Can AI Design and Build a Website (September 2026 Edition)

Today, AI can design and build a website, but not on its own or without careful management from someone with expertise in building websites. It’s more capable than most people in my industry want to admit. But what comes out depends entirely on who’s operating it. Someone who understands web standards, accessibility, UX/UI conventions, SEO, conversion rate optimisation and brand positioning will get results that will match or exceed a fully manual process. So it needs more than 1 operator; it needs a multidisciplinary team. Without a high level of expertise, you will get something that might look finished but simply isn’t.

Notice how neatly that suits me. “Great tools, you just need experts” is exactly what you’d expect from a man who sells expertise! So don’t take my word for it; much of what follows is backed by research.

Credit where it’s due

Two years ago I’d have told you AI design output was a novelty. That’s not true anymore, and pretending otherwise would be daft. It’ll now generate a page layout that respects a grid, a usable component library, copy that isn’t glaringly machine-written, and a working front end, all from a screenshot. 72% of designers use generative AI, and 91% of those leaning in say it improves the quality of their work. This is not a piece about AI being rubbish – of course it isn’t. It’s about what happens after the good-looking bit.

In recent times, we have been approached by a few clients who have come to briefing calls armed with website mockups created in Claude Design – I will admit this freaked us out, but only until we looked a bit closer. The visual quality wasn’t bad; at best, they were a very valuable guide and conversation piece and in fairness to the clients, this was their purpose. They were a great briefing reference; we use AI design for this very reason ourselves.

Why every AI website looks the same

Any custom-designed website is made up of thousands of tiny decisions made across a team, from designers to developers, SEO and marketing, brand and sales, not to mention the users’ and clients’ input across research, process and delivery. That’s what AI is missing: human decision-making, individual creativity and diversity. Every output looks good but not great, and everything starts to look identical. A study in Science Advances gave writers random access to generative AI, then had 600 separate people score the outputs blind. The AI-assisted work scored higher for creativity, but the stories were also far more similar to each other – everything was just a bit samey or vanilla.

That was short fiction, not websites, so I’m applying it by analogy. But it matches what we see. Your site doesn’t need to be good in isolation; it needs to be better than your competitors, and if you know what a typical AI-built website looks like, you know. It’s the nuance that makes you credible online, and this builds trust; your brand needs to shine in a way that is unique to you.

The prompt is not the skill

The quality of AI-assisted design and development comes down to the quality of the prompting. Garbage in, garbage out. AI answers what you ask, not what you should have asked, and knowing what to ask is the whole job. Harvard Business School researchers put it well: “AI makes you feel like you can do anything. But can you do it as well as people whose job it is?” They ran a field experiment with 78 workers and found AI only got people close to expert quality where they already had expertise in the area. People further from the domain couldn’t match expert quality, because they couldn’t judge what they were being handed. That’s taste, and taste is recognition built over years of seeing what works and what doesn’t.

A multidisciplinary team will say no, challenge each other and ask the questions a prompt never thinks to. Put AI in the right hands, and you get something unique and effective. Put it in the wrong hands, and you get whatever the model felt like giving you, filling in the gaps itself, right or wrong.

Standards are slipping

Quality and standards are key; a good designer knows the conventions and knows when breaking them is the right call. We see accessibility standards falling at a time when the EU is enforcing them. WebAIM carried out a recent scan of a million homepages and over 95% now fail WCAG accessibility standards, up for the first time in six years, and they name AI-assisted coding as a likely cause.

One of the areas AI is strongest at is coding. But again, it needs human oversight. Veracode recently found 45% of AI-generated code fails OWASP security testing, and one study found five rounds of prompting to fix flaws left the code with 37% more critical vulnerabilities than it started with. AI is also far from capable at deployment and everything after it, and this one’s from experience rather than research. Hosting, DNS, environments, redirects, analytics, handover, cookies and support. We’re nowhere near AI agents being trusted to handle this delicate and important work on their own; the room for error is too high.

Where we use AI, and where the time goes

We were never in the business of cheap websites, the kind driven by templates and corner-cutting. Same today. We’re not in the business of single-prompt websites either. We still craft custom work using the same mix of research, brand positioning, stakeholder engagement and performance metrics. It’s AI-assisted now, but the process hasn’t changed.

Chatbots help with research, copy and strategy. Figma and Claude Design assist with wireframes, ideation and design system management. Claude Code helps with development and QA. But AI is the assistant, never the creator. Everything runs through our own SOPs, with guardrails, human moderation and constant correction along the way.

And yes, it saves time. The work is better for it too. But the time saved hasn’t gone back in the pot; it’s shifted into user engagement, prompting and re-prompting, iterations, QA, audits and validation. AI raised the standard; there is work in keeping it there.

So can most people with a decent level of AI expertise build a website? Yes. But there are websites and there are websites. Can they deploy it? Will it pass accessibility standards? Does it have the content and schema to be found in search and by LLMs? Is it tasteful, is it unique, and will it still stand up twelve months later when the content’s changed and the traffic’s grown?

As I mentioned in my article on the matter from last May, clients work with us for expertise, experience and trust. That hasn’t changed. Take the experts out, and it’s a race to the bottom. If you’re building something that matters commercially, the tools aren’t the risk. Going in without anyone who can judge the output is. Ask me again in three months; this stuff moves fast!

Get in touch with us today if you are at the start of your website journey. We can help.