{"id":"cx-csat-comment-themes-table","audience":"customer-support","category":"support-work","technique":"grounding","track":"support","level":"intermediate","title":"Count CSAT comment themes into a fixed table","free":false,"task":"Your manager wants this week's survey comments sorted into her four themes: Slow reply, Repeated myself, Helpful agent, Not about support. Make a table with exactly the columns Theme, Count, Example, one row per theme, where Example is a short word-for-word quote. Put each comment in exactly one theme, and add one line after the table saying which theme has the most comments. Plain text table, no bold or headings.","materialLabel":"Your material (attached to your prompt automatically)","material":"CSAT COMMENTS - week of Oct 13 (10 responses, score in brackets)\n\n1. [2] \"Waited two days for a first answer.\"\n2. [5] \"Lena sorted it out in five minutes, amazing.\"\n3. [1] \"I had to explain my issue three times to three people.\"\n4. [2] \"The courier left my package in the rain. Not happy.\"\n5. [3] \"Took forever to hear back, but fixed in the end.\"\n6. [5] \"Very helpful and patient agent.\"\n7. [1] \"Every new person asked me for my order number again.\"\n8. [2] \"No reply for 48 hours on a billing problem.\"\n9. [4] \"Marco was great, really knew the product.\"\n10. [2] \"Nobody answered my chat for an hour, then I had to repeat everything to the next agent.\"\n\nNote from last week's report: \"Slow reply\" was 5 comments.","rules":[{"type":"plainText"},{"type":"maxWords","max":140}],"mistakes":[{"title":"Letting the AI pick the themes","why":"The AI invents its own themes, so this week can't be compared with last week.","fix":"Give them: \"Use only these 4 themes: Slow reply, Repeated myself, Helpful agent, Not about support.\"","checks":["r0"]},{"title":"Paraphrasing the quotes","why":"The AI tidies quotes into its own words, which then aren't real customer quotes.","fix":"Say: \"Example is a short word-for-word quote.\"","checks":["r1"]},{"title":"Double counting","why":"A comment about waiting and repeating gets counted twice, so totals don't match the 10 responses.","fix":"Add: \"Each comment goes in exactly one theme; counts must add up to 10.\"","checks":["r1","r2"]},{"title":"Getting a report back","why":"AI writes headings, bold and a long analysis around the table.","fix":"Say: \"Output only the table and one line after it. No bold or headings.\"","checks":["plainText","maxWords"]}],"promptHints":[{"label":"Names the four themes","patterns":["slow reply","repeated myself","helpful agent"]},{"label":"Asks for exact quotes","patterns":["word.for.word","verbatim","exact quote"]},{"label":"One theme per comment","patterns":["exactly one theme","only one theme","add up to 10","count once"]},{"label":"Gives the columns","patterns":["theme.{0,5}count.{0,5}example","columns"]},{"label":"Asks for plain text","patterns":["plain","no bold","no markdown","no formatting","no headings"]}],"hint":"Counts only mean something if the buckets are fixed. Give the AI the themes, the columns and the rule that each comment is counted once.","rubricLabels":["The table has exactly the columns Theme, Count, Example and exactly the four themes Slow reply, Repeated myself, Helpful agent, Not about support","The counts add up to 10 and each Example is a word-for-word quote from a comment in that theme","Not about support has a count of 1 (the courier comment), and the line after the table names the theme with the most comments, consistent with the counts in the table"]}