We get asked this most weeks now, usually by someone who has already tried. So here is the answer up front: yes. Someone who has never written code can produce a presentable website in an afternoon with Claude or ChatGPT. That is ordinary now, not remarkable.
What sits between “produced” and “usable” is larger than most people expect.
What AI genuinely does well
Turning an idea into something you can look at. Say “I want a dental clinic site, clean, professional, with bookings” and a few exchanges later you have a page with layout, colour and copy. Two years ago that was two days of a designer’s time.
A few other things it does more reliably than most people: a serviceable first draft of copy, turning a rambling description of a business into a clear list of services, generating ten headline options to choose from, translating between English and Chinese without the result reading like a translation.
We use all of them daily.
The trouble starts after launch
A website is not a deliverable, it is something that has to keep running. AI-generated sites tend to fail in the same few places, and none of them are visible from looking at the page:
Load speed. Generated code tends to pull in a whole framework, a few hundred KB of icon fonts and uncompressed images. The page is quick on your laptop and takes five seconds on a customer’s phone over 4G. Most visitors decide whether to stay inside three.
Whether search engines can read it. Heading structure, structured data, hreflang across languages, canonical tags — none of this changes how the page looks, and all of it determines whether Google and AI assistants understand and recommend you. In generated pages it is usually missing or wrong, and being wrong produces no warning.
The cost of the second change. The first version is fast. The third revision is slower. By the fifth, fixing one thing breaks another, because nobody is maintaining a structure — things are only being appended.
Forms, payments and bookings, which is where real damage happens. They involve other people’s money and personal information. A form without abuse protection fills with spam within a fortnight. A payment flow handled carelessly leaves you explaining yourself to a customer and a bank.
A more useful question
Rather than “can AI build a website”, ask: when this site breaks, who answers for it?
If it is your own project and failure only affects you, building it with AI is a sound decision and a cheap one. Portfolios, small event pages, internal tools all sit here.
If the site has to take enquiries, take money, or represent the company to a customer, the question becomes: will you still want to look after it in three months? Who renews the hosting? Who investigates when Google quietly stops indexing it? Your chat history will not help you on any of those days.
Use AI first even if you plan to hire someone
We recommend this sincerely, even when the work ends up being ours.
Before you contact any web company, spend half an hour with an AI getting clear on: who my customers are, what they came to my site to resolve, what I want them to do, how my competitors handle it, and where I differ.
Tidy that conversation up and you have a brief. Arriving with one instead of “I’d like a nice website” saves weeks. You will also find that several features you assumed you needed are unnecessary, and that something you had not considered is the important part.
AI is badly undervalued at this stage. It is poor at deciding for you and exceptional at forcing you to say what you mean.
How we use it
AI handles requirements analysis, first-draft content, bilingual work, the first pass of code, and repetitive checking. It compresses projects that used to take weeks into days, which is why we can commit to three-day delivery at all.
What we do not hand over: structural decisions, the performance budget, the SEO and GEO foundation, how payments and forms are secured, and monitoring after launch. We check those by hand, because when they are wrong nothing raises an error — you simply lose customers quietly.
AI makes us faster. It does not make us less responsible. The speed comes from the tool; the reliability still has to come from people.
Where that leaves you
If you want to try, try. Building a site with AI is worth doing for its own sake — you come out knowing what you want and which parts are hard.
If the site then matters enough to your business, that is the point to discuss who maintains it. You will know what you are buying by then, which suits both sides. Everything we have built is in our client work — all of it live, all of it one click away.