How I Keep AI-Made Assets on Brand with a Claude Skill
I spent about a month building the perfect slide template in Google Slides. I mean perfect. Stakeholders had feedback on every slide, every layout, every pixel, and we went round after round until we landed on something that would work for all 130 people in the company. It was an intense process, and when it finally went live, I was proud of it.
One week later, Claude launched Claude Design.
From that moment on, I’m not sure anyone remembers we even have a Google Slides template. Every presentation is made in Claude now. They’re internal only (for now), but that’s still a lot of slides I never see coming. Yes, my job is to be the brand police. Of course it is. But policing only works when you can see what’s being made, and that stopped being true almost overnight. When anyone can produce a ten-slide deck in a few minutes, one person reviewing outputs can’t keep up. The numbers just don’t work. And I had better projects to work on than chasing salespeople about the wrong shade of blue.
So I stopped trying to fight it, and started building with it.
That’s when my job quietly changed. For most of my career, brand was something you designed and then defended, with guidelines, templates and reviews. With AI in everyone’s hands, defending stops working. The brand has to live inside the tools people already use, so that the on-brand version is simply what comes out by default. That’s what I mean by brand engineering. You stop checking the output and start designing the system that produces it.
So I started learning the program, beginning with the design presentations. It wasn’t smooth. There was a lot of trial and error, and a lot of “omg, why can’t you follow my instructions?” muttered at my screen. But that friction taught me the most important lesson of the whole project: a model doesn’t read brand guidelines the way a designer does. A designer gets the spirit of a rule and fills in the gaps. A model needs the gaps filled for it. Everything a brand team usually leaves to taste and judgment has to be spelled out, tested, and spelled out again, until the model can’t misunderstand it.
Templates helped, but they weren’t easy enough for people. To use a template, you have to remember it exists, find it and start from it. Most people just open Claude, type what they need and hit enter. I could have spent my energy trying to change that habit. Instead, I designed around it. The best brand system is the one nobody has to think about, so rather than asking 130 people to change how they work, I changed where the brand lives.
I built a brand skill and made it available across the whole organization. Now the brand isn’t a file someone has to go looking for. It’s already there in the background whenever someone asks Claude for a deck or a document.
The skill was only part of the work, though. I also build on-brand templates in the AI tools people reach for (Claude, Lovable, ChatGPT), and I coach people on how to use them, so the brand holds up no matter where someone starts. A system only works if people trust it, and they only trust it once they’ve seen it work for them.
Looking back, the biggest shift wasn’t the tooling. It was my role.
I spend far less time approving things and far more time designing how things get made. Instead of being the bottleneck at the end of the process, I’m built into the start of it. And that gives me my time back for the work that actually needs a designer.
And I’m proud of it. I still get a real kick out of seeing someone in sales or HR make a document that’s actually on brand. Are those documents beautiful? Not really. To me they still look a bit flat and soulless. But nobody’s inventing new colors and fonts anymore... and for brand police, that’s a very good day.