Table of Contents
Good data starts with good questions. If you want to learn how to write survey questions that actually measure what you care about, the craft matters more than the tool you build them in. A clumsy question quietly biases every answer, and no amount of clever analysis fixes it afterward.
The good news: most survey mistakes come from a short list of predictable traps. Once you can spot them, your questionnaires get shorter, clearer, and far more reliable.
This guide walks through the whole craft. You will learn how to tie every question to a research goal, pick the right question type, dodge the classic wording traps, design answer options that behave, order questions so they do not bias each other, and pilot the whole thing before you launch.
Start with your research variables, not your questions
Before you write a single question, write down what you are actually trying to learn. Each item on your survey should map to a specific variable or decision. If a question does not connect to something you plan to analyze, cut it.
This one habit prevents the most common problem in amateur surveys: the bloated questionnaire full of "nice to know" items that nobody ever looks at. Every extra question costs you completions and attention. Spend that budget only on questions that earn their place.
The one-sentence test
For every question, finish this sentence: "I am asking this because I need to know ___ so I can ___." If you cannot fill both blanks, the question is probably filler. Cut it or rewrite it.
Once your list of variables is clear, each variable tells you what kind of question you need. A yes/no fact needs a different format than an attitude or a preference ranking. That is where question types come in.
How to write survey questions: the main question types
You do not need dozens of formats. Four cover the vast majority of research surveys, and knowing when to reach for each keeps your data clean.
Multiple choice (single or multiple answer)
Use these for facts, categories, and clear either/or choices: which product someone uses, which country they live in, which features they want. Keep options mutually exclusive for single-answer questions, and label multi-answer questions clearly ("Select all that apply").
Closed questions like these are easy to answer and easy to analyze. They should be your default whenever the set of possible answers is knowable in advance.
Likert scale questions
Likert scales measure attitudes and degrees of agreement, satisfaction, or frequency. A statement plus a symmetric scale ("Strongly disagree" to "Strongly agree") lets you quantify something fuzzy like opinion. They are the workhorse of attitude research.
Two rules keep them honest. First, use a balanced scale with an equal number of positive and negative options. Second, keep the wording of the statement neutral so the scale, not the sentence, carries the measurement. If you build a lot of these, our Likert scale generator produces clean, balanced scales in seconds.
Ranking questions
Ranking is useful when you need to know relative priority: which three features matter most, in order. Ranking forces trade-offs that rating scales hide, because respondents cannot call everything "very important."
Use ranking sparingly. Asking someone to rank more than five or six items gets tedious and the lower ranks turn to noise. For long lists, ask people to pick their top three instead of ordering all of them.
Open-ended questions
Open text is where you learn the things you did not think to ask. Use it for the "why" behind a rating, or to surface language and reasons you can later turn into closed questions. The trade-off is real: open answers are slow to write, so they lower completion, and they take real effort to code and analyze.
Use one or two well-placed open-ended questions, not ten. A single "What is the main reason for your score?" after a rating often delivers more insight than a page of free text.
The wording traps that quietly ruin your data
Most bad questions are grammatically fine. They fail because of how they nudge the respondent. Here are the traps to watch for, with fixes.
Wording traps to avoid
- Leading questions. Wording that pushes toward one answer. "How much did you enjoy our excellent new checkout?" assumes enjoyment. Ask neutrally: "How would you rate our new checkout?"
- Double-barreled questions. Two questions crammed into one. "Was the staff friendly and knowledgeable?" cannot be answered by someone who found them friendly but not knowledgeable. Split it into two questions.
- Loaded assumptions. Questions that presume something is true. "How often do you use our mobile app?" assumes the person uses it at all. Add a filter question first, or include a "I do not use it" option.
- Vague quantifiers. Words like "often," "regularly," or "a lot" mean different things to different people. Replace them with concrete ranges: "How many times in the last 7 days?"
- Jargon and insider language. Acronyms and technical terms that respondents may not share. Write for the least specialist person in your sample, or define the term inline.
Two of these deserve a full before-and-after, because they are the most common and the easiest to miss in your own writing.
Fixing a leading question
Before: "How helpful was our friendly support team in solving your problem quickly?"
This packs in three assumptions (friendly, helpful, quick) and pushes for praise.
After: "How would you rate the support you received?" followed by a balanced scale from "Very poor" to "Very good." Then ask an open follow-up: "What is the main reason for your rating?"
Fixing a double-barreled question
Before: "How satisfied are you with the price and quality of the product?"
Price and quality are different things, and a respondent may feel very differently about each.
After: Split into two clean items. "How satisfied are you with the product's quality?" and "How satisfied are you with the price?" Each gets its own scale, and now you can tell which one is the problem.
Designing answer options that behave
The question is only half the item. The answer options carry just as much risk.
Keep scales balanced, with the same number of positive and negative points and a clear midpoint if you want one. An unbalanced scale ("Good, Very good, Excellent") manufactures positive results.
Make options mutually exclusive and collectively exhaustive. For a single-answer age question, do not let "18-25" and "25-35" overlap on 25, and make sure every plausible answer has a home. Add an "Other" field when you cannot list every case.
When to include an escape hatch
For sensitive or not-applicable questions, give people a "Prefer not to say" or "Not applicable" option. It reduces random guessing and drop-off. But do not add it everywhere: on simple factual questions it just invites lazy non-answers.
One more small thing with a big effect: label every point on a scale where you can, not just the ends. Fully labeled scales are easier to answer consistently and reduce the guessing that muddies your results.
Question order and grouping
The order of your questions changes the answers you get. This is not a minor detail.
Earlier questions prime later ones. If you ask about a product's problems and then ask for an overall rating, the rating drops, because you just reminded people of everything wrong. Ask the general question first, then drill into specifics.
Group related questions together so respondents are not mentally switching topics on every screen. Put easy, engaging questions near the start to build momentum, and save sensitive questions (income, personal details) for the end, once trust is established. If people are going to abandon over a touchy question, you would rather they abandon after giving you the rest.
Keep it short
Completion rate falls as surveys get longer. The exact drop-off depends on your audience and topic, but the direction is reliable: every extra minute costs you respondents, and the ones who push through a long survey often speed up and answer carelessly toward the end.
Aim for the shortest survey that still answers your research questions. A focused 5-minute survey almost always beats a sprawling 20-minute one, both in completion and in data quality. If you are collecting responses through a survey exchange, a short survey is also simply kinder to the people answering it, and kindness gets rewarded with better answers.
How to write survey questions people finish: pilot test first
Never send a survey to your full sample without testing it first. Give it to 5-10 people who resemble your real respondents and watch what happens.
A pilot catches the things you cannot see in your own writing: a question everyone interprets differently, an answer option that is missing, a scale that confuses people, a survey that takes twice as long as you thought. Ask your testers to think aloud as they answer, and ask afterward whether anything was unclear or annoying.
Fixing a broken question after you have collected 200 responses is expensive and sometimes impossible. Fixing it after a 10-person pilot is free. This step pays for itself every single time.
That is the whole craft of how to write survey questions: tie each one to a variable, pick the right type, dodge the wording traps, and pilot before you launch. Once your questions are clean, the rest of the process gets easier. For more on turning good questions into good data, see how to collect high-quality data in survey research. To make sure people actually finish, read how to increase survey response rates. And before you launch, check how many responses you need with our survey sample size guide.
Frequently asked questions
How many questions should a survey have?
As few as possible while still answering your research questions. There is no magic number, but a survey that takes 5 minutes or less will almost always get more completions than a long one. Map each question to a variable you plan to analyze and cut anything that does not earn its place.
What is a double-barreled question?
A double-barreled question asks about two things at once but only allows one answer, like "Was the service fast and friendly?" A respondent who found it fast but unfriendly cannot answer honestly. The fix is always to split it into two separate questions.
When should I use open-ended questions?
Use open-ended questions sparingly, usually to understand the reason behind a closed answer or to discover language and issues you did not anticipate. They lower completion rates and take real effort to analyze, so one or two well-placed open questions beat a page of free-text boxes.
How do I know if my survey questions are any good?
Pilot test them. Give the survey to 5-10 people who resemble your target respondents, ask them to think aloud, and note every question that confuses them or that they interpret differently than you intended. A short pilot catches problems that are invisible to the person who wrote the questions.