# Traceable Executive Summary Prompt

> A reusable prompt that turns a long draft into a short executive summary where every claim cites the source paragraph it came from.

## Install

Copy the content below into your project:

# Traceable Executive Summary Prompt

A reusable prompt that turns a long draft into a short executive summary where every claim cites the source paragraph it came from.

## Start here

Paste the full long draft into an ordinary AI chat that accepts text. Number the paragraphs first (P1, P2, P3…) or let the model number them as they appear. Then paste the complete prompt appended with this asset, replacing the input lines with your own:

- **SOURCE DRAFT** — the long text to summarize.
- **SUMMARY AUDIENCE** — who will read it (manager, client, committee).
- **SUMMARY LENGTH** — a target word or bullet count.
- **READER GOAL** — what the reader must decide or understand.
- **SENSITIVITY NOTES** (optional) — names, figures or claims to flag rather than include.

## What you get back

The prompt returns four fixed sections: EXECUTIVE SUMMARY with [Pn] citations, a CLAIM TRACE table, OPEN GAPS, FLAGGED ITEMS, plus a short COVERAGE NOTE.

## How to check the output

1. Every [Pn] citation points to a real, correctly numbered paragraph.
2. Every quoted phrase in CLAIM TRACE appears word-for-word in your draft.
3. No fact, figure or recommendation appears in the summary that is missing from CLAIM TRACE.
4. The summary fits your target length.
5. Anything the source treats as uncertain stays uncertain in the summary.

## Prerequisites, permissions and limits

No terminal, account or API setup is needed — any chat that accepts a pasted text works. This prompt only produces text. It does not browse, open files, contact anyone, schedule anything, or send or publish the summary; those actions remain yours. Do not paste confidential material into a tool you do not trust. If the source is too long for one message, paste it in numbered chunks and ask the model to merge the per-chunk traces and mark paragraphs counted twice. source reviewed; runtime not tested.

## FAQ

**What if I leave out the audience or reader goal?** The prompt tells the model to stop and ask you for any missing required input before producing anything.

**What if two paragraphs contradict each other?** Both are placed in OPEN GAPS and the conflict is flagged, instead of the model quietly choosing one.

## Source and thanks

Original TokRepo prompt (CC BY 4.0). Reference context: [ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>). The worked example inside the prompt is labelled FICTIONAL MINI INPUT and is illustrative only, not real data.

## Complete reusable prompt

You are an editorial assistant helping me compress a long draft into a short executive summary whose claims I can trace back to the source. Work only from the material I paste below. Do not add facts, numbers, names, dates, recommendations or conclusions that are not already present in the source. This task is summarizing an existing draft, not researching, sending, publishing or making decisions — you only produce a draft summary for me to review.

INPUT I WILL PROVIDE
1. SOURCE DRAFT — the full long text to be summarized, broken into numbered paragraphs (I will number them, or you may number them P1, P2, P3… as they appear).
2. SUMMARY AUDIENCE — who will read the summary (for example a manager, a client, a committee).
3. SUMMARY LENGTH — a target word or bullet count.
4. READER GOAL — what the reader needs to decide or understand after reading.
5. SENSITIVITY NOTES (optional) — any names, figures or claims I want flagged rather than included.

YOUR TASK
Step 1 — Read the whole source draft and build a private claim map: list each distinct factual claim, figure, date, named party, or recommendation, and the paragraph number(s) where it appears. Do not force a fixed count; include what exists.
Step 2 — Select the claims that matter to the stated READER GOAL and drop the rest. If the source is silent on something the reader would obviously expect, do not invent it — list it under OPEN GAPS instead.
Step 3 — Write the executive summary within the target length, using clear plain prose. Every sentence that states a fact must carry a bracketed citation of the source paragraph(s) it came from, like [P4].
Step 4 — Add a short coverage note: which source paragraphs contributed and which were left out.

OUTPUT FORMAT (exactly these headings)
EXECUTIVE SUMMARY
<the summary, with [Pn] citations after each sourced claim>

CLAIM TRACE
A table with columns: Claim in summary | Source paragraph(s) | Exact quoted phrase from source (short).

OPEN GAPS
Bullet list of reader-relevant items the source does not address. Write "none identified" if empty.

FLAGGED ITEMS
Names, figures or claims matching my SENSITIVITY NOTES, quoted with their paragraph numbers, or "none".

COVERAGE NOTE
One or two sentences naming included and excluded paragraphs.

UNCERTAINTY AND MISSING INPUT
If SOURCE DRAFT, SUMMARY AUDIENCE, SUMMARY LENGTH or READER GOAL is missing, stop and ask me for it before producing anything. If the source is ambiguous on a point, mark it [UNCLEAR] rather than guessing. If two source paragraphs conflict, place both in OPEN GAPS and flag the conflict instead of choosing. If the draft is too short to compress meaningfully, say so and summarize it as-is without padding.

CHECKS BEFORE I USE IT
1. Every [Pn] citation points to a real, correctly numbered paragraph.
2. Every quoted phrase in CLAIM TRACE appears verbatim in the source.
3. No claim, figure or recommendation appears in the summary that is absent from CLAIM TRACE.
4. The summary fits the target length.
5. Drop any claim the source treats as uncertain, unless the summary carries the same uncertainty.

BOUNDARIES
Do not browse, open files, contact anyone, schedule anything, or send or publish the summary. You produce text only. Distinguish clearly between describing what the source says and recommending an action. Do not smooth over gaps with plausible-sounding filler. If a provider feature for summarizing or splitting long documents is unavailable to me, I can paste the source in numbered chunks and run this same prompt per chunk, then ask you to merge the per-chunk traces and mark any paragraph counted twice.

WORKED EXAMPLE (illustrative only, not real data)
Label: FICTIONAL MINI INPUT
SOURCE DRAFT (excerpt): P1 We tested the new intake form on thirty orders last week. P2 Each order took eleven minutes longer than before. P3 Two customers emailed to say the form was confusing. P4 We have not yet decided whether to keep it. P5 Costs rose by £40 total.
SUMMARY AUDIENCE: operations lead.
SUMMARY LENGTH: 3 bullets.
READER GOAL: decide whether to keep the form.
SENSITIVITY NOTES: flag any customer counts.

ILLUSTRATIVE OUTPUT SHAPE
EXECUTIVE SUMMARY
- The new intake form added roughly eleven minutes per order across thirty orders tested [P1][P2].
- Two customers reported confusion with the form [P3].
- No decision has been made on keeping the form [P4].
CLAIM TRACE table rows: "eleven minutes per order" -> P2 -> "eleven minutes longer"; "two customers confused" -> P3 -> "form was confusing"; "no decision" -> P4 -> "not yet decided".
OPEN GAPS: whether the extra time or cost is acceptable; P5 cost figure not used because it was outside the reader goal and target length.
FLAGGED ITEMS: "Two customers" (P3) matches the sensitivity note about customer counts.
COVERAGE NOTE: Included P1–P4; excluded P5 cost figures.

PASS/FAIL CHECKS ON THAT EXAMPLE
PASS if each bullet carries a citation and the flagged customer count appears in FLAGGED ITEMS. FAIL if the summary states the form saved money, invents a decision, or cites a paragraph that does not contain the quoted phrase.

## References and reuse

- [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) · Reviewed 2026-10-04

Original TokRepo prompt · [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). Reference documents retain their own rights.

---

<!-- ZH -->

# 可追溯的高管摘要提示词

一套可复用的提示词，把冗长草稿压缩成简短的高管摘要，并让每条事实都标注它所来源的段落编号。

## 开始使用

把完整的长草稿粘贴到一个可以接受文本的普通 AI 对话中。先给段落编号（P1、P2、P3……），或让模型按出现顺序自行编号。然后粘贴本资产附带的完整提示词，把其中的输入项替换为你自己的内容：

- **SOURCE DRAFT**——需要压缩的长文本。
- **SUMMARY AUDIENCE**——谁会读这份摘要（经理、客户、委员会）。
- **SUMMARY LENGTH**——目标字数或要点数量。
- **READER GOAL**——读者读完后需要决定或理解什么。
- **SENSITIVITY NOTES**（可选）——需要被标记而不是直接写进摘要的人名、数字或说法。

## 你会得到什么

提示词会返回固定结构的几部分：带 [Pn] 引用的 EXECUTIVE SUMMARY、CLAIM TRACE 表格、OPEN GAPS、FLAGGED ITEMS，以及简短的 COVERAGE NOTE。

## 如何检查输出

1. 每个 [Pn] 引用都指向真实且编号正确的段落。
2. CLAIM TRACE 中每一句引用的原文，都在你的草稿里逐字出现。
3. 摘要中的任何事实、数字或建议，都不会在 CLAIM TRACE 中缺席。
4. 摘要符合你的目标长度。
5. 原文中不确定的内容，在摘要里同样保持不确定。

## 前提、权限与限制

无需终端、账号或 API 设置——任何能粘贴文本的对话都可以用。该提示词只产出文本，不会浏览网页、打开文件、联系他人、安排日程，也不会发送或发布摘要，这些动作仍由你自己完成。不要把机密材料粘贴到你并不信任的工具里。如果原文太长、一条消息放不下，可分编号区块粘贴，再让模型合并各区块的追溯表，并标出被重复计算的段落。源代码已核对；未做运行时测试。

## 常见问题

**如果漏填受众或读者目标怎么办？** 提示词要求模型在产出任何内容之前停下来，向你询问缺失的必填输入。

**如果两个段落互相矛盾怎么办？** 两者都会被放入 OPEN GAPS 并标记冲突，而不是让模型悄悄选一个。

## 来源与致谢

原创 TokRepo 提示词（CC BY 4.0）。参考背景：[ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>)。提示词内部的示例已标注为 FICTIONAL MINI INPUT，仅作演示，并非真实数据。

## 完整可复制提示词

你是一名编辑助理，帮助我把一份长草稿压缩成一份简短的执行摘要，并且其中的每一项说法我都能追溯回原文。请只依据我粘贴在下面的材料来工作。不要添加原文中不存在的事实、数字、名称、日期、建议或结论。这项任务是总结一份已有的草稿，而不是做研究、发送、发布或做决定——你只需生成一份供我审阅的摘要草稿。

我将提供的输入
1. SOURCE DRAFT——需要被总结的完整长文本，已拆分为带编号的段落（我会自行编号，或者你可以按它们出现的顺序编号为 P1、P2、P3……）。
2. SUMMARY AUDIENCE——谁会阅读这份摘要（例如经理、客户、委员会）。
3. SUMMARY LENGTH——目标字数或要点数量。
4. READER GOAL——读者读完后需要决定或理解什么。
5. SENSITIVITY NOTES（可选）——我希望被标记而不是直接写进摘要的任何名称、数字或说法。

你的任务
第 1 步——通读整份源草稿，建立一份私有的说法映射：列出每一项独立的事实性说法、数字、日期、具名方或建议，以及它出现的段号。不要强行规定固定数量；有什么就收录什么。
第 2 步——选出对所述 READER GOAL 重要的说法，其余丢弃。如果原文对读者显然会期待的内容保持沉默，不要凭空编造——把它列入 OPEN GAPS。
第 3 步——在目标长度内撰写执行摘要，使用清晰平实的行文。每一句陈述事实的句子都必须带上其来源段落的方括号引用，如 [P4]。
第 4 步——添加一段简短的覆盖说明：哪些源段落有贡献，哪些被略去。

输出格式（严格使用这些标题）
EXECUTIVE SUMMARY
<摘要正文，每项有来源的说法后附 [Pn] 引用>

CLAIM TRACE
一张表格，列包括：摘要中的说法 | 源段落 | 原文中的逐字引用（简短）。

OPEN GAPS
以要点列表列出与读者相关、但原文未涉及的事项。若没有则写 "none identified"。

FLAGGED ITEMS
与我的 SENSITIVITY NOTES 相匹配的名称、数字或说法，连同其段号一并引用，或写 "none"。

COVERAGE NOTE
一到两句话，说明哪些段落被纳入、哪些被排除。

不确定性与缺失输入
如果 SOURCE DRAFT、SUMMARY AUDIENCE、SUMMARY LENGTH 或 READER GOAL 有缺失，请停下来，在产出任何内容之前向我询问。如果原文在某一点上含糊不清，请标记为 [UNCLEAR]，而不要猜测。如果两个源段落互相矛盾，请将两者都放入 OPEN GAPS 并标记冲突，而不是替我做选择。如果草稿短到无法有意义地压缩，请如实说明，并按原样总结，不要注水填充。

我使用前的检查
1. 每个 [Pn] 引用都指向真实且编号正确的段落。
2. CLAIM TRACE 中的每一句引用都在原文中逐字出现。
3. 摘要中没有任何说法、数字或建议在 CLAIM TRACE 中缺席。
4. 摘要符合目标长度。
5. 丢弃原文视作不确定的任何说法，除非摘要带有同样的不确定性。

边界
不要浏览网页、打开文件、联系他人、安排日程，或发送、发布摘要。你只产出文本。请清楚区分描述原文说了什么与建议采取行动。不要用听起来合理的填充内容来抹平空缺。如果我无法使用某家服务商提供的长文档总结或拆分功能，我可以把原文按编号区块粘贴，对每个区块运行同样的提示词，然后请你合并各区块的追溯表，并标出被重复计算的任何段落。

WORKED EXAMPLE（仅作演示，并非真实数据）
Label: FICTIONAL MINI INPUT
SOURCE DRAFT（摘录）：P1 我们上周在三十个订单上测试了新的录入表单。P2 每个订单比之前多花了十一分钟。P3 两位客户发来邮件说这个表单令人困惑。P4 我们还没有决定是否保留它。P5 总成本增加了 40 英镑。
SUMMARY AUDIENCE：运营负责人。
SUMMARY LENGTH：3 个要点。
READER GOAL：决定是否保留该表单。
SENSITIVITY NOTES：标记任何客户数量。

ILLUSTRATIVE OUTPUT SHAPE
EXECUTIVE SUMMARY
- 新的录入表单在测试的三十个订单上，每个订单多花了大约十一分钟 [P1][P2]。
- 两位客户报告说表单令人困惑 [P3]。
- 尚未就是否保留该表单做出决定 [P4]。
CLAIM TRACE 表格各行："eleven minutes per order" -> P2 -> "eleven minutes longer"；"two customers confused" -> P3 -> "form was confusing"；"no decision" -> P4 -> "not yet decided"。
OPEN GAPS：额外时间或成本是否可接受；P5 的成本数字未被采用，因为它超出了读者目标和目标长度。
FLAGGED ITEMS："Two customers"（P3）符合关于客户数量的敏感性说明。
COVERAGE NOTE：纳入了 P1–P4；排除了 P5 成本数字。

对该示例的 PASS/FAIL CHECKS
如果每个要点都带有引用，且被标记的客户数量出现在 FLAGGED ITEMS 中，则 PASS。如果摘要声称该表单节省了成本、凭空杜撰了一个决定，或引用了一个并不包含所引短语的段落，则 FAIL。

## 参考资料与复用

- [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) · Reviewed 2026-10-04

TokRepo 原创提示词 · [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)。参考资料保留各自原有权利。

---
Source: https://tokrepo.com/en/workflows/traceable-executive-summary-prompt-f14b6ebe
Author: Prompt Lab