What's genuinely automatable now, what still isn't, and why the honest answer to "which parts" matters more than the marketing version of it.
Short answer: the mechanical, time-consuming parts of a simple site — a first-draft layout, copy written from real facts, picking a design that suits the trade — can now genuinely be automated well, when the tool doing it is built to refuse inventing what it doesn't know. What still needs a person: bespoke brand judgement, complex custom functionality, deciding which facts about a business actually matter, and a final honest check that nothing on the page is wrong.
Every claim about "AI and web design" sits somewhere between two extremes — "it's already replaced designers" and "it's a gimmick that'll never really work" — and the honest answer is neither. Here's what's actually changed, and what hasn't.
A person picks a template and writes every word by hand, or doesn't.
A model helps write or rewrite copy a person still directs and assembles.
A full first draft built from real facts, reviewed by a person before it ships.
The real risk in AI-generated web content was never that it would look obviously artificial — it's that a model asked to fill in gaps will often invent something plausible-sounding rather than admit it doesn't know. On a business's own website, that's not a stylistic flaw, it's a false claim with the business's name on it: a service they don't offer, a price that isn't real, a testimonial nobody said. Whether a generator is built to refuse that outright, the way ours is, is a bigger difference between tools than any amount of visual polish. See our guide on how AI website generation actually works for how that gets enforced in practice, not just claimed.
Not entirely — it's already replaced a lot of the first-draft, template-assembly work for simple sites, but judgement calls about brand, bespoke functionality, and genuinely knowing a specific business still need a person, and someone still has to check an AI-written page is actually correct about the business.
A model inventing something about the business that isn't true — a service it doesn't offer, a price that isn't real, a testimonial nobody said. A generator built to refuse that and flag unknowns as placeholders instead is solving the actual risk; one that fills gaps with plausible-sounding invention is creating it.
It can vary layout and style meaningfully within a template-like structure, and do so faster than a person starting from a blank page. A fully bespoke, from-scratch design built around a specific brand identity is still more reliably a human designer's work, at least for now.