AI is the best thing to happen to small business web design.

The work that used to eat the budget

For most of the last decade, a large share of any small business web project went on work the client never saw and would have been annoyed to find itemised. Boilerplate markup. Wiring a contact form. Rebuilding the same layout four times for four screen widths. Chasing a browser quirk. Exporting every image at six sizes.

None of that is design. It is assembly. It is genuinely necessary, it consumes real hours, and it produces nothing a customer will ever notice or thank you for.

That is the layer AI compressed, and it compressed it hard. Work that used to fill three weeks now fills a few days. The interesting question is not what that saves — it is what it frees up.

Days rather than weeks changes what you can afford

The obvious reading is that sites got cheaper. Sometimes they did. The more useful reading is what the freed budget now buys.

A project that spends seventy per cent of its hours on assembly and thirty on thinking produces a site that works. Invert that ratio and you get a site that works and also says the right thing to the right person. Same budget, different result.

Strategy, proper photography of the actual premises, copy written the way the business actually talks — these were always the first things cut when money got tight, because assembly could not be cut. It had to happen or there was no site. Now it is the cheap part.

You can iterate before you commit

Under the old economics, changing direction once the build started was expensive. So you committed early, choosing a direction from a static mockup and a moodboard, and then lived with it.

Building a real, working version of two different directions is now cheap enough to do before deciding which one is right. Clicking through something real — at the actual speed, on an actual phone, with the actual words — is a far better basis for a decision than approving a picture of a website.

That is a quality improvement, not just a speed one, and it is the one clients tend to notice most.

The checks nobody used to run

Accessibility and performance auditing used to be specialist work, billed separately, and quietly dropped from small budgets. It was the line item that got cut in the second meeting.

Much of it is now automated and runs on every build rather than once, nervously, the week before launch:

• Colour contrast ratios against WCAG thresholds
• Missing or unhelpful image alt text
• Heading order that skips levels, which breaks screen reader navigation
• Keyboard traps and focus states that go invisible
• Core Web Vitals, image weight, and render-blocking resources

Being fair about what those checks catch

Automated accessibility testing catches somewhere between a third and a half of real problems. It is very good at the mechanical ones and blind to the rest. It will not tell you whether somebody using a screen reader can actually complete your booking form, because that requires a person to try.

So it is not a substitute for testing with real people, and anyone selling it as one is overstating it. But a third of the issues caught automatically on every single build is a great deal better than a thorough audit that never made it into the budget. The floor moved up even if the ceiling did not.

What AI did not change

It still cannot decide what your business should look like to your customer. It can produce something in the general shape of your industry — the layout your competitors have, the stock photography your competitors use. Deciding that a family law firm should feel calm rather than combative, and then working out what calm means in type, spacing, colour and photography, is judgement about one specific business and one specific customer. That is the part that was always the actual work.

It also cannot tell you what you are really selling. Most small business websites fail in the first screen, not in the code — a headline that could belong to any competitor in the country, and no clear reason to keep reading.

And left to itself it will cheerfully commit an API key to a public repository, which is the subject of the companion piece to this one.

The standard went up, not down

Ten years ago a small business could reasonably ship something slow and templated, because doing better genuinely cost more than it was worth to them. That was a fair trade at the time.

That excuse has gone. Fast, accessible, properly structured is now the floor rather than the ceiling, and a site that misses it is not saving anybody money — it is just behind.

Which means the thing that separates one site from another has moved. When competent execution is affordable for everyone, what is left is judgement: knowing what to say, to whom, and what the business should feel like when someone arrives. AI lowered the cost of meeting the standard, and in doing so quietly raised what counts as meeting it.