The prompt
Score 10 of 10, Excellent
Use it when You have rating-scale and multiple-choice answers from a staff, customer or student survey and need percentages, comparisons and a summary you can defend.
- {{survey_description}}
- {{response_counts}}
- {{groups}}
- {{min_group_size}}
- {{audience}}
- {{app_and_version}}
- {{results}}
Full prompt and why it works
Prompt
The survey and who answered: {{survey_description}} How many were invited and how many responded: {{response_counts}} Groups to compare, if any: {{groups}} Smallest group to report on its own: {{min_group_size}} responses Who will read the summary: {{audience}} App I am using: {{app_and_version}} The results (counts per answer for each question, or the raw responses): """ {{results}} """ Task: Summarize the closed-ended results above for {{audience}}, so that they see what the survey shows and how much weight each result can bear. Open-ended comments are out of scope here. Format: 1. Response: the response rate, and a sentence on who may be under-represented, if the data shows it. 2. Results table: for each question, how many answered it, the percentage for each answer, and for rating scales the share who agreed or were satisfied (the top two boxes) beside the share who disagreed. Whole-number percentages, each with the count it is based on. 3. Group comparisons: only where each group has at least {{min_group_size}} responses; mark smaller groups "too few to compare" and do not show their percentages. 4. Key findings: up to 5, one sentence each with its numbers, highest and lowest results first. 5. Formulas: the {{app_and_version}} formulas that reproduce the percentages and top-two-box shares from raw responses, with one question worked through so I can check them. Requirements: - Under 450 words, plus the table. - Calculate each percentage from the people who answered that question, not from everyone invited, and say so in a note under the table. - Do not average percentages across groups of different sizes; recalculate from the counts. - Give a mean rating only beside the percentages, because a mean can hide a split between strong agreement and strong disagreement. - Say "percentage points" for the difference between two percentages. - Do not explain why people answered as they did; the numbers do not show reasons. A good summary is one where every percentage can be traced back to a count in the results. Example of one finding's form only, with invented numbers: "Agreement that people have the tools they need (Q4) is 15 percentage points lower in the warehouse team (51% of 47) than in the office team (66% of 61)." If the counts for a question do not add up to the number who answered it, or a scale's labels are missing, flag it and leave that question out of the findings until I confirm.
Variables
- {{survey_description}}
- For example "annual staff engagement survey, all employees, May 2025".
- {{response_counts}}
- For example "412 invited, 287 responded".
- {{groups}}
- For example "department and length of service". Write "none" for no comparisons.
- {{min_group_size}}
- For example 10. Many organizations use 5 or 10 to protect anonymity.
- {{audience}}
- For example "the leadership team" or "all staff".
- {{app_and_version}}
- For example "Excel for Microsoft 365" or "Google Sheets".
- {{results}}
- Counts per answer for each question with the question wording, or the raw responses with one row per person. No names.
Why it works
- What good looks like. Every percentage shows the count it is based on and is calculated from those who answered, so readers can judge how much weight each number bears.
- What to avoid. It will not average percentages across uneven groups, report a mean without the spread, or guess at reasons, the usual ways survey summaries mislead.
- If unsure. Groups below your minimum size are marked too few to compare, and counts that do not add up are flagged before they reach a finding.