Evaluate Your Own Designs
Run three forced-perspective critiques on any design: the confused user, the skeptical engineer, the impatient PM. Get specific problems, not encouragement.
The situation
You have finished a design and want honest feedback before sharing it. You know what happens when you ask Claude "what do you think?": positives first, two gentle suggestions buried at the end, nothing you can actually act on.
You'll run three forced-perspective critiques and walk away with a prioritised list of real problems ranked by user impact, not a list of encouragement.
What you walk away with
Three forced-perspective critiques: the confused user, the skeptical engineer, the impatient PM
A ranked list of real problems sorted by user impact, not design effort
A synthesis you can present to a team without making the design sound broken
Set up the critique
If you are in the app, paste a screenshot or describe the screen in detail. In the terminal or IDE, describe the interface: what is on it, what the user is trying to do, what the flow looks like.
# Describe the design for Claude to critique:
"Here is the design I want you to critique.
Screen: [Name of screen, e.g. 'onboarding step 2 - account setup']
User goal: [What the user is trying to accomplish on this screen]
What's on the screen: [List the UI elements, labels, layout]
Platform: [Mobile / web / tablet]
User: [Brief description of the target user]"Run the confused user perspective
This perspective takes the viewpoint of a real user who does not understand digital conventions. It surfaces copy problems, interaction confusion, and assumed knowledge that should not be assumed.
# Prompt for the confused user:
"You are a first-time smartphone user with low digital literacy.
You have never set up an account before. You don't know what a PIN is or why you need one.
Review this screen and tell me every point of confusion you would experience.
Stay in character. Don't suggest solutions. Just describe the confusion."Run the engineer and PM perspectives
The engineer catches what you will have to build that you have not designed for. The PM asks whether this screen earns its place in the flow.
# Prompt for the skeptical engineer:
"You are a frontend developer who has to build this screen.
Tell me every edge case, error state, and implementation problem you see.
What have I not designed for?"
# Prompt for the impatient PM:
"You are a PM focused on conversion and retention.
Is this screen earning its place in the flow?
What friction would you cut? What would you move to later?"Synthesise into a prioritised action list
Ask Claude to take everything across all three perspectives and produce a ranked list of fixes, ranked by user impact, not design effort.
Turn your three-perspective critique into a prioritised action list.
Group by: must fix before sharing, should fix before handoff,
consider for next iteration.
Include the heuristic or principle each item violates.Common mistakes
Four failure modes for this process.
- Asking for a general review instead of a forced perspective. "Review my design" produces hedged praise. "You are a confused first-time user: describe every point of confusion" produces usable findings. The perspective constraint is the mechanism.
- Letting Claude out of character. If Claude starts suggesting solutions mid-critique, bring it back: "Stay in character. Only describe the problem, not the fix." Solutions come in the synthesis step.
- Running all three perspectives in one prompt. If you ask for all three at once without separating them, the output blends perspectives and the engineer issues get mixed with the user issues. Run them separately, then synthesise.
- Skipping the synthesis step. Three separate critiques are hard to act on. The synthesis groups them by urgency and gives you something you can put in a sprint. Do not skip it even if the critiques feel complete on their own.
What's next?
Do a heuristic evaluationNew guides, when they ship
One email, roughly weekly. CLAUDE.md templates, workflows I actually use, and the cut-for-length stuff that does not make the public guides. One-click unsubscribe.
Or follow on Substack