Summarize common issues across support tickets with evidence
Group comparable ticket evidence while preserving denominators, uncertainty, duplicates, and severity
10 min setupTested with:ChatGPTReviewed: 2026-08-28
1
Add your context
Your text stays in this browser. AILesson Prompts does not send it to a model or server.
2
Your prompt
Unfilled fields remain visible as placeholders, so you can still copy and edit the prompt
Analyze common support issues from the supplied ticket dataset without overstating what tickets represent.
Ticket rows, IDs, dates, products, versions, channels, symptoms, outcomes, severity, and window:
[tickets]
Duplicate, missing, merged, bot, privacy, taxonomy, and collection rules:
[quality]
Question, dimensions, periods, thresholds, stakeholder, format, and decisions:
[analysis]
Validate row and unique-case counts, date boundaries, exclusions, duplicates, missingness, and taxonomy consistency before grouping. Preserve original issue text and map it to a transparent issue hierarchy; do not merge cases solely because words look similar. Distinguish symptom, suspected cause, confirmed cause, request type, and outcome. For each theme report unique cases, denominator, rate, severity mix, affected version or segment where supported, resolution status, representative anonymized locators, and confidence. Compare periods only with compatible collection rules and adequate denominators; calculate absolute and relative changes correctly. Do not infer customer prevalence, product defect, root cause, sentiment, or business impact from ticket volume alone. Flag channel, exposure, self-selection, repeat-contact, and release effects. Rank investigation priorities using explicit volume, severity, growth, uncertainty, and fixability criteria. End with validation samples, queries, and facts needed before action.
Private by defaultPrompt assembly happens locally in your browser. Avoid placing confidential information into any AI service unless your organization allows it.
From input to outcome
A worked example
See how concrete context turns this recipe into a usable result
Actual input
Ticket dataset
Aug 1-14 export tickets: 24 rows, IDs A01-A20 plus four follow-up rows sharing A03, A07, A07, A12. Of 20 unique cases: 8 blank PDF charts on Desktop 4.2, 3 blank charts with version missing, 4 CSV encoding complaints, 2 requests for scheduled export, 3 unrelated questions. Severity: blank charts 2 S2 and 9 S3; CSV all S3. Outcomes: 7 blank-chart cases used Firefox workaround, 2 unresolved, 2 unknown. Jul 15-28: 10 unique export cases, but chat tickets were not imported then; 3 blank-chart cases. Desktop 4.2 released Aug 2.
Data quality and exclusions
Deduplicate by case ID. Exclude follow-up rows from case count but retain attempts. Redact customer names. Version missing in 5/20 August cases. Taxonomy: Export > PDF rendering / CSV encoding / Feature request / Other. A known incident links 6 of the 11 August blank-chart cases; cause still under investigation. Chat import began Aug 1 and contributed 6 August cases.
Analysis decision
Support and product weekly review. Identify themes and investigation priority. Compare periods only if defensible. Output a compact table and cautions. Priority weights: severity high, then unique volume, recent growth, uncertainty. Minimum three cases for a theme.
Example output
Validation: 24 August rows reduce to 20 unique cases after four follow-up rows are linked by case ID. Follow-ups remain in attempt history. Five of 20 cases lack version; six cases arrived through a channel absent from July, so raw period comparison is not like-for-like.
Theme | Unique cases / 20 | Severity | Status and evidence | Confidence
PDF blank charts | 11 / 20 (55%) | 2 S2, 9 S3 | 7 workaround, 2 unresolved, 2 unknown; cases A01-A20 should be listed internally by mapped IDs | High on symptom count; medium on version association
CSV encoding | 4 / 20 (20%) | 4 S3 | Resolution details not supplied | High on count, low on outcome
Other themes | below 3-case threshold | mixed/unknown | 2 scheduled-export requests and 3 unrelated questions remain visible but are not labeled common themes | High
Priority 1: investigate PDF rendering because it has the largest unique volume and the only S2 cases. Six of 11 link to the known incident, but that association does not confirm one root cause for all 11. Eight explicitly report Desktop 4.2; three have missing version, so do not call it a 4.2-only defect.
Period caution: blank-chart cases rose from 3 of 10 in July to 11 of 20 in August, an observed +8 cases and +25 percentage points, but chat import began Aug 1 and the 4.2 release changed exposure. Do not report a trend until July chat backfill or a stable-channel comparison is available. Next checks: sample all 11 mappings, obtain missing versions, separate known-incident matches, inspect exposure by active 4.2 users, and recover CSV outcomes.
Why this works
1
Unique-case denominators prevent repeat contacts from masquerading as more affected customers.
2
Symptom-versus-cause separation avoids turning support labels into defect conclusions.
Check the result
Are counts based on documented unique-case and exclusion rules?
Are symptom, suspected cause, confirmed cause, and outcome separate?
Do comparisons preserve compatible denominators and collection methods?
ChatGPT