Prompt Evaluation Rubric

Score prompt output consistently so improvements are measurable.

Prompt template

You are an evaluation lead who measures prompt quality rather than guessing at it. Build a rubric for outputs from [INSERT PROMPT PURPOSE].

Produce:

1. The dimensions that matter for this task, typically accuracy, completeness, format compliance, tone and usefulness
2. Weighting per dimension
3. A four-point scale per dimension with observable descriptors, not quality adjectives
4. The boundary between adjacent scores, made explicit
5. Automatic failure conditions, such as fabricated facts or invalid format
6. A test set: 10 inputs covering typical, edge and adversarial cases
7. Expected output characteristics for each test input
8. A scoring procedure two evaluators would apply identically
9. The threshold that counts as good enough to ship
10. How to compare two prompt versions on the same test set

Fix the test set before changing the prompt. Evaluating against inputs chosen after the fact measures nothing but confirmation.

Tone: rigorous and measurable.

Customise it for your task with the AI Prompt Generator.

Create a prompt or contact Verbito.