# Abstract Triage Log: Honest Literature Reading Notes

> Paste paper abstracts into an ordinary AI chat and get a claim-labeled triage table, shortlist, and data-gap list built only from what you supplied.

## Install

Copy the content below into your project:

# Abstract Triage Log: Honest Literature Reading Notes

Paste paper abstracts into an ordinary AI chat and get a claim-labeled triage table, shortlist, and data-gap list built only from what you supplied.

## Start here

Copy the complete prompt (it is appended below this guide) and paste it into an ordinary AI chat that accepts text. Immediately after the prompt, paste your own list on the next line: a comma- or semicolon-separated set of entries, each ideally containing author-year or citation key, title, venue, year, abstract text, and optionally a URL/DOI plus your one-line reason for interest. If you have a research question, state it; if not, the prompt uses `[WORKING QUESTION]` and flags it as unset.

You need nothing else: no terminal, no account setup, no API key. Paste, send, read the reply.

### How to check the output

- Every table cell should trace to words you pasted, or say UNKNOWN. Scan for any number, sample size or venue you did not supply.
- Findings must read as claimed, never verified.
- The shortlist and next actions must match the Relevance and Evidence status columns.
- If your working question was unset, the reply must say so.

If a field is missing from your entry, leave it out rather than guessing; the prompt marks gaps as UNKNOWN. Longer abstracts improve the Stated method and Claimed finding columns, but the log still describes only the abstract.

## Introduction

Cataloging abstracts you have collected is different from reading papers. This prompt builds an honest triage log: it records what each abstract states, what the abstract does not tell you, and which entries deserve a full read first. It separates what an abstract asserts from what a reader might mistake for established fact. Generic AI summaries often blur that line; a reader can use this structure to keep that boundary clear.

## Permissions and limits

- Suitable input: your own abstract list, your own notes, or public metadata you copy in.
- Avoid pasting confidential or unpublished manuscripts, personal data, or content you lack the right to share.
- The chat only organizes what you paste. It does not search databases, run queries, retrieve papers, or send anything.
- Findings stay labeled as claims; factual verification needs the full text.
- Fictional example: a made-up entry "Kim 2021; Abstract: survey of 214 remote knowledge workers" would be logged as Abstract-only with its response rate listed as missing.

## FAQ

**Can I paste fewer fields per entry?** Yes. Missing title, venue or abstract fields are recorded as UNKNOWN and grouped in the Data-Gaps note so you can chase the metadata later.

**Does the log judge paper quality?** No. The shortlist is a provisional triage tied to your working question, not a quality rating. Citation counts and prestige are excluded.

## Verification note

Source reviewed; runtime not tested. The prompt text and reference were inspected; no live run was performed.

## Source and thanks

Original TokRepo prompt, CC BY 4.0, category research. Reference context: [ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>) (reviewed 2026-10-04). The reference is context for a general-purpose AI workspace; this prompt depends on no specific new feature.

## Complete reusable prompt

You are helping me organize a literature-reading log from abstracts I paste in. I have NOT read the full papers. Your job is to build a structured, honest log I can use to triage what to read next, not to summarize papers as if abstracts were the complete text.

INPUT I will paste below a comma-or-semicolon-separated list of entries. Each entry may include: citation key or author-year, title, source or venue, year, abstract text, and optionally a URL or DOI and my own one-line reason for interest. Some fields will be missing. Do not fill gaps with guesses; mark them UNKNOWN.

MY GOAL is to answer: which abstracts actually address my question, which only seem to, and which I should read in full first. State my research question if I give it; otherwise use the working question '[WORKING QUESTION]' and flag that it is unset.

TASK
1. For each entry, output a row in a Markdown table with these columns exactly:
   - Key (use my citation key or a short author-year label you build from the title; if neither exists, use ENTRY-n)
   - Stated aim (one sentence, only what the abstract says)
   - Stated method (only what the abstract states; UNKNOWN if absent)
   - Claimed finding (only what the abstract asserts; mark as claimed, not verified)
   - Relevance to working question (High, Medium, Low, Unclear, with a one-clause reason)
   - Evidence status (Abstract-only; Full text needed; Field/definition ambiguous)
   - What the abstract does NOT tell me (specific missing inputs: sample, setting, measurement, comparison, limitations)
   - Next action (Read full text first / Skim methods / Defer / Verify citation)
2. After the table, add a SHORTLIST of the three entries most worth reading first, with a one-sentence justification each tied to the working question, and label the ranking as a provisional triage, not a quality judgment.
3. Add a CLAIMS-TO-VERIFY list of any statement that a reader might mistakenly treat as established fact but that only an abstract supports. For each, name what evidence in the full paper would settle it.
4. Add an OPEN QUESTIONS block listing what the batch as a whole cannot answer and what input I would need to supply (e.g., full texts, a numeric extract, a comparison table).
5. Add a DATA-GAPS note listing every UNKNOWN field across entries so I can chase missing metadata.

RULES
- Never state or imply that you read the full papers, ran a search, or accessed a database. You only organize what I pasted.
- Do not add papers, citations, numbers, or findings that are not in my input.
- If an abstract is ambiguous about whether a result is a finding or a hypothesis, mark it ambiguous and explain in one clause.
- Keep language plain and non-promotional; no impact-factor, popularity, or citation-count claims.
- If two entries look like the same work (preprint and published version), flag the possible duplicate and do not merge them.
- Output must be self-contained text I can paste into my notes. Do not schedule, send, or store anything.

REVIEW CHECKS before you finish:
- Does every table cell trace to words I supplied, or to UNKNOWN?
- Is every finding labeled as claimed rather than verified?
- Did you avoid ranking by perceived prestige?
- Are the shortlist and next actions consistent with the Relevance and Evidence status columns?
- If the working question was unset, did you say so clearly?

BOUNDARY: You are drafting an organizing log for me to review and act on; you are not performing research, retrieving sources, or making decisions for me.

FICTIONAL EXAMPLE INPUT (labelled):
Working question: 'How do remote teams maintain informal knowledge sharing?'
Entry A: 'Kim 2021; Informal knowledge flows in distributed teams; Journal of Org Work; 2021; Abstract: We survey 214 remote knowledge workers and report that ad hoc chat use correlates with perceived team memory. No sample breakdown or response rate given.'
Entry B: 'Untitled preprint; no year; Abstract mentions that shared documents may substitute for hallway conversation; no method stated; no sample.'

ILLUSTRATIVE OUTPUT SHAPE (not real data): a table row for Entry A with Stated method 'survey of 214 remote knowledge workers', Claimed finding 'ad hoc chat use correlates with perceived team memory', Evidence status 'Abstract-only', 'What the abstract does NOT tell me' listing response rate and sample breakdown, and Next action 'Skim methods'; a row for Entry B with UNKNOWN fields and Next action 'Verify citation'; then a Shortlist, Claims-to-Verify, Open Questions and Data-Gaps section.

PASS/FAIL CHECKS against the fictional example:
- PASS if Entry A's row labels the correlation as claimed and lists response rate as missing.
- PASS if Entry B is marked UNKNOWN/ambiguous rather than given an invented method.
- FAIL if any row invents a sample size, venue, or finding.
- FAIL if the shortlist implies a paper was read in full.

## 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 -->

# 摘要分诊日志：诚实可核的文献阅读笔记

把论文摘要粘贴进普通 AI 对话，即可得到标注为“声称”而非“已验证”的分诊表、短名单与数据缺口清单，全部内容仅来自你提供的材料。

## 开始使用

把完整提示词（附在本指南之后）复制并粘贴进一个接受文本输入的普通 AI 对话。紧接着在下一行粘贴你自己的清单：用逗号或分号分隔的一组条目，每条最好包含作者-年份或引用键、标题、来源/期刊、年份、摘要原文，以及可选的 URL/DOI 和你自己一句话的兴趣理由。如果你有研究问题，请写明；如果不写，提示词会使用 `[WORKING QUESTION]` 并标注其未设定。

此外无需任何准备：不需要终端、不需要账号配置、不需要 API 密钥。粘贴、发送、阅读回复。

### 如何检查输出

- 每个表格单元格都应能追溯到你粘贴的文字，或标注 UNKNOWN。检查是否出现你未提供的数字、样本量或期刊名。
- 结论必须表述为“声称”，绝不能表述为“已验证”。
- 短名单和下一步行动必须与“相关性”和“证据状态”两列一致。
- 如果你未设定研究问题，回复必须明确指出这一点。

如果某条条目缺少字段，请留空而不是猜测；提示词会把缺口标为 UNKNOWN。摘要越长，“所述方法”和“声称结论”两列越充实，但日志描述的仍然只是摘要内容。

## 简介

整理已收集的摘要与阅读论文是两回事。这条提示词会生成一份诚实的分诊日志：记录每条摘要说了什么、摘要没有告诉你什么，以及哪些条目最值得先读全文。它把摘要所声称的内容与读者可能误当作既定事实的内容区分开。通用 AI 摘要常模糊这条界线，而这份结构能帮助读者保持边界清晰。

## 权限与限制

- 适合输入：你自己的摘要清单、你自己的笔记，或你复制进来的公开元数据。
- 避免粘贴机密或未发表的手稿、个人数据，或你无权分享的内容。
- 对话只整理你粘贴的内容，不会检索数据库、运行查询、获取论文或发送任何内容。
- 结论始终标注为“声称”；事实核查仍需全文。
- 虚构示例：一条编造的条目“Kim 2021；摘要：调查 214 名远程知识工作者”会被记为仅凭摘要，并把缺少应答率列为缺口。

## 常见问题

**每条条目可以少粘一些字段吗？** 可以。缺失的标题、期刊或摘要字段会被记为 UNKNOWN，并汇总到数据缺口清单中，方便你日后补齐元数据。

**这份日志会评判论文质量吗？** 不会。短名单只是围绕你的研究问题做出的临时分诊，不是质量评级，也不包含引用次数或声望。

## 核查说明

来源已审阅；未进行运行时测试。已检查提示词文本与参考文献，未做实际运行。

## 来源与致谢

TokRepo 原创提示词，CC BY 4.0，类别为研究。参考背景：[ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>)（审阅于 2026-10-04）。该参考仅为通用 AI 工作环境的背景，本提示词不依赖任何特定新功能。

## 完整可复制提示词

你在帮我整理一份基于我粘贴进来的摘要的文献阅读日志。我并没有读过全文。你的任务是构建一份结构化、诚实的日志，供我对接下来该读什么进行分诊，而不是把摘要当作完整正文来总结论文。

输入：我将在下面粘贴一份用逗号或分号分隔的条目清单。每条条目可能包含：引用键或作者-年份、标题、来源或期刊、年份、摘要原文，以及可选的 URL 或 DOI 和我自己一句话的兴趣理由。有些字段会缺失。不要用猜测填补缺口；将其标为 UNKNOWN。

我的目标：回答以下问题——哪些摘要真正回应了我的问题，哪些只是看起来回应了，以及哪些我应该优先读全文。如果我给出研究问题，请写明；否则使用工作问题 '[WORKING QUESTION]' 并标注其未设定。

任务
1. 为每条条目在 Markdown 表格中输出一行，必须恰好包含以下各列：
   - Key（使用我的引用键，或你根据标题构建的简短作者-年份标签；若两者都不存在，使用 ENTRY-n）
   - 所述目标（一句话，只写摘要所述内容）
   - 所述方法（只写摘要所述内容；若无则标 UNKNOWN）
   - 声称结论（只写摘要所声称的内容；标注为声称，而非已验证）
   - 与工作问题的相关性（高、中、低、不明确，附一句原因）
   - 证据状态（仅凭摘要；需要全文；领域/定义含糊）
   - 摘要没有告诉我的信息（具体缺失的输入：样本、情境、测量、对照、局限）
   - 下一步行动（优先读全文 / 略读方法 / 暂缓 / 核实引用）
2. 在表格之后，添加一个短名单，列出最值得先读的三条条目，各附一句与工作问题挂钩的理由，并将该排序标注为临时分诊，而非质量评判。
3. 添加一个声称待核实清单，列出任何读者可能误当作既定事实、但仅有摘要支持的陈述。对每一条，指明全文中的什么证据可以对其进行确证。
4. 添加一个开放问题区块，列出这一批条目整体无法回答的内容，以及我需要补充提供的输入（例如全文、数值提取表、对照表）。
5. 添加一个数据缺口说明，列出各条目中所有 UNKNOWN 字段，以便我补齐缺失的元数据。

规则
- 绝不要陈述或暗示你读过全文、运行过检索或访问过数据库。你只整理我粘贴的内容。
- 不要添加我的输入中不存在的论文、引用、数字或结论。
- 如果摘要对某个结果是结论还是假设含糊不清，标为含糊，并用一句解释。
- 语言平实、不吹捧；不要提影响因子、人气或引用次数。
- 如果两条条目看起来是同一项工作（预印本与已发表版本），标记可能的重复，但不要合并它们。
- 输出必须是可粘贴进我笔记的自足文本。不要安排日程、发送或存储任何内容。

完成前的审查检查：
- 每个表格单元格是否都能追溯到我提供的文字，或为 UNKNOWN？
- 每个结论是否都标为声称而非已验证？
- 你是否避免了按感知声望进行排序？
- 短名单和下一步行动是否与相关性和证据状态两列一致？
- 如果工作问题未设定，你是否明确说明了这一点？

边界：你是在为我起草一份供我审查和行动的组织日志；你并非在执行研究、获取来源或替我做决定。

虚构示例输入（已标注）：
工作问题：“远程团队如何维持非正式知识共享？”
条目 A：“Kim 2021；分布式团队中的非正式知识流动；Journal of Org Work；2021；摘要：我们调查了 214 名远程知识工作者，并报告临时聊天使用与感知到的团队记忆相关。未给出样本细分或应答率。”
条目 B：“无标题预印本；无年份；摘要提到共享文档可能替代走廊交谈；未说明方法；无样本。”

示意性输出形态（非真实数据）：条目 A 的一行，所述方法为“调查 214 名远程知识工作者”，声称结论为“临时聊天使用与感知到的团队记忆相关”，证据状态为“仅凭摘要”，“摘要没有告诉我的内容”列出应答率和样本细分，下一步行动为“略读方法”；条目 B 的一行，字段为 UNKNOWN，下一步行动为“核实引用”；然后是短名单、待核实声称、开放问题和数据缺口部分。

针对该虚构示例的通过/失败检查：
- 如果条目 A 的行将相关性标注为声称，并列出缺少的应答率，则通过。
- 如果条目 B 被标为 UNKNOWN/含糊，而不是被赋予虚构的方法，则通过。
- 如果任何一行虚构样本量、期刊或结论，则失败。
- 如果短名单暗示某篇论文已被完整阅读，则失败。

## 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/abstract-triage-log-honest-literature-reading-notes-f0ff624c
Author: Prompt Lab