Add Success Criteria to a Prompt
If you can't say what makes the output right, the model can't either. Success criteria turn 'make it good' into a checkable definition of done.
Overview
The quiet defect in most prompts: no definition of done. 'Good', 'high quality', and 'useful' feel like criteria but check nothing. A success criterion is a sentence you could evaluate the output against — 'a reader can act on it without follow-up questions', 'every claim has a number or a source'. This resource loads a prompt whose only quality bar is the word 'good' and rewrites it so the bar becomes explicit, checkable, and yours.
Workflow
-
Rewrite the loaded prompt
'Good' becomes a success-criteria slot; 'covers what they need to know' becomes a coverage list you name.
-
Write the criterion as a test
Good test: 'a new hire can complete day one without asking their manager anything in the guide'. Bad test: 'comprehensive'.
-
Reuse the pattern
Any prompt with an approval step deserves a criteria line — it's the line reviewers actually use.
Why This Works
- A checkable criterion changes both sides: the model optimizes toward it and you evaluate against it
- Criteria written as tests catch the gap between plausible and correct
- The rewrite leaves the criterion as a placeholder because 'done' is a decision, not a default
Best for
- Prompts whose output gets reviewed or approved by someone
- Recurring prompts where 'is this right?' keeps coming up
- Deliverables with a real reader who can be failed
Not for
- Scoring two prompts against each other — that's the Prompt Comparator
- Prompts for pure exploration where 'done' genuinely doesn't apply
Use cases
- Making AI output evaluable instead of just plausible
- Aligning a team on what a shared prompt is supposed to produce
- Converting quality adjectives into acceptance checks