De-identify and label each message
Remove handles, names, IDs, private links and quoted personal details. Give each message a local label such as M1; keep exact wording after the label.
Follow the steps
Step 1 of 6
Your worksheet stays in this page and is lost when you leave or reload. Copy it before leaving. Only guide completion is saved. Do not enter sensitive information.
Read all steps
De-identify and label each message
Remove handles, names, IDs, private links and quoted personal details. Give each message a local label such as M1; keep exact wording after the label.
Try a synthetic message set
M1: “The export stopped at 80% twice today.” M2: “Could you add a CSV option?” M3: “I finally found the filter, but the labels were confusing.” M4: “The export stopped at 80% twice today.”
Copy the evidence-preserving prompt
Group each labelled message under problem, suggestion and/or expression of feeling. Quote the exact wording and keep mixed messages in every relevant group. Flag exact duplicates. Count messages, not people. Do not infer identity, severity, root cause or popularity. End with questions for follow-up. Messages:
Compare with a bounded reference
Worked example — four supplied messages, not four people Problems • M1: “The export stopped at 80% twice today.” This reports an export stopping. M4 repeats the exact same words; retain both labels and flag the pair as an exact text duplicate. We do not know whether they describe the same attempt, separate attempts or separate people. • M3: “I finally found the filter, but the labels were confusing.” This reports difficulty finding or understanding the filter. The filter was eventually found; do not turn this into “filter unavailable”. Suggestion • M2: “Could you add a CSV option?” This requests an option. It does not establish whether CSV already exists, why it is wanted or how many people want it. Feeling or reaction • M3: “I finally found the filter, but the labels were confusing.” Keep it here as well as under problems. The wording describes an experience; a precise emotion or sentiment score is not supplied. Counting note There are four labelled messages and three distinct wordings. M1/M4 are one exact-text duplicate pair. M3 belongs in two groups but remains one message. The number of people, independent incidents and affected users is unknown. Category totals overlap; do not add them up as an audience count. Questions to investigate • M1/M4: Which export, at what time, and with what visible error? Ask only for relevant technical details through the agreed channel, with personal or secret data removed. • M2: What would the CSV be used for, and which fields are needed? Check the existing export options before deciding this needs a new feature. • M3: Which screen and labels were confusing? Ask what wording the reader expected. Repair example Wrong: “Four users reported a critical export outage and want CSV urgently.” Better: “Four messages were supplied. M1 and M4 share the same export-stopping report; M2 requests CSV; M3 describes confusing filter labels. People count, severity, cause and urgency are not established.” A duplicate message is not a surprise new customer.
Audit labels, quotes and claims
Check your draft against the messages, one label at a time. 1. Can you find M1, M2, M3 and M4, with their original wording? Keep the duplicate label instead of silently deleting it. 2. Does M3 keep both its usability problem and reaction, without turning one message into two people? 3. Are message count, distinct wording and unknown people count clearly separated? “Twice” is part of the report, not proof of two independently verified incidents. 4. Have you removed invented causes, severity, popularity, sentiment scores and delivery promises? For your own messages, repeat these checks against your own labels; do not copy the example’s counts. If a quote contains identifying details, redact those details and mark the redaction before using an assistant. Keep the remaining words faithful. Fix the draft first, then tick the checks below.
Choose the next evidence step
Copy a handoff containing the grouped quotes, duplicate labels, counting note and open questions. This worksheet does not save your edited draft; keep a copy before leaving. Choose one follow-up to start with. For this example, investigate the export report, clarify the CSV use case or examine the filter labels. Record who will check it only after someone agrees; no owner or deadline is supplied here. These are proposed investigations, not promised fixes or messages already sent. If you need to compare themes across a larger set, continue to the related feedback-analysis Task with the same checked messages and duplicate notes. Do not treat its fictional example as additional feedback. You can keep using your chosen assistant or work manually; changing tools is optional. You are finished when another reader can trace every summary back to a message and see what still needs checking. This does not mean the product issue is fixed or the community has received a reply.
Finished this guide?
Saved for this browser session. Sign in keeps it attached to you on this device.
Synthetic sample. Counts describe only the supplied messages; they do not measure unique people, prevalence, severity or root cause.
Report a correction
Point out a missing step or an unsafe claim. Your note is saved in your workspace for review.
