Designs unbiased surveys - question types, Likert and frequency scales, screener-to-demographics flow, and a cognitive pilot plan - that produce trustworthy quantitative data. Use when someone says "write survey questions about X", "review my survey for bias", "how long should this questionnaire be", or wants to quantify attitudes, behaviors, or satisfaction. Do NOT use for full study design with hypotheses, sampling method, and power analysis - use primary-research instead; for qualitative interview scripts, use interview-guide-builder; for segment profiles built from the results, use user-persona.
Click to play with sound.
---
name: Survey Designer
description: Designs unbiased surveys - question types, Likert and frequency scales, screener-to-demographics flow, and a cognitive pilot plan - that produce trustworthy quantitative data. Use when someone says "write survey questions about X", "review my survey for bias", "how long should this questionnaire be", or wants to quantify attitudes, behaviors, or satisfaction. Do NOT use for full study design with hypotheses, sampling method, and power analysis - use primary-research instead; for qualitative interview scripts, use interview-guide-builder; for segment profiles built from the results, use user-persona.
---
# Survey Designer
A poorly designed survey produces precise measurements of the wrong thing - and because the numbers look rigorous, teams act on them with more confidence than they ever would on a hunch. This skill covers question construction, scale selection, and flow decisions that determine whether survey data is trustworthy, and it catches the leading, loaded, and double-barreled questions that quietly bias results before fielding, when fixes are free.
## Inputs to collect
1. The decision the survey informs, and the 2-4 constructs to measure (satisfaction, frequency, intent). A survey without a decision attached collects trivia.
2. The audience and how they will be reached (panel, in-product intercept, email list). This sets the length budget.
3. Expected responses. Under ~100 completes, differences between groups will rarely be meaningful - flag it and consider qualitative methods instead.
4. Any questions that must be kept for trend continuity - historical comparability beats wording perfection; never rewrite a tracking question mid-trend.
5. Length budget. Default: under 10 minutes for general audiences, under 5 for transactional intercepts. Roughly 3-4 closed questions per minute, so a 10-minute survey is ~30 closed items maximum.
## Operating procedure
Order matters: constructs before questions, questions before flow, pilot before fielding - reversing any pair produces rework or, worse, fielded bias.
1. **Map each construct to question types.** Closed questions (single-select, multi-select, scale) for anything that must be quantified and compared; open-ended sparingly - they inflate completion time and require qualitative analysis. A well-designed survey is mostly closed with one or two open "anything else" fields. Matrix questions look efficient but inflate satisficing (respondents clicking down one column); cap at 5 rows.
2. **Choose scales.** For attitudes and satisfaction: 5-point Likert, labeled endpoints, neutral midpoint. Do not reverse-code items without a clear analytical need - it confuses respondents more than it catches inattention. For frequency: concrete anchors ("never", "once a month", "once a week", "daily"), never vague ones ("rarely", "sometimes", "often") - respondents map vague anchors to different realities. Reserve 0-10 NPS for likelihood-to-recommend only; never repurpose it as a general satisfaction scale.
3. **Run every question through the bias tests:**
- *Leading* - embeds an assumption. Fails: "How much did you enjoy the new feature?" Passes: "How would you rate your experience with the new feature?"