You don’t need a designer, a design degree, or even a drop of talent to get genuinely good design out of your AI. Most people who try to design with AI fail because they ask once, get a generic result, and assume that’s the ceiling. It isn’t. There’s a process behind the good stuff, and it takes about an hour to learn.
Why AI design looks so generic
Large language models are prediction machines. They look at your prompt and produce the most statistically likely output, which is another way of saying “the most average output.” Ask for a logo, a landing page, or a slide deck and you’ll get something competent, recognizable, and completely forgettable. It looks AI-generated, because it literally is.
Here’s the part people miss: when you design with AI, the dullness isn’t the model’s fault. It’s the prompt’s fault. If you paste a vague request into a prediction machine, you get the average of everything, which is blandness. The fix isn’t a secret prompt hack. It’s a process that forces the AI to reflect your taste instead of the average taste of the internet.
This process comes from a great breakdown by the folks at Lenny’s Newsletter, and it applies to ChatGPT, Claude, or any capable model. Three phases: discover, define, build.
Step 1: Discover with deliberately bad prompts
Your first instinct is to ask for a finished design. Kill that instinct. The first pass should be a brainstorm, not a deliverable.
Ask the AI to list a bunch of ideas and explicitly tell it to keep them rough and lacking detail. Twenty rough directions, one line each. The goal here is not to get good ideas from the AI. The goal is to spark your imagination. Speed is the point: you want a wall of raw possibilities so your brain can react: “that one’s interesting, that one’s boring, wait, combine those two.”
Interesting thing happens in this phase. The model’s junk ideas are actually useful, because your irritation with them tells you what you don’t want, and that’s half of taste right there.
Step 2: Define with your own taste
Now you steer. This is where most beginners chicken out and just accept whatever the AI suggested in step one. Don’t. Bring your own references.
Think of an inspiration you actually love: a video game’s UI, an interior design trend, an art installation you saw once, a magazine layout that stuck with you. Describe it to the AI and explain how you want that influence applied to your project. Be ambitious here. Specific weird instructions beat generic polish every time.
Then iterate. Show the AI what you like, what you don’t, and why. Once you’re close, ask it to write the final build prompt itself, with all your constraints baked in. That prompt becomes your blueprint, and you can reuse it for every iteration. That’s how pros design with AI: they steer instead of settling.
The rule of thumb: if you paste the AI’s own ideas back into the AI, you’ll get something anyone could have made. If you push your taste through it, you end up with something only you would have created. That’s the entire game.
Step 3: Build with a design critic
When you design with AI and add a critic pass, the quality jumps noticeably. Here’s the trick that separates pros from amateurs: never let the builder judge its own work. When your AI produces a design, have a second pass evaluate it before you accept it.
The critic needs clear, objective criteria. “Make it look beautiful” is useless, the result varies wildly from run to run. Instead try: “Check the hierarchy. Is the headline the most prominent element? Are there more than three fonts? Does the color contrast meet accessibility standards?” Concrete checks produce consistent improvements.
You can even run the critic as a separate agent or chat session with a stronger model, since bigger models generally have better design judgment. Start with one or two critique rounds, look at whether the feedback is actually converging on something better, and stop when it is. The goal is momentum, not perfection loops.
A real example: your landing page in one sitting
Let’s say you want a landing page for your side project, and you’ve never designed anything before. You can still design with AI from scratch.
Round one: you ask the AI for fifteen rough landing page concepts, one sentence each, with no detail required. One of them mentions a brutalist layout with a single giant headline, and you feel a jolt of “yes, that.”
Then round two: you tell the AI you want it to feel like the title screen of Inside, that minimalist game, mixed with a Product Hunt launch page. You iterate on the headline, the section order, the vibe. After three rounds you ask it to write the build prompt.
And round three: you generate the page, then open a fresh chat and ask the critic: “Is the value proposition visible in five seconds? Is there exactly one call to action? Does it work in black and white?” You fix the two real problems it finds, and you’re done.
That’s a complete design workflow, and it cost you an afternoon you would’ve spent scrolling anyway.
Common mistakes
The biggest one is vague critique. “Make it prettier” loops forever because prettier isn’t measurable. Fix the criteria, fix the loop.
Another one: using a weak model as your critic. The small free models are fine for generating drafts, but judgment benefits from scale. Use the same model you’d trust to summarize a complex document.
And stop expecting one shot to work. Design with AI is iterative by nature. A design that needed five rounds isn’t a failure, it’s a process working as intended. Pros who make this look easy aren’t prompting better, they’re iterating more.
The bottom line
You already pay for the tool. The missing ingredient is the process: brainstorm rough, define with taste, build with a critic. Run one real design task through it this week, even something small like a social post or a slide deck, and the difference will be obvious.
If you want to go deeper, our guide on Claude prompts for visualizations covers the research-to-graphics side, and when you’re ready to pick tools, Google Pics vs Canva and the Grok vs Photoshop comparison sort out which design tools are actually worth your time. Start with the process, then let the tools fight it out.
