AI and web development: what the evidence actually says

Ask ten people in web design what they think of AI and you'll get ten different answers, delivered with roughly the same level of certainty. To some, it's the biggest shift the industry has seen since responsive design. To others, it's a fast way to produce mediocre websites at scale. Both camps can point to evidence. That's usually the sign of a genuinely mixed picture rather than a simple one.

So rather than take a side, it's worth looking at what's actually been measured — what AI tools are good at, where they tend to fall down, and what that means if you're trying to make sense of the hype.

How AI is actually being used in web development

"AI in web development" covers a few quite different things, and it's easy to conflate them. There are code assistants that suggest or autocomplete snippets as a developer types. There are AI website builders that generate a full layout, and sometimes copy, from a short prompt. There are AI tools for writing content, generating images, and running chatbots once a site is live. Each has a different track record.

Adoption of these tools, especially among professional developers, has moved quickly. Stack Overflow's 2024 Developer Survey — one of the largest annual surveys of working developers — found that 76% were using or planning to use AI tools in their development process, up from 70% the year before.¹ A controlled study by GitHub found that developers using its Copilot assistant completed a sample coding task 55% faster than a control group working without it.²

That last figure, from McKinsey's 2024 survey on the state of AI, points to something worth noting: this isn't a niche experiment confined to hobbyists.³ Gartner has predicted that by 2026, more than 80% of enterprises will have used generative AI APIs or deployed a generative AI application, up from under 5% in 2023.⁴ Whatever view anyone takes on the technology, the adoption curve is real.

The case for it

The strongest argument for AI in web development is speed. Repetitive, well-understood tasks — boilerplate code, first-draft layouts, generating a batch of alt text, restructuring a paragraph — are exactly what these tools are built for, and the productivity numbers above reflect that. For a small business with no in-house developer, an AI-assisted builder can also lower the barrier to having some kind of web presence at all, which for a business with literally nothing online is better than the alternative.

It's also fair to say the tools have improved fast. Early AI-generated code and copy was easy to spot; the more recent generation is harder to distinguish from work a person did by hand, at least on the surface.

Where it falls short

The concerns aren't hypothetical either, and a few are backed by research that's specific enough to take seriously.

On security: a study from Stanford and NYU asked developers to complete a set of programming tasks, some with access to an AI coding assistant and some without. The group using the assistant wrote measurably less secure code across most tasks — and were more likely to believe their code was secure than the group who didn't use it.⁵ That combination — lower quality plus higher confidence — is arguably the more concerning finding of the two.

On accessibility: WebAIM's annual "Million" report, which audits the home pages of the top million websites, has found for several years running that the vast majority — over 90% — contain detectable accessibility failures.⁶ That's a long-standing problem across the web generally, not something AI invented. But it does mean that generating a page faster doesn't, on its own, generate an accessible one — the underlying issue needs deliberate attention regardless of how the site was built.

On search visibility: Google has been explicit that it doesn't penalise content simply for being AI-assisted, but it does continue to penalise content — AI-written or otherwise — that's judged to be low-value or produced primarily to manipulate rankings rather than help a reader.⁷ In practice, that puts the pressure back on quality and originality, which is a harder thing for a generic tool to guarantee than speed.

So is it a good thing or not?

Genuinely, it depends what it's being used for. As an assistant for a developer who already knows what "good" looks like — someone who can read the suggested code, question it, and reject the parts that don't hold up — the evidence for productivity gains is solid. As a replacement for that judgement entirely, the evidence is a lot shakier, particularly around security and accessibility, where the mistakes are the kind a visitor won't notice but a business will eventually pay for.

It's also worth separating "AI was involved somewhere" from "AI made every decision." Most professional use of these tools today sits closer to the first — a person using a fast tool, then checking its work — rather than the second.

What it means if you're weighing up a website

If you're a business owner trying to work out whether a website — or a company offering to build you one — has used AI well or badly, the technology itself is the wrong thing to focus on. A few more useful questions:

• Does the design actually look like your business, or could it be swapped onto any other company's site without anyone noticing?
• Has the content been checked for accuracy — prices, claims, and details that are specific to you, not generic filler?
• Has anyone tested it properly — on a phone, with a screen reader, on a slow connection — rather than just looked at it on a large monitor?
• Is someone accountable if something breaks, or is it precisely as good as the day it was generated and no better?

Those are the questions that determine whether a website actually works for the people using it. The tooling behind it, AI-assisted or not, is a means to that end rather than the end itself.

— Jamie, Mortar & Pixel