A client recently gave me some honest feedback on a website mockup:

“I have some concerns that if the design is AI-generated, it might look too similar to other websites using the same generator. The gradient and headline formatting, for instance, seem quite standard.”

They were right to ask. I respect that they did, because it’s often hard to tell where the human stops and the AI begins.

Yes, we use AI. We don’t hide it, and we use it with purpose.

Our aim has always been to use the best tools available to get the biggest result for our clients. I’m happy to hand AI the grunt work: first drafts, variations, reading through mountains of data and writing repetitive code. The rest stays with people: understanding the problem, working out the strategy, talking to your team, testing the work and deciding what’s right for your audience.

In practice, I use lots of different AI tools. ChatGPT image generation is decent for mockups, Claude Design and Leonardo have a place for design systems and image generation, and I use other tools for copy and code. But equally, if a project needs a dedicated designer, I have one on retainer. Which path I take depends on the project and the budget.

What you can hold me to

Human accountability. If AI helped make something, I’ve checked it and I answer for it. Nothing goes on your site without a person approving it.

Transparency. If AI has shaped something you’re looking at, such as a mockup, an image or a draft, I’ll say so.

AI amplifies, not delivers. An AI first pass is never the final design. It’s a starting point for a conversation.

Client data protection. Your information stays out of the tools’ training. Everything I use for client work runs on paid accounts with model training switched off, so what I send to Claude or ChatGPT isn’t used to train their models. I don’t put passwords, payment details or personal information about your customers, members or donors into these tools. If a job ever needs that, I’ll talk to you first.

Where AI earns its place

Design mockups are the most visible use, but they’re a small part of it. Here are four others, from the everyday to the serious.

Placeholder content that makes sense

I use your own content wherever I can, and the best sites are built with it. But sometimes I’m building out a page, or a whole website, and the content isn’t ready yet. Then I have a choice: fill the page with grey boxes and lorem ipsum, or use AI to write a close approximation in your organisation’s voice, and sometimes find a placeholder image to match. If it’s almost as quick to do the second, I will. Lorem ipsum can feel cold and makes a custom site look like a template, and plenty of clients struggle to picture a finished site from it. A few images that are close and text that is roughly right let someone view the page and understand how it flows.

It’s clearly temporary and gets replaced with your words and your photos. In the meantime, the layout gets tested against something close to the real thing, and review meetings run faster.

Finding the SEO opportunities

On a website I manage, I run a weekly review that pulls ranking data from SE Ranking (a paid SEO tool), search data from Google Search Console and visitor data from Google Analytics, then looks for patterns. Reading that row by row would take me hours, because it’s thousands of lines of data. The AI does it in minutes and flags what deserves a closer look: a page with plenty of impressions but few clicks, a keyword that dropped, something ranking that we never targeted.

Then I decide what matters. A single-week dip is usually noise, and knowing which movements are real is the part that takes experience. In August, that same process helped me cut a tracked list of 222 keywords down to the 151 that matched what the business actually sells.

Custom plugins that extend what a site can do

For a national retail wholesaler, AI helped me build a custom plugin that extends their membership platform. It lets us display things like a member’s name and the content created since their last login, directly on the pages they use. Nothing crazy, but all really useful data that isn’t normally displayed.

I could have built it myself, but it was far quicker with AI. I tasked Claude Code with reading the documentation for MemberPress and WordPress, finding the simplest ways to connect the pieces, and writing the plugin that would allow me to display it. Then I tested it thoroughly, approved it and deployed it.

Stress-testing a login page

For a public voting site I built for a regional event, we expected thousands of people to log in within a 12-hour period. Under that load the login page is the front door. If it falls over, or a voter can reach somewhere they shouldn’t, the whole event suffers.

So I needed to test two things. The first was speed: consecutive and concurrent hits on the site, to make sure it wouldn’t trip up under the crowd. That’s a test I can’t run by hand. The second was a code and credential review, to make sure a voter could never jump across into the areas competitors or administrators use.

AI did the volume work, the thousands of simulated visits and the line-by-line reading. I read the results, decided what was a real problem and fixed it. Then we ran the tests again and again until I was confident we’d picked up all the issues.

I only use AI to test sites I’ve built and am authorised to test. It’s the same mindset behind the WP-SHELLSTORM campaign I wrote about recently: known weaknesses stay dangerous until somebody fixes them.

In each of these, the AI supplies speed and volume. I supply judgement, testing and accountability.

Why AI mockups can look the same

Let’s go back to the client question at the top of this article, because there are some real limitations to using AI.

The client’s concern has a real cause. AI design tools are trained on an enormous amount of existing design, so left to themselves they gravitate to whatever is most common: the gradient, the bold headline, the row of three cards. A first pass is an average of the web. If that were the finished site, you’d be right to worry about disappearing into a sea of similar ones. The same can be said of virtually any generative AI output: designs, copy, even strategic recommendations.

Here’s what I do instead. I treat the AI’s first pass as a guide to the final design.

Most clients can’t tell me exactly what they want. In fact, most people say that they’ll know it when they see it. You can probably relate to that.

Last week I spent around four hours in Figma building a homepage mockup for one client. We had a style guide, a logo and good photos. I used my eye for design, my knowledge of good practice and design fundamentals, a solid understanding of their clientele, and massaged the design until it was just right. The output: they were happy with the direction, and we refined it with our typical two rounds of revisions. Job done, ready for production.

The same week, another client kicked off a project with less design direction. This time I used ChatGPT to generate five mockups in about 15 minutes. What I had then was five things for the client to react to. We spent a good hour talking through the designs and they pointed to a layout they liked in one, an image treatment in another and elements to avoid in the rest. I combined the strongest ideas into one refined mockup, and we only ended up needing a single round of revisions to polish it up.

It’s like choosing paint. You can spend an hour debating “warm white” against “soft white”, or you can put five samples on the wall and see which one you keep looking at.

I’m not saying that AI is always cheaper, faster, or easier. But getting to the right direction sooner leaves more of the project for the parts that matter: the layout, the user experience, how the site works and whether the message lands.

And does it matter if a mockup uses a popular design trend? In my opinion, less than you think. What makes a finished site stand out is the words and the story. In my experience, visitors don’t stay on a website just because it has an unusual layout. They stay for a clear message that only your organisation could be sending, on a page that’s easy to read. A strong message with an average font will usually outperform a unique font and a bland message.

That’s not to dismiss the concern at the top of the article. I’d just turn it around:

  • Use AI mockups to speed up the initial brainstorming and land on a design direction that feels right.
  • Spend the time that frees up refining the content, the message, the story and the call to action.
  • Have a human (like me) take the AI’s first draft and refine, polish and tweak it for your audience.

That’s how I use AI on client work. Here’s how you might use it in yours.

How you can use AI too

You don’t have to use AI. But if you’re curious, here’s where I’d start.

I get asked every week how to apply some of these practices to other people’s businesses, and my recommendation is often this. Start with the grunt work. Write down the jobs in your week that are repetitive and low-risk: the first draft of a newsletter, tidying up meeting notes, summarising a long email thread, sorting a pile of survey responses. That’s the same kind of work I hand to AI, and the same rule applies. The tool does the first pass, and you decide what’s right.

One word of warning before you start: check the settings. Both ChatGPT and Claude let you decide whether your conversations are used to train their models, so make sure that’s switched off in each. In ChatGPT it’s Settings, then Data controls, then “Improve the model for everyone”. In Claude it’s in your privacy settings. It takes a minute, and it was the first thing I did. Keep passwords, payment details and personal information about your members or donors out of these tools altogether.

If you’re not sure where AI fits in your organisation, that’s the kind of question the Digital Strategy Workshop is built for. You’ll work out what you actually need and leave with a 12-month roadmap. It might include AI, and it might not.

And if you have questions about how I use AI, email me and I’ll tell you honestly what I think.