Prompt library
Summarize survey results without flattering them
Leading with response reality and demanding a base for every percentage inoculates the readout against the overclaiming that makes survey data untrusted. Typifying verbatims and minimum cell sizes keep the qualitative and subgroup layers evidence-grade.
Last reviewed July 17, 2026
The prompt
Summarize the survey results below.
Survey purpose and who was asked: {{purpose}}
Results: {{results}}
Structure:
1. Response reality first: sent, completed, completion rate, and who the respondents over- and under-represent relative to the population. Every finding below inherits this caveat; state it once, precisely, here.
2. Headline findings: 3-5, each with the number, the base ("62 percent of 214 respondents"), and the question wording it comes from, because wording drives answers.
3. Cross-cuts: only the differences that survive a sanity check on subgroup size (state the minimum cell size you used). Differences on 12 respondents are anecdotes, and get labeled as such.
4. The open-text mine: cluster free-text answers into themes with counts and one verbatim quote each. Verbatims are the most persuasive artifact a survey produces; pick the ones that typify, not the most extreme.
5. What the survey cannot tell us: the questions people will ask in the meeting that this instrument cannot answer, so nobody stretches it.
6. The 3 decisions this data actually supports.
Never report a percentage without its base.Run in idaptOpens a new chat with the prompt prefilled. Nothing sends until you press send.
Fill in the variables
| Variable | What it is | Example |
|---|---|---|
| {{purpose}} | Why the survey ran, who got it | churn-risk survey to 800 customers, 214 completed |
| {{results}} | The export | [paste the results export or attach the CSV] |