colour, an accent colour, a background colour and a text colour as hex codes, plus a heading font and a body font from Google Fonts. One line on why each fits. Keep it [calm and minimal]."

That's your whole website, minus the scaffolding — and unlike the HTML, it's portable. It goes into any platform, ours or anyone's.

Then put it somewhere it can live

1. Create a free [OhhWells] account
2. Pick a template built for your kind of business
3. Paste in the copy from Prompt 1, set the colours and fonts from Prompt 2
4. Publish

Live on random-words-12345.ohhwells.site in about a minute. Free, permanently. Booking is built in and actually books. And when your prices change on a Tuesday, you change them yourself in the editor — no chat, no code, no HTML at 11pm.

Step 5 — Put your own domain on it

Time: 2 minutes to connect it, once you've got a domain

Applies whichever path you took. random-words-12345.ohhwells.site on your business card isn't the move.

First, buy a domain if you haven't got one — Namecheap, Cloudflare, or a local registrar for .my. Roughly [$10–15/year — VERIFY .com and .my]. Budget [10 minutes]; it's mostly deciding on the name.

Then connect it — this is the two-minute part:

1. In OhhWells: [Settings → Domains → Add domain — GOR: confirm path and wording]
2. Follow the DNS instructions it gives you, at your registrar
3. Done. Usually live within minutes, occasionally up to a day — that wait is DNS, not us.

The security certificate (the padlock) is handled for you.

This is the paid step. [OHHWELLS CUSTOM DOMAIN PRICING — GOR.] Everything before it — building, editing, publishing on a subdomain — is free. You pay at the point you want it to look like your business rather than a test page.

(On the Step 3 path, you'd point the domain at Netlify instead — same DNS idea, and their docs cover it.)

Step 6 — The parts Claude can't do

You now have a real website, live, on your own domain, for the cost of the domain. That's a genuinely good outcome, and for a lot of businesses it's enough.

Here's what it isn't.

Your booking button doesn't book anything

The most common misunderstanding on this page — and it applies to Claude's output whichever path you took.

Claude built you a beautiful booking form. It's a picture of a booking form. No calendar behind it, no availability rules, no confirmation email, no reminder, no reschedule, no payment, no record that the booking happened.

Real booking is software with a database, not a page. Claude can build that too — for a developer, over weeks. Not from one prompt.

If you took Step 3, this is yours to solve: embed a third-party tool (Calendly, or a salon-specific system) or a WhatsApp link. Both work. Both mean your site and your bookings live in different places and you're the integration between them.

If you took Step 4, it's already handled.

Changing it next Tuesday

Your prices change. A stylist leaves. Raya hours go up.

You can fix it — that's worth being clear about. You go back to Claude, explain the site again, ask it to regenerate the page with the new price, download it, upload it again. Fifteen minutes, if it goes well.

The problem is that it's fifteen minutes, and it's always you. Not from your phone between clients. Not delegated to whoever's on the desk. And each regeneration is a small gamble that the bit you liked survives.

This is where AI-built sites quietly die — not at launch, at the first edit. Not because the edit is impossible, but because it's just annoying enough that you don't do it. The Raya hours never go up. The old price sits there for a year.

It's the entire reason Step 4 exists. If you skipped it, this is the bill.

Nothing tells you if it works

Claude has no idea whether your site converts. It can't. It's never seen a visitor land on it.

It will happily give you a hero with no phone number, a booking button below the fold, and three paragraphs nobody reads. It looks professional. It's optimised for looking like a website — that's what it learned from. And most websites don't convert either.

You won't get an error. You'll just get fewer bookings than you expected and no explanation.

Everything after

Uptime. Mobile rendering on the one phone your customer actually uses. Page speed. Someone to call when it's down before a launch.

Not glamorous. All of it the actual job.

What it really costs

The $1/month is true. It's also not the whole number.