Structured Summary Prompt — the Anatomy
The blocks a reliable summary prompt needs: source guidance, a fixed section skeleton, length budgets, fidelity rules, and quote handling.
Overview
Most summary prompts are the word "summarize" with hope attached — and the result skips points, pads sections, and invents context. The reliable anatomy has five blocks: SOURCE guidance that teaches how to read this kind of text, an OUTPUT STRUCTURE with named sections in a fixed order, LENGTH RULES with countable budgets, FIDELITY RULES that draw the invention line, and QUOTE RULES. This resource loads a support-case handoff — sectioned by the source's own logic (Issue, What Was Tried, Resolution State, Follow-Ups) — the cleanest demonstration of every block doing its job.
Workflow
-
Read each block's job
SOURCE teaches reading, STRUCTURE fixes the shape, LENGTH caps the budget, FIDELITY draws the invention line, QUOTES decide what stays verbatim.
-
Note the source-specific sections
Issue / What Was Tried / Resolution State / Follow-Ups exist because the source is a support conversation — switch the source type and watch the skeleton change.
-
Swap in your own scenario
Change the goal and source; the structure preview shows the new skeleton before you generate.
Why This Works
- Named sections in a fixed order make summaries comparable across runs
- Each block closes a specific failure mode — skipped points, padding, invention, paraphrased commitments
- Source-aware skeletons beat one-size-fits-all summary templates
Best for
- Anyone whose summary prompt is currently one word long
- Teams standardizing summaries across people and models
- Sources with their own logic — cases, incidents, threads
Not for
- Pulling discrete values into fields — that's the Extraction Prompt Generator
- Generating new documents — summarization compresses an existing source
Use cases
- Learning the structure before writing your own summary prompts
- Auditing an existing summary prompt against the five blocks
- Handing off support cases in a shape the next agent can trust