There are dozens of tools for building websites with AI and their marketing is interchangeable — a website in minutes, all of them. Trying each takes real time and leaves you no clearer.
A more useful division is by what you end up holding. There are four categories, suiting very different situations. Tools will change; the categories will not.
1. Chat assistants — Claude, ChatGPT
Not website tools, but widely used as them, and they work.
What you get: code. You describe what you want and receive HTML, CSS and JavaScript, which you save as files and put somewhere yourself.
Suits: anyone willing to touch a little technical detail; unusual requirements templates cannot cover; or an existing site where you only want to change part of it.
Does not suit: anyone who wants nothing to do with files or servers. It handles neither deployment nor domains.
The most undervalued use: not writing code with it but working out what you need. Have it interview you until it can describe your business. That step is often worth more than the code — we wrote separately on how to do it.
2. AI prototyping tools — Lovable, v0, Bolt and similar
A category that appeared in the last couple of years: a chat panel beside a live preview that updates as you talk.
What you get: a working project, usually with one-click deployment. Say “make the button blue” and the page changes immediately.
Suits: prototypes, testing an idea, internal tools, small marketing pages. Fast feedback, visible changes, very low cost of being wrong.
Where they end: easy to launch, hard to maintain. Generated projects usually depend on a full modern front-end framework, pages are heavier than necessary, and the SEO-related parts — structured data, multiple languages, sitemaps — depend on you knowing what to add.
Also worth expecting: things slow down and tangle around the fifth or sixth revision, because each change layers onto the last and nothing consolidates the structure. That is the curve for this whole category rather than a fault of any one product.
3. AI features inside site builders — Wix, Squarespace, Shopify
Established platforms with AI attached: answer a few questions, receive a site, then edit it in their existing editor.
What you get: a complete hosted service. Domain, hosting, SSL, backups and an admin panel, none of which you manage.
Suits: anyone wanting no contact with technical detail, with standard requirements, who wants to be live quickly. A perfectly reasonable choice.
Two things to know first:
One, you cannot take it with you. Sites here generally cannot be exported intact, so moving means rebuilding. Ask before committing what you would receive if you left.
Two, the platform sets your optimisation ceiling. Speed, structured data, caching — you adjust whatever switches exist and no further.
What AI does on these platforms is really automating template selection and content filling. Do not expect it to match custom design.
4. AI coding assistants — Claude Code, Cursor and similar
What you get: an assistant that operates on your project files directly. It reads the codebase, edits multiple files, runs commands, reads errors and fixes them.
Suits: people who already have a codebase, or complex functionality. This category is not comparable to the first three — it handles changes like “add multilingual routing across the whole site”, not “generate a page”.
Does not suit: anyone with no technical background. It assumes you can follow what it is doing and know when to stop it.
This is what we use daily. It requires guards around it, though: automatic checks before changes land, automatic verification after, version control and one-command rollback. Without those, moving fast only means breaking things faster — we wrote about an outage caused exactly that way.
Choosing
Three questions, each pointing at a category:
How do you intend to maintain it?
Not at all, let a platform handle it → category three.
Happy to republish after each change → one or two.
Technical yourself or have someone → two or four.
How much does this site matter to the business?
Internal, experimental, personal → any of them, pick the fastest.
Taking enquiries, representing the company → what happens after launch matters far more than how it was generated, so choose something you can control long-term.
How much time will you spend learning?
None → three.
A few hours → two.
Ongoing → one or four, where the ceiling is highest.
Four things no tool will do for you
Common blind spots across all of them, because getting each wrong raises no error:
Confirm the content is really in the HTML. If key content is rendered by script, search engines and AI cannot read it. Use view page source — not inspect element — and search for text you can see.
Confirm nonexistent addresses return 404. Many generated sites return 200 for anything, which is a soft 404 and drags on the whole site’s search performance.
Confirm images are sized sensibly. Uploading photos straight off a phone is the most common performance problem; one 5MB image adds seconds before anything appears.
Confirm you have backups and can roll back. How do you return to the previous version? Answering that before an incident costs considerably less than answering it during one.
We covered these four at length elsewhere, each with a check requiring no technical background.
A practical route
If you have not started, our suggestion: use category one to work out what you need, category two to produce a prototype you can look at, and then decide whether to continue yourself or hire someone.
That takes a few days and tells you what you actually need. Arriving at a conversation with a prototype beats arriving with “I’d like a nice website”, whoever ends up building it.
We welcome clients who arrive with a prototype. Talking it through first is more useful than quoting first, and that stage is free.