Review edition · Complete text lesson. Platform screenshots are being prepared.
Suppose you could…
Find the story in customer feedback
Turn a pile of comments into clear themes and a useful next step.
Group qualitative feedback into traceable themes, verify counts and distinguish a pattern from a representative conclusion.
Time is a practice estimate. Pause or return whenever you need.A useful little possibility.
A pile of feedback can make the loudest comment feel like the whole story. You’ll practise on eight fictional comments and create a short insight brief with supporting evidence. Your result: a theme summary and one small experiment to investigate, not a claim about all customers.
Let’s make a start.
Minute 0–3: read the sample before involving AI. Mark the comment IDs that concern finding or following booking instructions.
Minute 3–7: submit the prompt with the sample. Ask for categories and comment IDs, not only an executive summary.
Minute 7–11: recount the proposed groups manually. Check whether any comment is omitted, duplicated without explanation, or interpreted too strongly.
Minute 11–15: write a three-sentence brief with an observation, evidence and a proposed test. Save it with the source comments so someone else can check it.
Your sample material
Fictional training material. No real customer information needed.
FICTIONAL MEMBER COMMENTS — HARBOR WORKSPACE C1. I like the space, but finding the booking guide took too long. C2. A picture showing how to book a room would help. C3. The front-desk team answered my question quickly. C4. I could not find the first step for booking a room. C5. The welcome email arrived a day late. C6. The front-desk team was friendly. C7. A short getting-started guide would help me book a room. C8. The workspace was clean when I arrived.Download sample .txt ↓
Group the numbered fictional comments below into a small number of themes. For each theme give: a neutral label, exact supporting comment IDs, number of distinct comments, and one short exact quote. A comment can appear in more than one theme, but explain overlap and do not count it twice within one theme. Separate observations from interpretations. Do not invent comments, sentiment or population-level conclusions. Suggest one small test for the clearest problem, with uncertainty stated. COMMENTS: [Paste the sample input here.]
Keep personal and confidential information out of external AI tools. Start with the sample material.
Turn a first try into your thing.
If AI puts C3 under “confusing instructions,” challenge that classification: C3 describes a quick answer, not the cause of the question. Ask it to revise using explicit evidence only.
Write three sentences: what this sample suggests, which comments support it, and what you would test. Do not describe this as a survey of all members.
Pause for the “aha.”
Theme names involve judgement; exact wording can differ. Counts and cited IDs must still be reproducible. Four of these eight selected comments concern booking guidance, but the sample is too small and its selection too unclear to estimate a population percentage.
An AI summary is useful when it makes the evidence easier to inspect, not when it hides it behind a confident conclusion.
Useful beats convincing.
Check these against your actual output. Your answers stay on this device; this is self-review, not automated grading.
Finished your first try? Reveal a worked example ↓
One valid grouping:
• Booking guidance: C1, C2, C4, C7 — 4 comments. Quote: “A picture showing how to book a room would help.”
• Helpful front desk: C3, C6 — 2 comments. Quote: “The front-desk team was friendly.”
• Welcome-email timing: C5 — 1 comment.
• Cleanliness: C8 — 1 comment.
Three-sentence brief: Four comments in this small sample concern finding or following room-booking guidance. C1, C2, C4 and C7 support that theme, including a request for a picture of the booking process. We could test a clearer guide with a few new members and observe whether they can find the first step; this sample does not establish how common the problem is across all members.
Why it works: the recommendation is a test, and the supporting IDs make the interpretation inspectable.
A new case. Your own approach.
Apply the review principles to these new facts, not the previous sample’s specific numbers or answers.
WITHOUT THE STARTER PROMPT: Classify these new notes: A “Opening hours are hard to find”; B “I need weekend hours”; C “The staff are kind”; D “Where are opening hours listed?” Decide whether A/B/D belong together or in separate findability and availability themes. Explain your choice, count the IDs and suggest what you would need to learn next. Different defensible groupings can pass.
One possibility leads to another.
Apply the method to five non-sensitive meeting questions. Ask for “open questions” rather than “customer themes.” Before using real feedback in an external tool, remove identifiers and check that your organisation permits that use.
Your reusable workflow card ↓
WORKFLOW: EVIDENCE-BACKED THEMES Use when: a small set of comments needs an inspectable summary. Inputs: anonymised, numbered comments and the question you want to explore. Output: theme / IDs / distinct count / exact quote / uncertainty / possible test. Review: recount IDs, verify quotes, inspect ambiguous classifications, explain sample limitations. Done when: another person can reproduce each count and trace the recommendation to evidence. My category rules: __________ My next research question: __________Download workflow .txt ↓
You tried something. That counts.
Keep a note of what you made or what surprised you. It stays in this browser.