Which AI to build a site when starting from zero
You are starting from nothing and every article recommends a different tool. Some talk about code, others about drag and drop, all of them swear theirs is the simplest. You cannot tell which suits someone who has never built anything, and you are afraid of choosing wrong immediately.
Every serious general-purpose AI can produce a decent site today, so raw power is not what separates them. What separates them is what remains in your hands at the end: the code, the address, the data. Choose the exit first, and only then choose the entrance you go in through.
Why “which one is best” is a trap
It is the first thing everyone types, and the least useful.
Every serious general-purpose AI can now produce a decent site from a description. The quality gap between them is real, it shows on hard cases, and it is not your limiting factor when you are starting out. Your limiting factor is your ability to describe what you want.
There is a second, harsher reason: the ranking changes every three months. Choosing on “which is strongest” means choosing on a fact that will be wrong before your site is finished.
What does not change is what remains in your hands at the end. That is what to decide on.
Question 1 — do I leave with my site?
This question outranks every other, and it is the one nobody asks at the moment of choosing.
Some tools hand you a folder of standard files, readable by any developer and servable by any host. Others hand you only part: the appearance yes, the machinery no.
The publishers’ own documentation is the place to check, and it is clearer than people assume — some state explicitly that the code produced is yours on every plan and syncs with an outside code space. The detail of the three exit doors is in Lovable, Bolt or Base44.
Why it comes first: because it is the only one whose answer is irreversible. Everything else can be changed along the way.
Question 2 — am I building, or having it built?
Two families of tools exist, and they address different moments.
Website-specific tools. You describe, an application appears in the browser, hosting is included, you publish in one click. The first screen arrives in minutes. In exchange, your project is born inside a platform.
General-purpose AIs. They write your site’s files on your machine, as someone you employed would. Nothing is included: you choose the hosting, you publish yourself. Slower to start, and what you end up with belongs to nobody but you.
The right criterion is not your technical level — both work without knowing how to code. It is your horizon. A demo to show next week calls for the first family. Something that must run in two years calls for the second.
Question 3 — how much time do I actually have?
A trivial-looking question, decisive in practice.
A specialised tool wins you the first hours. A general-purpose tool wins you the last months. With three evenings in total, the first wins easily. With three months, the advantage flips — and what you learn stays useful, whereas platform-specific know-how evaporates with the platform.
On budget the gap is smaller than people imagine: entry plans sit around twenty dollars a month on either side, and hosting a brochure site is free at several providers. The costed detail is in what a website really costs.
Question 4 — what do I do when it does not work?
This one never appears in comparisons, and it is where the real experience is decided.
All these tools share exactly the same blind spots: they forget your instructions between sessions, sometimes change what you did not ask for, and announce “done” without proof.
None of those three depends on the brand. Switching tools will not make them go away — you will meet them again within the week. What makes them go away is what you put around the tool: written instructions it re-reads at startup, and a way back.
The grid, in one minute
You want to see something tonight, and the project is not meant to last. Take a specialised tool, publish on its address, do not overthink it. That is a perfectly legitimate use.
You are building something that must exist in two years. Get a domain of your own on day one, whatever the tool. It is the decision that matters most and it costs about ten euros.
You are torn between two tools that both let you leave with your code. Try both for one evening on your real project. The free versions are enough, and one evening will teach you more than any comparison — this one included.
You are torn between a tool that lets you leave and one that keeps half. There is no hesitation to have there, as long as your project has a future.