# De-Mix Interview Notes: Observation vs Interpretation

> An original prompt that turns raw interview or field notes into a table separating participant wording from analyst judgment, plus themes and gaps.

## Install

Copy the content below into your project:

# De-Mix Interview Notes: Observation vs Interpretation

An original prompt that turns raw interview or field notes into a table separating participant wording from analyst judgment, plus themes and gaps.

## Start here

Paste your interview or field notes into an ordinary AI chat that accepts text. Optionally add your interview guide and participant ID codes. Then paste the prompt (appended separately). If you paste no notes, the prompt will ask you for them and stop.

**How to check the output**

- Quotation-marked cells contain only the participant's words.
- Every INTERPRETATION row cites support or says NO SUPPORTING QUOTE.
- Every MIXED row became two labelled rows.
- Every theme cites row numbers and states contradicting rows.

## What this prompt does

Raw notes usually blend what a participant said with what the note-taker concluded. This prompt sorts each meaningful unit into OBSERVATION, INTERPRETATION or MIXED, keeps short verbatim fragments with participant ID and timestamp, and closes with WHAT I STILL NEED and VERIFICATION NOTES. Themes cite row numbers, and confidence reflects only the rows present.

## Prerequisites and permissions

- No terminal, API or account setup is needed; a plain text chat is enough.
- You need your own notes. The prompt adds no facts, motives or demographics beyond them.
- Identifying details are marked [REDACT IF SHARED], but review before sharing notes with any tool.

## Limitations

- It only structures what your notes contain; it does not give hiring, diagnosis, legal or performance advice.
- It does not claim themes represent any population, and one participant stays one participant.
- Source reviewed; runtime not tested.

## FAQ

**Can I use it for several participants at once?** Yes, the input accepts notes for one or more participants; keep participant IDs consistent.

**What if I have no verbatim wording?** The quote field is marked UNAVAILABLE; wording is never invented.

## Source and thanks

Original TokRepo prompt (CC BY 4.0). Context reference: [ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>). The complete prompt and translation are appended.

## Complete reusable prompt

You are helping me de-mix interview notes into observation and interpretation. Your job is to keep the participant's words separate from any analyst judgment, and to flag where I am guessing rather than reporting.

INPUT I WILL PASTE
- Interview or field notes for one or more participants, in whatever rough form I have them (fragments, quotes, my asides, timestamps, comments).
- Optional: the interview guide or question list, and any participant ID codes I use.
- If I paste nothing, ask me for the notes and stop.

TASK
1. Read the entire pasted material once. Do not summarize it yet.
2. Split every meaningful unit into exactly one of these rows:
   a. OBSERVATION - something the participant said or did, close to their own words.
   b. INTERPRETATION - a meaning, cause, motive or category I (the note-taker) appear to have added.
   c. MIXED - the note bundles a statement and a judgment in one line and must be separated.
3. For every OBSERVATION row, keep a short verbatim fragment in quotation marks plus its participant ID and, if present, a timestamp. If the exact words are missing, mark the quote field as UNAVAILABLE and do not invent wording.
4. For every INTERPRETATION row, state what evidence in the notes supports it, or write NO SUPPORTING QUOTE.
5. For every MIXED row, produce the two replacement rows (one OBSERVATION, one INTERPRETATION) and label the row number it came from.
6. End with a short section titled WHAT I STILL NEED, listing only questions that would change how a row is classified: missing verbatim wording, unclear speaker, ambiguous referent, missing context before or after.

OUTPUT FORMAT
A Markdown table with columns: Row | Type | Participant | Quote or Statement | Source location | Analyst note | Support in notes.
Then a second table titled CANDIDATE THEMES with columns: Theme | Supporting observation rows | Contradicting rows | Confidence (High/Medium/Low) | What would raise confidence. Themes must cite row numbers, and confidence must reflect only the rows present.
Then the WHAT I STILL NEED list.
Then a short VERIFICATION NOTES section describing any place where I had to guess and any place where two rows conflict.

RULES AND BOUNDARIES
- Never put my interpretation inside quotation marks.
- Never upgrade a single participant's comment into a general finding.
- Never add facts, motives or demographics not present in the notes.
- Preserve the participant's own wording, including hesitations, hedges and informal phrasing, where I supplied it.
- Mark any name, company or identifying detail as [REDACT IF SHARED] instead of repeating it.
- Do not give advice on hiring, diagnosis, legal matters or performance decisions; only structure what the notes contain.
- Do not claim the themes are representative of any population.
- If I ask later to send, publish or schedule anything based on this table, treat that as a separate step and do not perform it here.

WORKED EXAMPLE (fictional, for format only)
Input fragment: 'P3, ~04:12 - said they stopped using the onboarding checklist because it felt pointless; honestly seems like the team just doesn't care about new users.'
Expected output rows:
- Row 1 | OBSERVATION | P3 | 'stopped using the onboarding checklist because it felt pointless' | ~04:12 | none | direct quote.
- Row 2 | INTERPRETATION | P3 | 'the team just doesn't care about new users' | ~04:12 | my inference from one comment | NO SUPPORTING QUOTE.
A CANDIDATE THEMES row might read: Onboarding checklist relevance | Rows 1, 4, 9 | Row 12 | Medium | a verbatim reason from a second participant.

CHECKS BEFORE YOU FINISH
- Every quotation-marked cell contains only participant words from my notes.
- Every INTERPRETATION row either cites supporting text or says NO SUPPORTING QUOTE.
- Every MIXED row has been replaced by two rows and labelled.
- Every theme cites at least one row and states contradicting rows if present.
- No theme claims to represent a population.
- My input notes remain the only source of content.

## 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 -->

# 访谈笔记拆分：区分观察与解释

一份原创提示词，把原始访谈或田野笔记整理成表格，将受访者原话与分析者判断分开，并列出候选主题与待补信息。

## 开始使用

把访谈或田野笔记粘贴到可以接受文本的普通 AI 对话中。可选：附上访谈提纲和参与者编号。然后粘贴提示词（单独附上）。如果没有粘贴笔记，提示词会先向你索取笔记并停止。

**如何检查输出**

- 带引号的单元格只包含受访者原话。
- 每条 INTERPRETATION 行都注明依据或写 NO SUPPORTING QUOTE。
- 每条 MIXED 行都被替换为两行并标注来源。
- 每个主题都引用行号，并写出矛盾行。

## 这份提示词做什么

原始笔记常把「参与者说了什么」和「记录者得出什么结论」混在一起。这份提示词把每个有意义单元归为 OBSERVATION、INTERPRETATION 或 MIXED，保留带参与者编号和时间戳的简短原话，并以 WHAT I STILL NEED 和 VERIFICATION NOTES 收尾。主题引用行号，置信度只反映现有行。

## 前置条件与权限

- 不需要终端、API 或账号设置，纯文本对话即可。
- 需要你自己的笔记。提示词不会添加笔记以外的信息、动机或人口统计特征。
- 身份信息会标为 [REDACT IF SHARED]，但把笔记交给任何工具前仍请自行检查。

## 限制

- 它只整理笔记内容，不提供招聘、诊断、法律或绩效建议。
- 它不声称主题代表任何群体，一名参与者始终只代表一人。
- 已审阅来源；未做运行测试。

## 常见问题

**可以一次处理多名参与者吗？** 可以，输入接受一名或多名参与者的笔记，请保持参与者编号一致。

**如果没有原话怎么办？** 引号字段会标为 UNAVAILABLE，绝不编造措辞。

## 来源与致谢

TokRepo 原创提示词（CC BY 4.0）。背景参考：[ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>)。完整提示词与译文单独附上。

## 完整可复制提示词

你正在帮助我把访谈笔记拆分还原为观察与解释。你的工作是让参与者自己的话与分析者的判断保持分离，并标出哪些地方是我在猜测而不是在转述。

我将粘贴的输入
- 一名或多名参与者的访谈或田野笔记，形式随意，按我手头的原样（片段、引语、我的旁批、时间戳、评论）。
- 可选：访谈提纲或问题清单，以及我使用的任何参与者编号。
- 如果我没有粘贴任何内容，请向我索取笔记并停止。

任务
1. 把粘贴的全部材料通读一遍。暂时不要总结。
2. 把每个有意义单元归入且仅归入以下其中一行：
   a. OBSERVATION（观察）——参与者说过或做过的事，贴近其原话。
   b. INTERPRETATION（解释）——我（记录者）似乎添加的某种含义、原因、动机或类别。
   c. MIXED（混合）——笔记在一行里把陈述和判断混在一起，必须拆分。
3. 对每一行 OBSERVATION，保留一段加引号的简短原话片段，并附上参与者编号，如有时间戳也一并附上。如果原话缺失，把引语字段标为 UNAVAILABLE，不要编造措辞。
4. 对每一行 INTERPRETATION，说明笔记中有哪些证据支持它，或写 NO SUPPORTING QUOTE。
5. 对每一行 MIXED，产出两行替换行（一行 OBSERVATION，一行 INTERPRETATION），并标注它来自第几行。
6. 以一个小节结尾，标题为 WHAT I STILL NEED，只列出会改变某行分类的问题：缺失的原话、说话者不明、指代不清、前后语境缺失。

输出格式
用 Markdown 表格，列包括：Row | Type | Participant | Quote or Statement | Source location | Analyst note | Support in notes。
然后第二个表格标题为 CANDIDATE THEMES，列包括：Theme | Supporting observation rows | Contradicting rows | Confidence (High/Medium/Low) | What would raise confidence。主题必须引用行号，置信度必须只反映现有行。
然后是 WHAT I STILL NEED 列表。
然后是一个简短的 VERIFICATION NOTES 小节，描述任何我不得不猜测的地方，以及任何两行相互冲突的地方。

规则与边界
- 绝不把我的解释放进引号内。
- 绝不把单个参与者的评论升级为普遍结论。
- 绝不添加笔记中不存在的事实、动机或人口统计特征。
- 在我提供了原话的情况下，保留参与者自己的措辞，包括犹豫、模糊限定词和非正式表达。
- 把任何姓名、公司或身份细节标为 [REDACT IF SHARED]，不要重复它。
- 不要就招聘、诊断、法律事务或绩效决策提供建议；只整理笔记内容。
- 不要声称主题代表任何群体。
- 如果我之后要求基于这张表发送、发布或安排任何事情，把它视为单独步骤，不要在这里执行。

示例（虚构，仅用于展示格式）
输入片段：'P3, ~04:12 - said they stopped using the onboarding checklist because it felt pointless; honestly seems like the team just doesn't care about new users.'
预期输出行：
- Row 1 | OBSERVATION | P3 | 'stopped using the onboarding checklist because it felt pointless' | ~04:12 | none | direct quote.
- Row 2 | INTERPRETATION | P3 | 'the team just doesn't care about new users' | ~04:12 | my inference from one comment | NO SUPPORTING QUOTE.
一行 CANDIDATE THEMES 可能写成：Onboarding checklist relevance | Rows 1, 4, 9 | Row 12 | Medium | a verbatim reason from a second participant.

完成前的检查
- 每个带引号的单元格只包含我笔记中的受访者原话。
- 每条 INTERPRETATION 行要么引用支持性文本，要么写 NO SUPPORTING QUOTE。
- 每条 MIXED 行都已替换为两行并标注。
- 每个主题至少引用一行，并在存在矛盾行时写出矛盾行。
- 没有任何主题声称代表某个群体。
- 我的输入笔记仍然是唯一的内容来源。

## 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/de-mix-interview-notes-observation-vs-interpretation-5f35a4da
Author: Prompt Lab