To write a good survey question, pick one specific behavior or opinion, ask about it in plain words a 12 year old could read, and offer a balanced set of answer options that cover every reasonable response. That is the short version of how to write survey questions that produce data you can actually act on. The longer version follows the APP formula: agree on the single decision the question must inform, promise the respondent it will take less than a minute, and preview what kind of answer you expect. Get those three right and your response rates climb, your data stops contradicting itself, and your team stops arguing about what the numbers mean. If you want a faster path, sign up for PollPe Survey Builder and let Aria draft a first pass in two minutes.
Key takeaways
- A good survey question covers one idea, uses plain language, and offers a balanced scale. Vague time frames and double-barreled wording are the two fastest ways to corrupt your data.
- Match the question type to the decision you need to make. A 1 to 10 rating belongs in tracking studies, while a ranking or MaxDiff exercise belongs in prioritization work. Our guide to survey question types walks through each format.
- Pilot test every survey with five real respondents before you send it. You will catch confusing wording, missing options, and ordering bias in 20 minutes.
- Sequence matters. Sensitive questions go near the end, easy questions go at the start, and demographics go last. This single change can lift completion rates by 10 to 15 percent, as documented in our notes on improving survey response rates.
- Avoid leading and loaded language. Words like "amazing," "frustrating," or "powerful" prime the answer before the respondent has thought about it.
- Use a 1 to 10 scale only when you actually need that resolution. For most attitude questions, a 5 point Likert scale is easier to answer and easier to chart.
What makes a good survey question
A good survey question has five properties, and missing any one of them weakens the response. First, clarity: the respondent understands what you are asking on the first read, with no rereads required. Second, neutrality: the wording does not push toward a particular answer, and the scale is symmetric around a true middle.
Third, single-idea: the question asks about exactly one thing, never two glued together with an "and." Fourth, a concrete time frame: "in the last 30 days" beats "recently" every time, because respondents anchor on different definitions of "recently." Fifth, an answerable scale: the options cover the full range of plausible answers, including "not applicable" or "prefer not to say" where relevant.
When you write a survey question that satisfies all five, you remove most of the noise that makes survey data look messy in analysis. The American Association for Public Opinion Research (AAPOR) has published similar guidance for decades, and the principles have not changed. What has changed is how fast you can apply them, which is where modern tooling earns its keep.
7 rules for writing survey questions
These are the rules I come back to every time I draft a questionnaire, whether it is a 5 question CSAT pulse or a 40 question pricing study. Each rule has a bad and good example so you can see the difference on the page.
1. Use plain language
Cut jargon, acronyms, and internal product names. If your respondent has not sat in your standup, they do not know what "the new flow" or "the v2 dashboard" refers to. Write the question the way you would ask a friend at dinner.
Bad: How satisfied are you with our recently launched omnichannel engagement platform? Good: How satisfied are you with the way we send you updates by email and text?
The good version names the actual behavior. The bad version requires the respondent to translate your marketing copy into something concrete, and many will just pick the middle option to move on.
2. Avoid double-barreled questions
A double-barreled question asks about two things at once and forces one answer. The classic example is "How satisfied are you with the price and quality of our product?" If the respondent loves the quality but hates the price, there is no honest answer.
Bad: How satisfied are you with the speed and accuracy of our support team? Good: How satisfied are you with the speed of our support team's first reply? (Ask accuracy as a separate question.)
Split every "and" question into two questions. You will get cleaner signal and the survey will still feel short if you have trimmed the rest.
3. Avoid leading and loaded language
Leading language plants an answer. Loaded language attaches an emotion. Both are common in product surveys written by the team that built the feature, because they cannot help cheering for their own work.
Bad: How much do you love our new AI assistant? Good: How would you rate our new AI assistant?
If you must use adjectives, balance them. "How easy or difficult was checkout?" is fair. "How easy was checkout?" assumes the answer.
4. Balance the scale
A balanced scale has the same number of positive and negative options around a neutral midpoint. A five point scale (Very dissatisfied, Dissatisfied, Neutral, Satisfied, Very satisfied) is balanced. A four point scale that drops "Neutral" forces a choice and is sometimes useful, but only if you have a reason.
Bad: Rate our onboarding: Excellent, Very Good, Good, Average, Poor. (Three positives, one neutral, one negative.) Good: Rate our onboarding: Excellent, Good, Neutral, Poor, Terrible.
Unbalanced scales inflate scores. Pew Research has shown that even small asymmetries can shift means by 0.3 to 0.5 points on a 5 point scale, which is enough to make a flat trend look like a win.
5. Match the question type to the decision
Pick the format that fits what you will do with the answer. If you need to track a single metric over time, use a 1 to 10 rating or a Likert. If you need to prioritize features, use a ranking or a MaxDiff. If you need price sensitivity, use Van Westendorp.
Bad: An open text box asking "What features should we build next?" Good: A MaxDiff exercise asking respondents to pick the most and least important feature from sets of four, repeated six times.
The open text box generates a wishlist nobody can prioritize. The MaxDiff produces a ranked list with utility scores. Our breakdown of the 1 to 10 rating scale covers when that specific format earns its place.
6. Pilot test with at least five people
Send the draft to five people who match your target respondent before you launch. Watch them fill it out, or ask them to think aloud over a call. You will find confusing wording, missing answer options, and questions that take twice as long as you expected.
Bad: Launch to 10,000 customers and hope the data makes sense. Good: Pilot with 5 customers on Monday, fix the three problems they surface, then launch on Wednesday.
Two days of piloting saves two weeks of "why does this number look weird" in analysis.
7. Sequence matters
Put easy, engaging questions first. Put sensitive or demographic questions last. Group related questions together so the respondent does not have to context switch every screen. And never start with a screening question that feels like an interrogation.
Bad: Q1: What is your annual household income? Q2: How satisfied are you with our app? Good: Q1: How often do you use our app? Q2: How satisfied are you with our app? Q15: What is your annual household income? (Optional.)
The order you ask in changes the answers you get. This is called order effects in the methodology literature, and SurveyMonkey's own research team has published examples where flipping two questions moved CSAT scores by 8 points.
Question wording examples by use case
These are starting points. Adapt the wording to your product, your customer, and the decision you are trying to make. Every example below is one I have used or seen used in production surveys.
Customer satisfaction
- How satisfied are you with [product] overall? (1 = Very dissatisfied, 5 = Very satisfied)
- Thinking about your last 30 days using [product], how well did it meet your needs?
- What is the one thing we could change to make [product] more useful for you? (Open text, optional.)
The first two give you trackable numbers. The third gives you the qualitative color that explains movements in the numbers.
Product feedback
- How often do you use [specific feature] in a typical week? (Never, 1 to 2 times, 3 to 5 times, 6 or more times)
- When you used [specific feature] most recently, did it do what you expected? (Yes / No / I have not used it)
- If you could change one thing about [specific feature], what would it be? (Open text.)
Name the feature. Do not ask about "the product" if you want feedback on a specific surface.
Employee engagement
- I would recommend [company] as a place to work. (1 = Strongly disagree, 5 = Strongly agree)
- In the last 7 days, I received recognition for good work. (Yes / No)
- I have a clear understanding of what is expected of me at work. (1 to 5 agreement scale)
These three are adapted from the Gallup Q12 family and they work because each one names a specific, recent behavior.
Pricing research (Van Westendorp)
The Van Westendorp Price Sensitivity Meter uses four questions to find an acceptable price range. Ask them in this order, after describing the product clearly:
- At what price would you consider [product] to be so expensive that you would not buy it?
- At what price would you consider [product] to be priced so low that you would feel the quality could not be very good?
- At what price would you consider [product] starting to get expensive, so that it is not out of the question, but you would have to give some thought to buying it?
- At what price would you consider [product] to be a bargain, a great buy for the money?
Plot the four cumulative curves and read off the acceptable price range at the intersections. This single technique replaces hours of pricing debate.
NPS follow-up
The NPS score itself is the easy part. The follow-up is where the value is.
- On a scale of 0 to 10, how likely are you to recommend [product] to a friend or colleague?
- What is the single most important reason for the score you gave? (Open text.)
- What is one thing we could do to earn a higher score from you? (Open text, shown only to scores of 0 to 8.)
The conditional third question is the one most teams skip and the one that produces the most actionable feedback.
Common mistakes that ruin survey data
Most bad survey data comes from a small number of repeat offenders. Watch for these:
- Jargon and internal product names the respondent does not recognize.
- Vague time frames like "recently," "often," or "in the past." Use specific windows.
- Social desirability bias, where respondents answer the way they think you want. Reduce it by promising anonymity and avoiding loaded adjectives.
- Hypothetical framing ("Would you use a feature that..."). People are terrible at predicting their own future behavior. Ask about past behavior instead.
- Asking about things the respondent cannot remember. Nobody knows how many times they opened your app last month.
- Acquiescence bias, where respondents agree by default. Mix positive and negative statements to break the pattern.
- Forced-choice questions with no "not applicable" option, which pushes respondents to invent answers.
- Surveys longer than 10 minutes. Completion rates fall off a cliff after the 7 minute mark.
- Showing the brand name on every question, which primes positive bias toward the sponsor.
If you can eliminate these nine mistakes from your next survey, your data quality will jump before you change anything else. For a deeper library of vetted wording, our survey research question examples collection is a good next stop.
How PollPe Survey Builder helps you write better surveys
Most of the work above is judgment, and judgment is what AI is now good at. PollPe Survey Builder ships with an AI agent called Aria that turns a one sentence research goal into a complete, balanced question set in about two minutes. You tell Aria "I want to understand why trial users are not converting to paid," and it returns a 12 to 15 question draft with the right mix of closed and open questions, sensible scales, and a logical sequence. You edit, you do not start from a blank page.
Beyond Aria, the builder includes more than 20 question types, including Matrix, MaxDiff, Van Westendorp, Ranking, and Heatmap, so you can match the format to the decision instead of forcing every question into a 1 to 5 scale. The free tier includes unlimited responses, which matters because Typeform's free plan caps you at 10 responses per month and that is not enough to pilot anything real. On the Business plan you also get 15 languages, including Hindi, Telugu, and Tamil, which is the difference between running a survey in India and translating your survey by hand into a spreadsheet. See PollPe pricing for the full breakdown.
FAQ
How many questions should a survey have?
Aim for 5 to 12 questions for most use cases. Completion rates drop sharply past the 7 minute mark, which usually corresponds to 15 to 20 questions depending on complexity. Tracking surveys (CSAT, NPS) should be 3 to 5 questions. Discovery surveys can run to 20 if every question earns its place. If you have more than 20, split the survey into two waves or use logic to show each respondent only the questions that apply to them.
What is a double-barreled question?
A double-barreled question asks about two things at once and forces one answer. "How satisfied are you with the price and quality?" is the textbook example. If a respondent loves the quality but hates the price, neither a high nor a low rating is honest. Fix it by splitting into two questions, one for each attribute, and asking them on the same scale so you can compare.
How do you avoid leading questions?
Strip out adjectives that carry emotion, balance positive and negative phrasing, and offer a neutral middle option. Replace "How much do you love..." with "How would you rate..." Replace "How easy was..." with "How easy or difficult was..." Read every question out loud and ask yourself which answer it is steering toward. If you can hear the steering, rewrite it.
Should survey questions be open or closed?
Use closed questions for anything you want to count, chart, or track over time. Use open questions sparingly for the "why" behind a number, and always make them optional. A good ratio is 80 percent closed to 20 percent open. Open questions take 4 to 6 times longer to answer than closed ones, and they take even longer to analyze unless you have AI coding the responses.
Can PollPe Survey Builder write survey questions for me?
Yes. Aria, the built in AI agent, drafts a complete question set from a one sentence research goal in about two minutes. You describe the decision you need to make, Aria proposes the questions, scales, and sequence, and you edit from there. It is faster than starting from a template and produces better drafts than most first time questionnaire writers, because it has been trained on the rules in this post.
Conclusion
If you remember one thing from this post, remember that learning how to write survey questions is mostly about restraint. Cut the jargon, split the double-barreled questions, balance the scale, and pilot with five real people before you launch. Those four habits will lift your data quality more than any new tool or fancy analysis. The tooling matters too, because the faster you can draft, pilot, and iterate, the more research your team can run in a quarter. PollPe Survey Builder was built for that loop, with Aria drafting the first pass, 20 plus question types covering every decision, and unlimited responses on the free tier so you can pilot without watching a counter. Sign up free or see pollpe.com/pricing when you are ready to scale.
