# Score an AI output against the prompt that produced it

A requirement-by-requirement scorecard for an AI output, separating prompt problems from output problems, with prompt fixes.

- Topic: Working with AI
- Framework: RTF
- Level: Advanced
- Tags: AI & Prompting, Rewrite & Edit, Analyze, Fill-in Template
- Source: https://www.thinkwithprompt.com/prompts/score-an-ai-output-against-the-prompt-that-produced-it

## Prompt

```text
Grade an AI output against the prompt that produced it, so I can compare prompts or models and improve my results.

**The prompt:** [PASTE the full prompt, including any system or custom instructions]
**The output:** [PASTE]
**What a great result would do for me:** [the real goal - e.g. "an email my client would approve without edits"]
**Comparing against:** [optional: a second output from another prompt or model]

Steps:
1. List every explicit requirement in the prompt (format, length, sections, tone, inputs to use, constraints), then the implied ones.
2. Check each requirement - met, partly met, or missed - and quote the part of the output that shows it.
3. Score the output from 1 to 5 on:
   - **Instruction following:** did it do what was asked, in the requested format?
   - **Accuracy:** are facts, numbers, and claims correct and grounded in the inputs? Flag anything that looks invented.
   - **Completeness:** does it cover everything, without padding?
   - **Usefulness:** could I use it as-is for my real goal?
   - **Clarity and style:** readable, right tone, right length.
4. Separate problems caused by the prompt (ambiguity, missing context, conflicting instructions) from problems in the output itself.

Return:
- A requirements table: Requirement | Status | Evidence.
- Scores with a one-line justification each, plus an overall score.
- If comparing, a side-by-side verdict on which better serves my goal.
- The top 3 fixes to the prompt, with the revised wording.

Be strict and specific. A 5 means I could use it without changes.
```

## What to fill in

- PASTE
- optional: a second output from another prompt or model

Free to copy and use. From ThinkWithPrompt (https://www.thinkwithprompt.com/), a library of 2,663 curated AI prompts.