Most bad surveys do not look obviously bad.
They look normal. A few rating questions. A few dropdowns. Some open text. Maybe a matrix. Then the results come in and everyone treats the numbers as truth.
But survey data is unusually easy to damage. Question wording, order, response options, and layout all change how people answer. This is not an academic footnote, it is the difference between a decision you can trust and one you cannot. If you ask unclear questions, you get unclear data, no matter how clean the chart looks.
Here are the practices that hold up, with the evidence behind them.
Start with the decision, not the topic
Before writing questions, write the decision.
Bad starting point:
We want to ask users about onboarding.
Better:
We need to decide whether to simplify onboarding, add more guidance, or change the activation checklist.
The second version tells you exactly which answers you need, and which questions you can cut. Every question that does not move a decision is a question that costs you completion rate for nothing.
Respect the length the data supports
This is the practice teams ignore most. Vendor data is remarkably consistent:
- Drop-off jumps roughly 17% once a survey passes ~12 questions or ~5 minutes.
- Abandonment spikes noticeably beyond 7–8 minutes, and surveys past ~12 minutes (or ~9 minutes on mobile) see substantial break-off.
- In one randomized trial, short surveys hit a 64% response rate versus 51% for long ones among uncompensated respondents.
The practical target: 7–10 focused questions, under 5 minutes. If you need more, use logic to skip irrelevant paths so no single respondent answers everything.
One question should ask one thing
Double-barreled questions are everywhere.
Bad:
How satisfied are you with our pricing and support?
If the respondent loves support but hates pricing, there is no honest answer. Split it:
How satisfied are you with our pricing?
How satisfied are you with our support?
It feels obvious written out, yet this mistake shows up constantly.
Neutral wording is underrated
The fastest way to ruin feedback is to ask a question that already contains the answer you want.
Bad:
What did you love about our new dashboard?
Better:
How would you describe your experience with the new dashboard?
The first is a compliment trap. The second gives the respondent room to be honest, including the honesty you need to hear.
Keep scales balanced
If your scale has three positive options and one negative, you are not measuring fairly. Use balanced labels, make the midpoint meaningful, and never hide the negative choices.
Be careful with vague labels, too. "Good/bad" is weaker than the dimension you actually care about: "easy/hard" for product effort, "resolved/not resolved" for support.
Use open text deliberately
Open text is where the best insights live, and the highest effort sits. Pair it with a structured answer so the work has a clear job:
How satisfied are you with onboarding?
What is the main reason for that score?
Now you get a number and the explanation behind it. Random standalone open-ended questions read like homework and get one-word answers.
Design mobile-first
If a survey is painful on mobile, the survey is broken, and the data shows mobile-optimized surveys see roughly 10% higher completion than desktop-first designs. The worst offenders are long grids, tiny tap targets, huge option lists, and required text boxes. Conversational, one-question-at-a-time mobile flows can reach 80–85% completion when done well.
Assume the respondent is on a phone, distracted, and one bad interaction away from leaving. Design for that person.
Use progress cues honestly, and rethink the bar
A counterintuitive finding: progress bars can actually decrease completion, because a slow-moving bar signals "this is long." Textual cues often work better ("Just a few questions left"), and for longer surveys, named stages ("About you → Your workflow → Final thoughts") beat raw percentages.
Put sensitive questions later
Do not open with income, phone number, company revenue, or personal details. Start with easy, relevant questions, build a little trust, and ask sensitive questions only when necessary, with a one-line reason for why you need them.
The practical lesson
Survey design is product design. The respondent is doing unpaid work for you, and your job is to make that work feel clear, respectful, and worth finishing.
That is the default I want Formaly to encourage: shorter surveys, neutral wording, the right question types, and fewer ways to accidentally collect bad data.