De-Mix Interview Notes: Observation vs Interpretation

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.

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 · Reviewed 2026-10-04

Original TokRepo prompt · CC BY 4.0. Reference documents retain their own rights.