A survey is only as good as its worst question. One ambiguous item does not just waste its own data - it erodes the respondent's effort on everything that follows. These seven rules, each with a before-and-after, cover the mistakes we see most often in studies fielded on the platform.
Rule 1: One idea per question
The double-barrelled question is the most common flaw in first drafts. If a respondent agrees with half the question, no answer they give is true.
| Question | |
| ❌ Before | Was the delivery fast and was the packaging in good condition? |
| ✅ After | How would you rate the delivery speed? / How would you rate the condition of the packaging? |
Rule 2: Don't lead the witness
Any question that hints at the desired answer will get it - and the data will confirm whatever you already believed, which is worse than no data.
| Question | |
| ❌ Before | How much do you love our new faster checkout? |
| ✅ After | How satisfied or dissatisfied were you with the checkout process? |
Rule 3: Anchor time and quantity
"Often", "regularly", and "recently" mean different things to different people. Replace adverbs with numbers and windows.
| Question | |
| ❌ Before | Do you often shop online? |
| ✅ After | In the past 30 days, how many times did you make a purchase online? |
Rule 4: Make the options exhaustive and exclusive
Every respondent must have exactly one honest place to land. Overlapping ranges (18-25, 25-35) and missing options (no "never", no "other") force people to lie by structure. Always test: can someone answer this truthfully with one selection?
Rule 5: Match the scale to the judgment
Use frequency scales for behavior, agreement scales for attitudes, satisfaction scales for experience - never agreement scales for everything. And choose scale length deliberately: our Likert scale guide covers when 5 points beats 7.
Rule 6: Write for a tired reader on a phone
- •Questions under 20 words; options under 10
- •No negatives ("How much do you not disagree...") - and never double negatives
- •No jargon or internal product names the respondent may not share
- •Front-load the ask: "How satisfied were you with X?" beats "Regarding X, taking everything into account, how..."
Rule 7: Pilot before you pay
Field the survey to 10-20 people before the full launch. Look at the open-text answers of your pilot: if respondents are answering a different question than the one you meant, the wording failed - and it is far cheaper to learn that at n=15 than at n=1,500.
“If a question can be misread, it will be - at scale, and in your most expensive study.”
- Research team maxim
Ready to apply this? Walk through the first-time researcher's checklist before you hit Publish.