This guide gives you a reusable starting pattern. It is designed to help you see the work more clearly; it is not a substitute for judgment, source checking, or responsibility for the result.
Prepare the inputs before you ask for output.
The model only sees what you give it. Spend a few minutes naming the reader, desired result, and uncertain information. This makes a first draft easier to assess and reduces the need for decorative rewriting later.
Give the task a useful brief.
Help me create a reviewable theme map from customer feedback. Research question: [what we are trying to learn] Feedback records: [paste records with source labels, dates, and customer segment if known] Known limits: [sample size, source bias, missing groups] First, preserve each source label. Then return: (1) a small set of possible themes, (2) the records that support each theme, (3) counter-examples or disagreements, (4) wording that is directly quoted versus paraphrased, (5) what cannot be concluded from this sample, and (6) questions to verify next. Do not invent sentiment scores, counts, personas, or customer quotes.
Replace every bracketed field with your real context. Read the output before reuse.
Four steps that keep the result usable.
- 1
State the decision or research question before clustering comments.
- 2
Label every record with a source, date, and segment when available.
- 3
Ask for counter-examples and gaps alongside each proposed theme.
- 4
Review the original records before presenting a theme as a finding.
Themes need a trail back to the record
A useful theme map lets a reader move from a concise label back to the comments that support it. Keep source labels and a short evidence trail near the theme so a later decision does not depend on a persuasive but untraceable summary.
Absence is not agreement
A small sample cannot prove that people who did not respond agree. Notes from one channel can also over-represent a particular moment or customer type. Make those boundaries visible before comparing the loudest comments with the largest opportunity.
Look for the counter-example before naming the pattern
The fastest way to weaken an attractive theme is to ask what does not fit it. A different segment, a contrary comment, or an unclear source may change whether the next step is research, a product decision, or simply a better question.
A worked example: nine support interviews
Nine interviews mention “the search is slow” in different words. A theme map that labels each comment with its source and date can show that six complaints come from one heavy-usage segment, while three come from new users. That changes the next step from “make search faster for everyone” to a specific question about that segment’s queries.
Keep the sample visible
A theme map built from nine interviews is not a market verdict. Show the sample size, the channels, and the missing groups in the map itself. When a theme looks strong, the counter-examples and the boundaries of the sample are what keep a decision honest instead of convenient.
Run a human check.
- Can each theme be traced to labeled feedback records?
- Are direct quotations distinguished from paraphrases?
- Did the output avoid invented counts, sentiment scores, or personas?
- Are counter-examples and sampling limits visible to the decision owner?
AI is strongest here when it makes missing information, structure, and options easier to see. The moment an output becomes a claim, commitment, or decision, bring a person back into the loop.