Setting Output Expectations

Many users say "the output is bad" without defining what "good" looks like. In this lesson, we learn how to give the AI a Scoring Rubric.

1. The Rubric Technique

Instead of asking for a "good" article, tell the AI exactly what criteria to use.

Example Prompt:

*"Write a product description for [Product]. Your output will be evaluated on the following 0-5 scale:

• Clarity: Is the benefit immediately obvious?
• Tone: Is it premium and sophisticated?
• Action: Is there a clear Call to Action? Aim for a 5/5 score in every category."*

2. Defining "Style" by Example

Provide the AI with Anti-Examples.

• "Write this memo, but do NOT use buzzwords like 'synergy', 'leverage', or 'paradigm shift'. If you use passive voice, the output is unacceptable."

graph LR
    Goal[Vague Goal] --> Generic[Generic Output]
    Criteria[Specific Rubric] --> Target[Targeted Output]

3. Self-Grading

Ask the AI to grade itself before showing you the result.

• "Write the draft. Then, below the draft, provide a self-assessment of how well you followed my instructions on a scale of 1-10."

4. Expected Formats (The "Shape")

• "Provide the answer in a bulleted list of exactly 4 items. Each item must start with a verb."

---

Hands-on: The Rubric Challenge

1. Task: "Write a 1-sentence sales pitch for a pencil."
2. Criteria: It must be funny, under 10 words, and mention the eraser.
3. Your Role: Act as the judge. Did it meet all three criteria? If not, ask it to try again.

Key Takeaways

• Rubrics provide a map for the AI.
• Use Anti-examples to steer away from bad habits.
• Criteria-based prompts are much more reliable than emotional prompts.