Your sketch does not need to look impressive. It needs to make your idea understandable. This playbook helps you use AI to explore a flow, then learn whether people can use it.
WHAT YOU’LL LEAVE WITH
An interactive prototype, a research script, and a clear next decision.
Before you begin
- A photo of a rough sketch with screens and arrows labeled
- A specific audience, task, and uncertainty
- Known constraints and real or clearly labeled sample content
STEP 01
Name the question
Write one question this prototype should help answer. For example: Can a new user identify their next onboarding step without help? Keep desirability, usability, and feasibility questions separate.
STEP 02
Annotate your sketch
Label screens, actions, decisions, and states. Add a note explaining what must stay and what is open to change. A sketch shows spatial intent; your brief explains why it matters.
STEP 03
Check the AI’s interpretation
Upload the sketch to your approved visual AI tool. Ask it to describe the flow and identify ambiguity before creating the interface. Correct mistakes while they are still cheap.
STEP 04
Compare two approaches
Ask for two meaningfully different ways to complete the task. Discuss the tradeoffs with design and engineering where appropriate. Choose a direction based on the learning goal.
STEP 05
Build the smallest useful version
Use Claude Design or Figma Make for an interactive concept, or a coding assistant for a code prototype. Include the important empty, loading, error, and success states. Label sample data.
STEP 06
Observe people using it
Recruit relevant users for a small formative study. Give a task, avoid coaching, and record where help is needed. Check mobile use, keyboard navigation, labels, and recovery paths.
STEP 07
Make the next decision
Summarize what you observed, what remains unknown, and whether to continue, revise, or stop. Usability evidence does not establish willingness to pay. A prototype is not a production release.
MAKE THIS PROMPT YOUR OWN
Read the attached sketch and brief. Describe the flow you infer and list unclear details. Separate supplied facts from assumptions. Propose two ways to help [audience] complete [task], preserving [constraints]. After I choose, build a narrow interactive prototype with sample data and relevant states. Explain what it can test and what it cannot establish.
Check your work
- Does the prototype test the stated question?
- Can you trace requirements back to the brief?
- Have you checked failure and recovery states?
- Does the decision reflect observed behavior?
Further reading: official guidance / research ↗
Editorial guide · Reviewed September 27, 2026. Tool interfaces may change.