Note: contexts are purposely shortened, but - as usual - the more detailed the context, the more accurate the LLM will reply.

Generate Visual Concepts

#1 — Ask for assumptions

When key info is missing, don’t let the model fill the gaps silently. Ask it to list assumptions + unknowns + the questions that would change the solution. Fewer invented details, cleaner reasoning.

Good prompt example:

We have high drop-off in onboarding.
Propose a UX solution to improve completion.
Context: B2C mobile app, 4-step onboarding.

Before proposing solutions:
- List your assumptions if you’re missing info.
- If an assumption is critical, write “insufficient data” + the exact question to ask me.

Then propose a V1 (flow + screens + states) using only this context.
Don’t pretend you know our numbers. Don’t invent anything.

Problem Framing

#2 — Split complex tasks into 4 simpler steps

Get better outputs by forcing a sequence: 1) help reframe the problem, 2) propose options, 3) compare trade-offs, 4) deliver a final recommendation. Less generic advice, more actionable materials.

Bad prompt example: