My first proper job was as a junior web developer, working in a basement (literally) for a small agency doing a job that would be almost unrecognisable in today’s market. In those dark and distant days, instead of going down the pit, a designer would create a website design in some Adobe product or other and web developers like me would build the website, from those design files, with HTML that would often do unexpected things in Internet Explorer (no, Bill, we will never forgive you for IE5.5). We didn’t have to think about mobile phones or tablets, and we weren’t yet talking about user journeys, UX or analytics: our future was as bright as Google’s logo, as the new search engine on the block gathered pace.
Then came content management systems. CMSs brought a big shift in the jobs of designers and developers. We suddenly needed to work with databases, templating systems, and a growing number of languages and platforms. Instead of building fixed pages, we were feeding fragments of HTML into systems so clients could edit their own content – safely (we hoped). Complexity grew. We started thinking about UX, user journeys, accessibility, semantics, and how search engines understood our sites. And that technology has continued to evolve – headless CMSs, modern JavaScript frameworks – but the fundamentals remain the same. Most of the web still runs on CMSs, surrounded by an ever-growing ecosystem of UX thinking, analytics, search, and digital marketing.
Now, though, we have AI. A squillion Instagram posts and TikToks will tell you that this means websites can be made almost instantly, with the minimum of effort, and at a fraction of the cost. Marketing strategies around the AI gold rush platforms tell us that they can now design your logo, write your content, generate your images and produce your layout all with an unhealthy dose of congratulations for taking the right strategic approach. The question most organisations are exploring, especially if they are about to embark on a new web project, is are the claims being made about AI in web deisgn and development accurate?
And the answer isn’t simple, as AI itself will tell you.
As a consultant and all round digital bod, AI is genuinely very useful to me. It is particularly good at taking on some of the grunt work of repetitive tasks in coding. It’s great at turning a blank page into a non-blank page, at suggesting alternative approaches, helping to structure and rehouse content that is currently living across six Word documents, an Excel spreadsheet and an email from 2019. It helps me move more quickly – generating ideas, generating code, suggesting layouts, brainstorming brand approaches and generally reducing a not insignificant amount of staring into the middle distance.
For site owners, content editors and web folk alike, AI is very good at getting us started. It is very good at reviewing our progress. It is very good at coming up with alternative approaches, generating ideas and filling in the gaps. It came up with some great ideas to explore in this post. And a few terrible ones.
What AI is not so good at
We all know about AI hallucinations. We’re likely used to those surprisingly over-confident moments where AI tells us there really are only 2 As Alabama. Applied to code, you might find yourself trying to reference an entire library that has never existed or a plugin perfect for your purposes that has never been written. We see the big mistakes, but we may not see the smaller mistakes or the gaps, particularly in a field we don’t know. So how does this translate to web sites designed and developed to a greater extent by AI? As web bods, we sometimes see it in the most obvious places: with logos that look amazing full size on the brightest Macbook screen the marketing department could afford, but look like a bowl of soup when presented as a mobile logo on last year’s mid-range Samsung. In other ways, it becomes harder to articulate. We know something isn’t quite right, but it takes time to understand – and articulate – what. It’s like watching a very good but still obviously AI video: you know something is off but it’s not always immediately obvious what is.
At the root, AI does not completely understand your organisation or your customers and may not know have all the questions to ask in order to. It probably doesn’t spend enough time staring into the middle distance. It won’t wake up at in the morning with a Eureka idea for your project. It won’t call you after that third cup of coffee with a random magazine open at an article about yarn bombing to tell you about the great marketing strategy it brought to mind. And all of this is before the AI has filled the internet with its own content, and the learning material itself becomes even less… well, human.
Because AI has learnt from all of the written understandings of other people’s organisations and customers, as well as wrongly learnt some of the misunderstandings, it will often just not quite get the subtlety of your organisation or know the questions to ask to find out how you tick.
The role of challenge
A big part of my role is to challenge my customers and to advocate for their customers. While it’s more clearly my job to look at the bigger questions – who are you trying to attract? What’s changing in your strategy? What keeps the money coming in? – it’s also my job to look at a hundred much smaller questions and to challenge the bad ideas. It’s my job to, for example, point out that most of your visitors won’t care which department or committee curated a given programme of events. Your customers are more likely to care whether The Moscow Project is a jazz gig, a talk on Russian politics or an exhibition of post-modernist paintings than whether it was the Languages Committee or Global Folk Music Group that made it happen. The subtlety in how we present that is key – we certainly don’t want the Communist Brutalist Architecture Committee to be invisible, but we may want gently lessen their visual weight compared to other more use friendly approaches – and that subtlety might well be missed by AI, which can be far more prone to over-praise than erudite challenge. You might thing these differences seem small, but what if you were to miss a dozen, two dozen even, of these challenges? What if it risks changing your whole approach to menus and AI? To who you sell things to? Your whole visual identity, even.
And this is where the problem lies as we learn how to use this incredible, emerging, flawed technology. A website isn’t just a set of landing pages: it’s a system. And we still need experienced eyes and ears to know what to ask – and how to make that system work – for a wide range of users, journeys, and purposes. We need to understand subtle differences between apparently similar organisations and their apparently similar customers. And as AI increasingly becomes the gatekeeper of what users are shown – blurring the line between search engines and AI-generated answers – your distinctiveness, your expertise, and your character are going to matter more, not less.
