What parts of web design has AI genuinely sped up?
Turning a brief into a structure used to take a morning. You would read the notes, work out what the business sells, decide what needs its own page and sketch a sitemap. That step is now minutes rather than hours, and the first pass is usually close enough to argue with. Arguing with a draft is much faster than starting from a blank page.
Code is the other obvious one. Custom blocks, template changes, migration scripts and the one off fixes that used to be quoted as half a day now take about an hour. Small things that were previously not worth doing, like tidying a badly named set of fields or rewriting a fiddly bit of markup, now get done because the cost of doing them fell below the cost of explaining why they were left alone. The work that disappeared was not the interesting part. It was the typing.
How does AI change writing website content?
The useful change is not generation, it is restructuring. Most small business websites have the right information in the wrong order, buried under three paragraphs of preamble. Taking existing material and reshaping it into something a visitor can scan is work AI does quickly and well, because the substance already exists and only the arrangement is wrong.
Generating content from nothing is where it goes flat. Ask for a services page with no input and you get the average of every services page ever published. It is grammatical, plausible and says nothing a competitor could not also say. The raw material still has to come from the business.
Can AI design a layout?
It can produce variations, which is genuinely useful. Seeing six versions of a hero section in the time it used to take to build one means you explore more options before committing to any of them. Weak ideas get discarded earlier, and the version that survives has usually had to beat something rather than simply being the first thing anybody drew. That is a better process than the one it replaced.
What it cannot do is pick. Choosing between six competent layouts depends on who the visitor is, what they are anxious about, what the business wants them to do next and how the rest of the site will feel. That is a judgement, and the model has no stake in the outcome. It will happily recommend the third one if you ask, with reasons that sound right either way.
What has AI changed about auditing a website?
This is the change that gets least attention and matters most. Auditing used to mean sampling. You checked twenty pages on a two hundred page site, found the patterns, and assumed the rest were similar. Most of the time that assumption held. Sometimes it hid the one page with a broken canonical or a heading structure that made no sense.
Now every page gets read. Every title, every heading order, every internal link, every bit of duplicated copy, every page that still advertises something the business stopped offering two years ago. The result is less dramatic than a list of big problems and more useful, because the recommendations cover the whole site rather than the part that happened to be inspected. That is what an audit was always supposed to be.
- Full coverage instead of a sample of pages.
- Inconsistencies in tone, naming and claims across a whole site.
- Structural problems that only show up at scale.
- Stale content describing services that no longer exist.
What has AI not changed about web design?
Deciding what a business should say. That comes from understanding how it actually wins work, which usually contradicts what the owner puts on the homepage. The thing that convinces customers is often specific, slightly awkward and not obviously marketable, and it takes a conversation to get at it. No amount of processing power substitutes for that conversation.
Judging whether a page converts is the other one. A page can be well written, well structured and correct, and still fail because it asks for too much too early or answers the wrong worry. That judgement comes from having watched a lot of pages succeed and fail. It does not come from pattern matching against the average of the internet.
Taste belongs in the same category. Knowing when a layout is too busy, when a headline is trying too hard, when to leave a section out entirely, none of that can be specified in a prompt because the answer depends on everything else on the page. Ask for an opinion and you get a confident one, but confidence is cheap now and being right is not.
Does AI make web designers unnecessary?
It makes slow designers unnecessary, which is not the same thing. The parts of the job that were about effort are now cheap, so the value sits entirely in the parts that were always about judgement. Taste, restraint, knowing which trade off matters on this project for this business, and being willing to tell a client that the thing they asked for will not work.
The practical effect is that more of a project is spent on decisions and less on production. That is a better use of everybody’s money. It also means the gap between a considered site and a generated one is wider than it was, not narrower, because the generated one is now available to everybody.