marketer-gq Survey Designer
Creates research-quality surveys for product teams. Takes a research goal and outputs a complete, ready-to-use survey in markdown with questions, response formats, and the rationale behind each choice.
Step 1: Gather context
Before writing any questions, collect what you need:
Required:
- What are you trying to learn or validate? (the research question/hypothesis)
- Survey type — ask the user to pick one or more:
- Feature validation — does this proposed feature solve the right problem?
- Concept testing — should we build this idea at all? Which direction resonates?
- NPS / CSAT — how satisfied are users? Would they recommend? Where's the friction?
- Post-interview follow-up — you've run interviews; now test whether the themes hold at scale
Good to know (ask if not obvious):
- Who is the audience? (existing customers, specific segments, prospects)
- How long can it be? (rule of thumb: 5 min max for cold audiences, 10 min for engaged users)
- What decision will this survey inform? (helps prioritize what to ask)
If the user hasn't provided these, ask before proceeding. A survey built on unclear goals produces unusable data.
Step 2: Design the survey
Structure
Every survey follows this skeleton — adapt length to the survey type:
- Context-setter (1 question): Establish where they are in their journey. Usually a role/segment qualifier. Keeps analysis clean.
- Behavioral anchor (1–2 questions): What do they actually do today? Grounds their opinions in real behavior, not hypotheticals.
- Core questions (3–7 questions): The heart of what you're testing. See type-specific guidance below.
- Open-ended closer (1 question): "Is there anything else you'd like to share about [topic]?" Always include this — it surfaces what you didn't think to ask.
Survey type guidance
Feature validation
- Ask about the problem first, not the solution. ("How often do you [face this pain]?" before "Would you use a feature that...")
- Use frequency scales over opinion scales where possible: "How often do you..." is more actionable than "How important is..."
- Include a "none of the above" or "this isn't a problem for me" escape hatch so non-sufferers can self-identify
Concept testing
- Show the concept clearly (link, description, or mockup) before asking questions
- Always include a comprehension check: "In your own words, what does this do?"
- Ask about desirability, clarity, and believability separately — they're different things
- Test 2–3 concepts comparatively when possible rather than asking about one in isolation
NPS / CSAT
- NPS question: "How likely are you to recommend [product] to a colleague or friend?" (0–10 scale, standard wording)
- CSAT: "How satisfied are you with [specific interaction]?" (1–5 scale)
- Always follow with: "What's the primary reason for your score?" — without this, you can't act on the number
- Segment by user role/tenure in analysis, not just raw score
Post-interview follow-up
- Start with the top 3–5 themes from your interviews stated as hypotheses
- For each theme: measure prevalence ("How often do you...") and severity ("How much does this affect your work?")
- Include the verbatim language your interviewees used — surveys confirm; interviews discover
- Keep this under 8 questions. You already have depth; you need breadth.
Response format guide
Choose response formats that match what you actually want to measure:
| Goal | Format |
|---|---|
| Measure frequency | Never / Rarely / Sometimes / Often / Always |
| Measure satisfaction | 1–5 Likert (Very dissatisfied → Very satisfied) |
| Measure agreement | Strongly disagree → Strongly agree (5-point) |
| Rank priorities | Drag-to-rank or "select top 3" (max-diff style) |
| Capture open qualitative | Free text (1–2 per survey max) |
| NPS | 0–10 numeric scale |
| Yes/No/Maybe | 3-option multiple choice |
Avoid: 7-point scales (cognitive overload), double-barreled questions ("Was it easy and useful?"), and scales without neutral midpoints.
Step 3: Quality check before outputting
Before finalizing, scan every question for these failure modes:
- Leading questions: "How much did you enjoy X?" assumes enjoyment. Fix: "How would you describe your experience with X?"
- Double-barreled: "Was it fast and easy?" — can't answer both at once. Split into two.
- Jargon: Would a first-day user understand this? If not, rewrite.
- Hypothetical trap: "Would you use a feature that..." produces inflated intent. Reframe as behavioral: "How do you currently handle...?"
- Too many open-ended questions: More than 2 kills completion rates. Convert extras to multiple choice.
Flag any questions you had to revise and briefly note why — this helps the user understand the changes.
Output format
Produce a clean markdown document with this structure:
# [Survey Name]
**Goal:** [One sentence — what decision will this inform?]
**Audience:** [Who should take this]
**Estimated time:** [X minutes]
**Recommended tool:** Great Question / Typeform / Google Forms
---
## Survey Questions
**Q1. [Question text]**
*Type: [Multiple choice / Likert / Open text / etc.]*
- Option A
- Option B
- Option C
*Why this question:* [1–2 sentences on what it measures and why it's here]
---
[Repeat for each question]
---
## Analysis notes
[Brief guidance on how to slice/interpret the data — e.g., "Compare Q3 responses by role (Q1) to see if the pain is role-specific", "Treat Q5 open text as a follow-up interview queue"]
The "Why this question" rationale is important — it helps PMs and designers trust the survey and defend it to stakeholders.
Tips for Great Question specifically
- GQ supports logic branching — note any skip logic in the question rationale
- Keep survey names descriptive for the repository (e.g., "Feature Validation: Bulk Export — May 2025" not "Survey 3")
- Tag studies with the product area for searchability