AI website builders: What they ship, and what they leave you holding

An AI website builder will give you something that looks like a website in about four minutes. That is a real achievement and it is not a trick. The question worth asking is not whether the output looks good — it usually does — but which of the jobs a business website has to do are actually being done, and which have simply been left out of the picture.

What does an AI website builder actually produce?

It produces a visual layout with placeholder-grade copy, generic stock imagery and a working contact form — the easy half, built in about 4 minutes. What it rarely produces is the invisible half: accessible markup, structured data, a sensible URL structure, fast images, and code a human can safely change later.

The half you can see is the easy half

Layout is the part of web design that has been solved hardest. Grids, spacing, type scales and colour systems are well-understood problems with well-documented answers, and a model trained on millions of pages will give you a competent one. Nobody should be surprised that the picture looks fine.

The jobs that decide whether the site earns anything are all in the other half. Does Google understand what the business is and where it operates. Does the page load on a phone on mobile data outside a competitor’s shop. Can a customer using a screen reader complete the booking. Does the enquiry actually arrive in an inbox somebody reads.

None of those are visible in a preview, and none of them are what the builder is optimising for.

On some builders the gap starts before any of that: half the Lovable sites we fetched sent search engines an empty page, which is the problem our Lovable SEO fix is for.

Accessibility: the measurable gap

WebAIM tests the top million home pages every year. In their 2026 report, 95.9% of home pages had detectable WCAG 2 failures — up from 94.8% the year before, reversing six straight years of slow improvement. The average page carried 56.1 detected errors, a 10.1% rise on 2025.

WebAIM’s own list of the most common failures reads like an inventory of things a generated layout gets wrong because nothing in the prompt asked about them: low contrast text on 83.9% of pages, missing alternative text on 53.1%, missing form input labels on 51%, empty links on 46.3%, empty buttons on 30.6%.

Two of those are directly a form problem. If half of the million busiest home pages on the internet cannot label a form field, the odds that a four-minute build labelled yours are not good — and an unlabelled form field is a customer who cannot complete an enquiry.

The content problem nobody warns you about

Most AI builders will also write your pages for you, and this is where a second and much quieter risk sits. It is not a matter of taste or of house style. Google has written this one down in its published policies rather than leaving it to interpretation, which makes it checkable.

Google’s Search spam policies define scaled content abuse as many pages generated for the primary purpose of manipulating rankings rather than helping users, and name the practice explicitly: using generative AI tools or other similar tools to generate many pages without adding value for users. The stated consequence is that such sites may rank lower, or not appear in results at all.

The dangerous version is not one AI-written page. It is the builder that offers to generate a page for every service in every suburb you serve — forty pages, identical but for a place name. That is the shape the policy describes.

What you are actually buying at each price

The comparison that matters is not builder against agency, and it is not price against price. It is what each option leaves you holding twelve months later, when you need a new service page, a changed price and a booking integration. Here is that comparison, job by job.

Where a builder is genuinely the right answer

If you are testing whether a business idea has any pull at all, a builder is the correct tool and a studio would be a waste of your money. Put up one page, run it for a month, see whether anybody calls. That is a good use of four minutes and a small subscription.

The moment to move is not a design decision. It is when the site starts carrying something you would miss: bookings, payments, a phone that rings because of it. At that point the question changes from “does this look like a business” to “can this business survive the site being wrong”, and those need different answers.

How to check yours in ten minutes

Open your site on a phone, on mobile data, away from your own wifi, and try to find the phone number and tap it. Then try to complete your own enquiry form. Then turn images off and see whether the page still says what you do.

After that, run the boring diagnostics rather than guessing. Our nine-point audit checks the specific items WebAIM found most often, publishes the report, and costs nothing. If it comes back clean, you have a good site and you should keep it.

If it does not, what it costs to fix a website AI built sets out the realistic range, and our brackets start at $999.