01 / Vague askProperty: Symptom
Why are users dropping off?
Answer
They may be hitting friction. Review analytics, simplify the flow, and test clearer copy.
Question Quality Is Leverage
AI exposes your judgment.
When the answer is generic, inspect the question first.
Prompting is not product thinking. Better wording will not save an unresolved decision.
Narrowing field
The question is a boundary around what can count as a good answer.
A vague ask leaves too much room. It gives the model permission to produce almost anything coherent, which makes almost anything feel like progress.
The useful move is not a longer prompt. It is naming the judgment you expected AI to own.
Question threadResolution space
Possible answer spaceshrinks as judgment enters
- Symptom
A visible problem, but no boundary yet.
- Surface
The answer must stay inside onboarding.
- Audience
The answer must focus on new users.
- Outcome
The answer must relate to first value.
- Risk
The answer must explain trust loss, not just friction.
- Evidence
The answer must say what would change the call.
- Decision
The answer must point to a promise to remove or defend.
- Consequence
The answer must include what breaks if the decision is wrong.
If evidence shows it causes trust loss before the first useful result, remove or defend that promise before optimizing the flow.
Risk named. Evidence named. Decision owned.Checkpoint
What judgment did you expect AI to own?
That question is the work. It turns the model from a decision-maker back into a tool for inspecting context, constraints, criteria, and consequences.
Proof without theater
The trap and the checkpoint are closer than they look.
Trap
Build speed can make avoidance look productive.
- AI makes fast build loops easy.
- A working artifact can still expose a missing product judgment.
- The question was vague enough to let anything count as progress.
Checkpoint
The useful pause happens before the build gets expensive.
- New context can make an old risk boundary stale.
- A sharper question can name, replace, defer, or block the next move.
- A prompt can format judgment. It cannot replace it.
Focus test
Keep the answer in focus.
Same focus move, different object: an AI-built feature already works. The risk is mistaking a clean artifact for a reason to build.
The focused question asks what must change after use, what evidence would prove it, and what decision the team will own if the feature does not create that change.