AI Can Make Hundreds of Designs. You Still Have to Choose.
AI won't one-shot the right design, but it can one-shot hundreds. The work that matters is still the design process: a clear brief, lots of exploration, and the judgment to pick what fits the business.
When I started redesigning my own site with AI, I expected what most people expect: a good prompt should get you a good design in one shot.
It doesn’t. And even when a single prompt gave me something worth using, it was still close to the mean. It looked like what the model thinks a website should look like.
What worked was a lot more back and forth. It still takes a ton of iterations.
Hundreds of designs, then the design process
A while back I read Ideaflow, a book about the value of generating a lot of ideas. I think it still holds after AI.
The magic of AI isn’t that it can one-shot a design. It’s that it can one-shot hundreds of designs. Then you still use the design process to converge on the right one.
It starts with the client
My process always starts with the client: their goals, their dreams, their vision. This is not my art project.
I capture that in a simple design brief. The brief gives every round of exploration some context, and it records what the design needs to accomplish. I come back to it at the end.
Exploring
Then I explore. I need something visual to react to, and I hate designing in a chat window. So I work with real HTML pages I can open and compare.
I start from my own reference library at taste.plerom.io. I built a skill that takes websites I like, fingerprints their design style, and writes prompts for recreating that style. For each project, I choose the categories of direction that fit the client.
The tool I use most is Impeccable. It’s the most visual tool I’ve found for getting inside the design process. I can make changes across a whole page, or select one element and ask for several variations of just that part. It gets me out of the chat box.
I generate a lot of directions I can react to. I pick a few that look right and generate more iterations of those. Then I mix, match, and repeat until something feels close.
The illustration I couldn’t copy
I’ve been dogfooding this process on the Plerom site for several months. I landed on colors and typography early. I only recently found an illustration style I’m happy with, and I’m still playing with layouts.
For a while I tried to copy illustration styles I liked, ones I thought might work for my site. Getting the output to look like a very specific example was enormously difficult.
Then Impeccable used Higgsfield, an image-generation model, to produce a lot of diverse options. One of them used a completely different approach than anything I had tried. Visually, I could just tell it fit the look and feel I was after.
I found my own style almost by accident.
Knowing what the business needs
Exploring this way means I see a lot of designs that aren’t right for a client’s business or buyers.
I recently worked with an immigration lawyer. Many of the first directions came across as corporate and sterile, or bold. But immigration law is very different from typical law. The people looking for help are dealing with a very scary process.
So I built a warm, friendly direction instead, one that would make those visitors feel invited.
Type is where I can add something
Typography is a place where I can bring my own perspective. AI tends to pick free, open-source fonts, usually from Google Fonts. So I spend time with my favorite foundries, like Klim, OH no, and Displaay, looking for fonts with personality that match the business.
Knowing when to stop
When do you stop exploring and commit? That’s the million-dollar question. It’s often hard to know when to stop tweaking.
I go back to a book I read as a design manager, Discussing Design by Adam Connor and Aaron Irizarry. They define critique this way: “Critique is a form of analysis that uses critical thinking to determine whether a design is expected to achieve its desired objectives.”
The objective might be utility, like helping someone complete a task. It might be a metric, like more conversions on a particular call to action. Or it might be an experience, like making someone feel excited or surprised.
A lot of that should be in the brief. That’s why I write it first.
What it’s changed
Using this process on my own site first, and now with clients, has made me a much better web designer. It’s given me more confidence. Mostly, it’s made me even more disciplined about my design process.