Get Two or Three Genuinely Different Prototypes to Test

Introduced in “AI Helps Most When It Removes Handoffs Between Users and Builders.” Use it once a problem has been observed and written up, to get two or three genuinely different working prototypes a real user can try, instead of three versions of the first idea. Give the AI the observation or the Product Evidence Brief, the user role, and a description of the realistic or synthetic data it may use.

You help a product team build alternative prototypes for one observed problem.
Below this prompt are:
- the observation or Product Evidence Brief that describes the problem,
- the user role that has the problem,
- the data that the prototypes may use.

Rules:
- Use only the problem, the role, and the data below.
- Do not invent requirements, users, quotes, or numbers.
- Do not add a feature that the observation does not call for.
- Use only the synthetic or de-identified data below. Do not use real personal data.
- Do not choose a winner. Real users choose.

Steps:
1. State the problem in the words of the observation. Quote them exactly.
2. State the decision or task that the user must complete.
3. Write two or three prototype approaches.
   Each approach must make a different assumption about how the user does the work.
   A different layout or color of the same idea is not a different approach.
4. For each approach, state:
   - the assumption about the work that it makes,
   - the hypothesis that it tests,
   - what the user sees first,
   - what the user does to complete the task,
   - which data it uses,
   - what it leaves out,
   - what result in a user test would show that the approach is wrong.
5. Build each approach as a working prototype against the data below.
6. List each assumption that you made. Mark each assumption that a user test must check.

Output:
The approaches with the fields from step 4, the working prototypes,
and the list of assumptions.

Check the output before anyone sees it. If two approaches differ only in layout, ask for a replacement. Confirm that nothing in the prototypes uses data you didn’t provide, especially anything that looks like a real person’s record. Read the assumptions list: it’s the agenda for the user test. And remember that prototypes built by the team are still the team’s opinion until a real user at D0–D1 has used them.