Building a Website with AI: Why Good Design Still Comes First

AI has changed how quickly a website can go from idea to working build. A simple prompt can now produce layouts, components and functioning code in an afternoon. That's genuinely useful. But it has also created a misconception: that quick means good.

The real value isn't in how quickly you can build something. It's in knowing what to build, why you're building it, and how to make it work properly. Without the right foundations, AI can just as easily accelerate poor decisions as it can good ones. Here's the process that keeps speed working in your favour.

Why start with the brand, not the prompt?

Because the brand decides what the website should look, sound and feel like. AI can generate a logo or a colour palette, but that isn't a brand. Without clear positioning, audience, personality and tone, an AI-built site defaults to predictable layouts, interchangeable typography and generic imagery that could belong to anyone.

A cohesive identity sets the direction for every decision that follows: how the business presents itself, how it speaks, and what makes it different. If brand guidelines already exist, they're the starting point. If they don't, building the brand first will save far more time than it costs.

Why build a design system before building the website?

A design system turns the brand into reusable rules, so every page is built from the same parts. It covers design tokens (colour, typography, spacing and sizing) and components (buttons, forms, navigation and cards). Without one, small inconsistencies accumulate quickly, especially when elements are generated independently, prompt by prompt.

A system gives developers and AI tools the same framework to work from. It isn't about making every page identical. It's about making good decisions repeatable.

How do you choose the right technology for the job?

Start from what the website needs to do, not from whatever stack an AI tool defaults to. The questions that matter:

• Does the team need a CMS to manage content?
• Which third-party integrations are required?
• Will it handle bookings, payments or user accounts?
• Does it need a database, and how much does it need to scale?

These choices shape dependencies, ongoing maintenance and running costs for years. The right tools aren't the most advanced or the fastest to prototype. They're the ones that fit the job.

Why design the experience before generating the interface?

A prompt can produce a homepage, but it can't supply the context behind it. Good UX starts with questions only people close to the business can answer: who is the audience, what do they need, what stops them, what's the one action that matters most, and how does intent change from page to page?

Working through the sitemap, user journeys and key components in a design tool like Figma, before any build, is where human judgement adds the most value. It's far cheaper to change a wireframe than a finished build.

How should AI fit into the build itself?

Give AI a system to work within, not a series of disconnected prompts. The most reliable results come from supplying the design tokens, component library, layout rules, approved designs and sitemap up front, then building foundations first: global styles, typography, spacing, components and page layouts.

Every output still needs reviewing. AI can quietly introduce unnecessary complexity, duplicate code or regressions in parts of the site that were already working. AI can help you build faster. A well-defined system helps ensure you're building the right thing, consistently.

How do you make a website manageable after launch?

Plan the content model as part of the architecture, not as an afterthought. There's a difference between a site that's technically editable and one that's genuinely manageable. A headless CMS like Sanity works best when content types, reusable structures, relationships and editorial workflow are designed alongside the front end.

Seeding the CMS with realistic content before launch also exposes problems early: layouts that break with long titles, missing fields, or structures that don't match how the team actually works.

What should you test before a website goes live?

Test the experience, not just the code. Automated tools catch a lot, but manual testing on real devices catches what matters to real visitors.

Why is launch a milestone, not the finish line?

Because a website only proves itself once real people use it. After launch, the work shifts to monitoring performance, analytics and integrations; maintaining dependencies; improving accessibility (opens W3C in a new tab) and security; and optimising for conversion.

The best improvements are evidence-led: based on how visitors actually behave rather than assumptions made before launch.

So, can you build a website in ten minutes with AI?

For a prototype, yes. But getting something on screen isn't the same as building something fit for purpose. Skipping brand, systems, experience design and testing tends to lead to inconsistency, technical debt and costly rework later.

AI isn't the problem. Treating the build as a substitute for the thinking is. Used where it genuinely adds value, AI makes good teams faster.