APE
Action · Purpose · Expectation — TAG with the emphasis moved to the finish line: what to do, why, and how you will judge the result.
Use it for
Quick requests where the acceptance criteria matter more than the shape — analysis, recommendations, judgement calls.
Avoid it for
Work that depends on documents you must supply. Like RTF and TAG, APE has nowhere to put source material.
The slots
Each letter is a question the prompt has to answer.Action
What should the model do?
One verb, one deliverable.
Example: Review the pricing page copy below and identify what would make a first-time visitor hesitate.
Purpose
Why does this matter?
The business or personal reason. Tells the model which trade-offs to make.
Example: We convert 1.8% of pricing-page visitors and want to understand the drop-off before redesigning.
Expectation
What does a good answer look like?
Acceptance criteria, not format. "Every point tied to a specific sentence" is an expectation; "use bullets" is a format.
Example: Five findings maximum, each quoting the exact sentence it refers to, ordered by likely impact on conversion.
Strengths
- Expectation doubles as a self-check the model can grade itself against.
- Very short, and the three slots genuinely do not overlap.
Weaknesses
- No role, no context, no source.
- Least widely recognised of the short frameworks, so shared prompts need explaining.
Sources and grounding
- 1Prompt engineering — best practicesOpenAI (Developer docs)OpenAI recommends stating success criteria explicitly rather than implying them.