Support / PromptFreeIntermediateRun on ChatGPT, Claude and Gemini on 14 Nov 2026
Find the themes in a week of support tickets
Groups a week of tickets into themes with counts and customer quotes, and says which need a help article, a saved reply or a product fix.
Use it when
You want to tell your team or the product team what customers struggled with this week, with numbers rather than impressions.
You will need
- The week's tickets, exported with subject, message and date
- The main areas of your product or service
- Last week's themes, if you have them
The prompt
ROLE
You are a support team lead who turns tickets into clear priorities.
CONTEXT
Tickets: [UPLOAD OR PASTE THE EXPORT: SUBJECT, MESSAGE, DATE]
Product areas: [THE MAIN PARTS OF YOUR PRODUCT OR SERVICE]
Last week's themes: [PASTE THEM, OR WRITE NONE]
TASK
Find the themes in this week's tickets and say what to do about each.
STEPS
1. Say how many tickets you read. If that is fewer than the export holds, tell me and stop.
2. Group the tickets into themes. A theme needs at least three tickets.
3. Count the tickets per theme and pick one quote that sums it up.
4. For each theme, suggest one fix: a help article, a saved reply, or a product change.
5. Compare with last week's themes, if I gave them: new, growing, shrinking.
RULES
- Quote customers' exact words. Never reword a quote.
- Count, do not estimate.
- Do not guess the cause of a problem. If you suggest one, mark it [CHECK].
OUTPUT
A table with columns: theme, tickets, quote, suggested fix, trend. Under it, the one theme I should raise with the product team, and why.
Example output, shortened
Tickets read: 312.
Theme: cannot find the invoice. 23 tickets. "Where do I download a VAT invoice?" Fix: help article, and a link in the receipt email. Trend: new.
Sample for an invented accounting app's help desk. Your output will differ.
Make it better
- Write the help article for the biggest theme, in the words customers used.
- Which theme points to a product problem rather than a missing help article? Write a short note for the product team with the evidence.
Watch out for
With a long export, some tools read only part of it and the counts come out wrong. Compare the number of tickets read with your export, and spot-check two themes by searching your help desk.
How we test
On 14 Nov 2026, a script filled in this prompt with an invented work example and ran it on ChatGPT (GPT-6.1 Terra), Claude Sonnet 5.5 and Gemini 4 Hydrogen. A separate AI model then checked each answer against what this page says the prompt does and what to watch out for. The label only appears when most answers pass. No person tried it by hand, and your answers will differ with your own input.