One of the (few) things I miss about working at an agency or client side is the informal conversation around a design project. A passing comment or spontaneous question can expose something you missed or send the work in another direction.
Working primarily solo during iteration and refinement, I miss that extra pair of eyeballs. So I tried using AI for design critique. The results were different than I expected.
Let’s get the disclaimers out of the way.
-
AI is not a replacement for human art or creative direction.
-
Without explicit instruction, it tends toward sycophancy. That makes it useless for refinement.
-
It can identify problems but can’t decide which problems should be addressed.
-
It can suggest changes but can’t make the creative decision.
So what is it good for?
With the right instructions, AI can analyze design work against explicit criteria and articulate what it sees with precision.
Anyone who has spent enough time designing has gone blind to their own work. You make twenty versions of a logo. Then forty. Eventually you’re moving something three pixels because you can no longer tell whether the underlying idea works. AI can offer enough critique to get the process moving again.
AI can identify hierarchy problems, competing visual elements, legibility issues, unnecessary complexity and places where the execution has drifted from the original intent. You may disagree with the critique. Sometimes you should, just like with a human AD. But now you have something specific to react to. And that beats “it needs more pop.”
It also has one practical advantage that shouldn't be overlooked: availability. Late night. Weekend. Version 37. It doesn't care.
Let’s take a look at a couple of examples.
This is a concept logo I worked on for King of the Hammers, an off-road racing event in Johnson Valley, California.
Partial feedback from ChatGPT:
Keep the aggressive blackletter/custom-metal DNA, but reduce the number of “special” moments. Pick the K slash, the giant H and the terminal S as the major character events. Let more of the supporting letters behave.
The S is also over-dominant. It has a nice flame/speed shape, but visually it becomes a second symbol at the far right. In the simplified wordmark, your eye lands on H, then S, then backs into the rest. That’s a decent poster read, but weaker as an identity read.
Make OF THE clearly subordinate but cleaner. It can be smaller, but it needs more breathing room and less interference from the G/S shapes.
My ego was a little bruised after developing the custom type.
But the critique is specific. More importantly, it creates decisions. If the concept were moving forward, I'd have to decide whether the character in those letters is helping the identity or competing with it.
That's useful criticism.
You may disagree with the critique. Sometimes you should. But now you have something specific to react to.
See the full feedback thread →
The next one is a stalled work in progress for an armor plating company called Gorgon.
Partial feedback from ChatGPT:
The face is the main problem. The left eye has too much character and softness, which pulls it toward illustrated heroine or game faction. The right side collapses into black mass, so the split-face idea doesn’t resolve cleanly. At small size, that interior will turn into a blob with a couple of white cuts. The gray arcs and lower tabs add noise without clarifying armor, plating or impact absorption.
The outer ring is controlled, but it reads more like orbital hardware or generic cyber-security than layered armor. For an ablative plating company, the mark should probably feel more material, modular, sacrificial, engineered. Right now it feels like a name-driven emblem wearing defense-category styling.
The concept has a long way to go. But the critique gave me some ideas about how to handle the shield shape.
When I told ChatGPT the concept was based on the mythological Aegis shield, I got more directed critique:
That helps, and it actually makes the problem clearer: the Aegis idea is conceptually right for this company, but the mark isn’t reading as shield yet…
The armor/plating brief is still the missing bridge. Aegis gives you mythic protection. Ablative plating gives you layered sacrifice, impact, erosion, material failure by design. The current ring system gestures at layers, but in a decorative/graphic way, not in a “this surface is built to take the hit” way. That’s why it feels unresolved rather than just unfinished.
I was surprised when I asked for tactical next steps:
Strip the outer structure down. The three white nodes and concentric orbital rings are pulling it into aerospace/tech-logo territory. If those become shield hardware, they need to feel like fasteners, plates, impact rings or layered armor, not decorative satellites.
Use the ablative plating idea structurally. Layered plates, offset seams, worn edges, fractured outer bands, sacrificial surface logic. Not texture for texture’s sake, but construction that says “this thing is built to absorb punishment.”
That's where the thread became more interesting.
The first critique identified visual problems. Adding the Aegis concept gave the AI enough context to connect those problems to my intent. Asking for tactical changes pushed it another step.
Not genius-level creative direction. Useful enough to act on.
Lastly, I asked it to produce a sketch of what it was driving at. Let’s just say ChatGPT won’t be winning any design awards any time soon.
You can see what I mean in the full feedback thread →
Want to give it a try?
I turned the instructions I used for these critiques into a downloadable set of Creative Review Guidelines. You can add the PDF to an AI conversation or use it as a governing source document in a project.
It's designed to keep the AI analytical, specific and focused on the work rather than telling you how wonderful you are. Download it, add it to your AI chat or project, and modify it as needed.
Get the Creative Review Guidelines →
If you improve it, I'd like to hear what you changed. I'm going to keep working on it.