Sites a machine builds, and where we stop it

What the engine does

It produces a complete, structurally sound site from a brief: page structure, semantic markup, responsive layout, meta tags, structured data, sitemap and an accessible colour treatment. It does the work that is identical on every project and unrewarding to do by hand for the fourth hundredth time.

What it does not do

It does not decide what the business should say, which is the thing that actually determines whether a site works. It does not judge whether the result looks right, because it has no taste. And it never publishes anything on its own.

Every generated site is reviewed by a person before it goes anywhere near a client. That is not caution for its own sake: automated output fails in specific, recognisable ways: copy that is fluent but says nothing, an image that is technically correct and completely wrong for the trade, a layout that is fine except for the one thing a human notices immediately.

How that changes the price

It is the reason a properly built site starts at £500 rather than £2,000. The automation removes the repetitive hours; it does not remove the judgement. You are paying for the judgement, and paying much less for the typing.

Where the engine has got it wrong

Specific failures are more useful than a general claim of caution, so here are the recurring ones.

Generated copy is consistently fluent and occasionally empty, a paragraph that reads well and commits to nothing. It is the hardest failure to spot because nothing about it looks broken. Imagery is the second: technically appropriate and tonally wrong, which a person catches instantly and a machine does not. And structure is the third, a sensible page order for a generic business can be exactly wrong for a trade whose customers arrive with one urgent question.

We have also caught our own automated checks being wrong, which is worse than having none if you trust them blindly. A contrast checker on this site once reported the primary buttons as failing badly; the fault was in the checker, not the buttons. The rule we work to now is that when a check suddenly passes after you changed the check, you have to prove the new result is correct rather than simply enjoy the quiet.

This site is the proof

The pages you are reading are generated from a single content source by our own build tooling, and every one of them has to pass an automated gate before it can be published: exactly one heading, valid structured data, every internal link resolving, measured colour contrast, no layout overflow at phone width, a minimum amount of real content, and a check that no two pages are near-duplicates of each other.

That last check exists because mass-produced location pages are the most common way an agency quietly damages a client's domain. Our own pages are measured against it, and the worst-matching pair on this site sits at around two and a half per cent similarity.

Related case studies