The core of text summarization / explanation tasks: define the audience + define the purpose + define the output structure. The same source material will have completely different summary depths depending on whether it's for executives, product managers, or engineers. And "summarize" means different things when it's for decision-making vs. reporting vs. retrospectives vs. learning.

This page provides a set of reusable summarization templates to help you quickly compress "long text" into "actionable short info."

When You Need LLM Summarization

• Information overload: Meeting notes, long articles, research reports, customer interviews
• Quick decisions: Get key conclusions, risks, and next steps in 1-3 minutes
• External communication: Translate complex content for non-technical readers
• Structured output: Need JSON, tables, or bullet lists

Essential Elements (include these every time)

1. Role & audience: Who's reading this? (Executives / PMs / engineers / customers / students)
2. Purpose: What's it for? (Decision / report / retrospective / learning)
3. Length: How short? (80 words / 5 bullets / 1 table)
4. Format: How to output? (Bullets / table / JSON)
5. Notes: Preserve data / dates / citations / terminology?

Quick Templates (Copy & Paste)

Template A: Bullet Summary

Summarize the following into 5 bullet points for a product manager.
Requirements: each bullet max 20 words; preserve key data and dates.
Input:
"""
{{TEXT}}
"""

Template B: Decision Summary

You are the head of operations. Output a decision summary.
Format:
- Conclusion:
- Evidence:
- Risks:
- Next steps:
Input:
"""
{{TEXT}}
"""

Template C: Structured JSON

Output JSON with fixed fields:
{
  "summary": "string",
  "key_points": ["string"],
  "actions": ["string"],
  "risks": ["string"]
}
Input:
"""
{{TEXT}}
"""

Key Parameters That Affect Quality

• Compression ratio: Specifying "word count / bullet count / paragraph count" significantly stabilizes output
• Detail preservation: If numbers, dates, or names must be kept, say so explicitly
• Tone and style: Concise, formal, external-facing, internal -- spell it out in the instructions
• Structured format: Fixed fields or fixed ordering is best for reuse

Common Variations

1) Layered summary (progressive compression)

First summarize in 3 sentences, then output 5 bullet points.

2) Comparison summary

Output three sections: "Consensus / Disagreements / Unresolved issues."

3) Audience-specific versions

Output two versions:
1) For executives: 1 paragraph + 3 bullet points
2) For engineers: 5 bullet points + risk checklist

Quality Checklist

• Does it meet length/format requirements?
• Does it cover core conclusions and key data?
• Are irrelevant details removed?
• Can it be used directly for the next action?

Quick Tips

• Wrap input with delimiters (like """) to prevent model confusion
• Structure first, content second -- the more explicit the structure, the more stable the output
• When you need actionable output, prefer JSON or tables

Index

• /learn/prompt-master/prompt-text-summarization-explain-concept