🧭 Why It Matters

Good prompts don’t happen by accident. If you want outputs you can actually use, you need structure. That’s where R.I.S.E. comes in — four steps that make any prompt sharper.

🔑 The R.I.S.E. Framework

R — Role
Tell the AI who it is.

“You are a customer support agent trained to handle refunds politely.”

I — Instruction
State exactly what you want it to do.

“Summarize this email in 3 bullet points.”

S — Structure
Show how the answer should look (bullets, JSON, steps, etc.).

“Return the answer as: {‘summary’: ‘’, ‘next_step’: ‘’}”

E — Examples
Include one sample input/output if style or tone matters.

INPUT: ‘Hi, I need help canceling…’ → OUTPUT: {‘summary’: ‘Request to cancel’, ‘next_step’: ‘Explain refund policy’}

Before → After (Boundary-Setting in Action)

Vague (fails often):
“Summarize this customer email.”

Agentic (wins consistently):

<role>
You are a support triage summarizer for L2 engineers.
</role>

<rules>
- Extract only actionable info; no pleasantries.
- If version/build is missing, set status:"needs_more_info".
- Tone: concise, neutral. 3 bullet max.
</rules>

<schema>
{ "status":"success|needs_more_info",
  "summary":"string",
  "items":[{"label":"issue","value":"string"},
           {"label":"repro","value":"string"},
           {"label":"env","value":"string"}] }
</schema>

<examples>
INPUT: "Hi team, on v2.3.1 the export freezes after clicking 'CSV'…"
OUTPUT: {"status":"success","summary":"Export freezes on CSV click","items":[…]}
</examples>

<context>
<<< paste the raw email thread >>>
</context>

10-Point QA Checklist (Ship-Ready)

  • Task is first; context is separated with clear markers.

  • Role & audience named (not “be helpful”).

  • Positive, prescriptive rules (not just prohibitions).

  • JSON/XML schema provided and parsable.

  • Few-shot examples cover normal + edge cases.

  • Success criteria & test cases exist (tiny eval set).

  • “I don’t know / needs_more_info” path defined to reduce hallucinations.

  • Tone/verbosity constrained (mirror what you want).

  • Parameters aligned to task (e.g., low temperature for factual extraction).

  • Iteration plan: log failures → refine spec.

⚡ Try This Today

Write one prompt for a task you need today using R.I.S.E.

  • Add a Role

  • Write a clear Instruction

  • Define the Structure

  • Give an Example

Run it once. Then refine with feedback like:
“Make it shorter.” / “Add a friendlier tone.”

🛠 Pro Tip

Save your R.I.S.E. prompts in a Prompt Journal. Over time, you’ll build your own library of tested, reusable prompts.

Keep Reading

No posts found