These days, I design everything for my business with Claude: Hero images, infographics, landing pages, product mockups, PDFs, and full websites and apps.
I’m not an expert when it comes to professional design software like Adobe Illustrator and Figma, but I do have strong opinions about design. I typically know the aesthetic I want, and years of conversion optimization and product development work have shaped how I think about UX. Still, it’s never made sense for me to prioritize mastering design tools, so for years, my ideas stayed limited to whatever I could pull off in Canva or Google Slides (iykyk).
The reason Claude works so well for me is that I’ve treated it like a professional designer I’m working with. The results come out higher quality and much closer to what I picture than anything I made before…in a fraction of the time.
The way I work with Claude for design has grown into a repeatable system I use for every project, and this article documents every part of it. Note: This system is not an argument against professional designers, and a good designer could even build and run a system like this for you.
Why Most People Struggle to Design With AI
People struggle to design with Claude because the model starts every conversation knowing little to nothing about your brand, and that missing context leads to five common issues.
Graphics look generic. AI designs from the average of everything it has ever seen, so without a strong brand point of view, the results look like everyone else’s.
Graphics lack consistency. Even with your hex codes and logo pasted in, AI makes every other decision fresh each time: The layout, the type sizes, the spacing, the decoration. Each graphic can look fine on its own, but side by side they look like they came from different brands.
You keep having to repeat the same instructions. Every new chat needs the brand, the dimensions, and the specs explained before any real work starts. When all of that knowledge only lives inside your head, you are the one carrying it from session to session.
Corrections get forgotten. Feedback lives only in the chat where you gave it. Once that conversation ends, the AI goes back to making the same mistake.
There’s no quality control. A designer would check contrast, text size, and how the image will look on each platform before handing it over. Without a system in place, quality control falls to you, and you may not always be able to see the issues.
The five problems don't come from a lack of ability. AI can build almost anything you describe. Sharing your hex codes, fonts, and logo isn't enough, because colors and fonts are a small piece of what a designer knows about a brand. Your agent still guesses at everything else: Whether the brand should feel playful or serious, how you lay things out, what sizes work on each platform, and which ideas you would reject.
A human designer doesn't have any of these problems, because a human designer gets onboarded. On day one, you would hand them brand guidelines, examples of past work, and a sense of what good looks like for you, and once you corrected something, they would not do it again. AI typically only gets a two-sentence prompt.
My AI design system does that onboarding for me. When a new session starts, Claude has already read my brand materials and my saved feedback, and it checks its own work before showing me anything. Each of its five parts solves one of the five problems.
How My Claude Design System Is Organized
My Claude design system has five parts: A design folder, a brand guidelines document, a feedback loop, skills, and verification checks. Each part does one job. For example, my infographic skill contains the LinkedIn dimensions and my guidelines document contains my brand colors, so I can change my palette without touching the skill, and update an infographic treatment without any chance of messing up the brand everywhere else.
The components work together. When I correct something, the feedback gets saved as a rule that same day, and rules that generalize get folded into the guidelines document. When getting a new format right takes a lot of back-and-forth, everything we figured out gets packaged as a skill, so the next one takes a fraction of the time to produce at the same quality. The folder collects every finished asset along the way, so each new project starts with more reference material than the last.
The rest of this article walks through all five parts, with the exact specs and prompts from my working setup.
📁 The folder structure that turns your past work into Claude’s reference library.
📄 How to build brand guidelines Claude reads before every design task, and why HTML beats a brand PDF or a brand.json.
🔁 The feedback loop that turns your feedback into permanent rules, plus the sync protocol that keeps everything current.
✅ How to make quality control automatic, with Claude checking every design against your standards before you see it.
✍️ The CLAUDE.md rule that makes Claude save every correction automatically.
🛠️ Access to my Claude tools that build the system for you, including a design guidelines creator and a LinkedIn infographic creator.