Every questionnaire is built from about eight question types, and most survey problems trace back to using the wrong one. Here's each type with a real example, when it earns its place, and the wording mistakes that quietly bias answers.
By the Wavefield Research team · Published Aug 5, 2026
Every question type belongs to one of two families. Closed-ended questions give respondents fixed options — click, done — and produce structured data you can chart, crosstab, and significance-test the moment fieldwork closes. Open-ended questions collect free text in the respondent's own words, which is where the answers you didn't anticipate live.
The trade is precision versus discovery. Closed questions can only confirm or deny what you thought to ask; open questions surface what you missed but cost respondent effort and need coding before they become numbers. Professional questionnaires are mostly closed-ended, with a few deliberate open-ends doing the discovery work — not fifty-fifty.
The checkout process was easy to complete.
Five labeled points, one neutral midpoint, every point named — the conventions that make scale results comparable across studies.
These seven cover nearly every closed question in professional practice. Specialist designs exist beyond them — semantic differentials, constant-sum allocations, MaxDiff — but they are variations on the same choices: how many options, how many selections, and whether the answer is a category, a scale point, or an order.
Which of these brands did you purchase most recently? (Select one)
The workhorse: one question, one answer from a fixed list. Use it whenever the options are mutually exclusive — brand purchased, primary reason, age band. Keep lists exhaustive with an Other/None option, and randomize option order so position doesn't drive the answer.
Which of the following have you done in the past 30 days? (Select all that apply)
The checkbox variant for non-exclusive options. Analysis reports each option as its own percentage (they won't sum to 100). Beware recall limits: past roughly ten options, respondents satisfice and tick whatever they recognize first.
Have you purchased from us in the past six months? Yes / No
Exactly two options — Yes/No, True/False, Aware/Not aware. Dichotomous questions are the backbone of screeners and branching logic because they split cleanly, but they force nuance out: use them to route people, not to measure attitudes.
“The product is worth what I paid for it.” — Strongly disagree to Strongly agree
A statement rated on a 5- or 7-point agreement scale. The convention set: label every point, keep an odd number so there's a genuine midpoint, and keep the direction consistent across the survey. Reported as top-two-box shares or means.
How likely are you to recommend us to a friend or colleague? (0–10)
Numeric scales for satisfaction, likelihood, or importance. NPS is the famous special case: 0–10 likelihood to recommend, with 9–10 promoters, 7–8 passives, 0–6 detractors, and the score being promoters minus detractors.
Rank these five features from most to least important to you.
Forces trade-offs that rating scales let respondents dodge (where everything scores “very important”). The cost is effort: ranking is hard on phones and unreliable past six or so items. For long lists, MaxDiff designs do the same job better.
Rate each statement below on the same agree–disagree scale.
Several Likert items sharing one scale — compact and fast, but the top source of straightlining, where a respondent picks the same column all the way down. Keep grids under about seven rows and check speeders and straightliners before trusting the data.
An open-ended question is a text box and a prompt. The good ones are specific enough to answer in one sentence and open enough that you couldn't have listed the answers yourself:
Notice the pattern: each one probes a why or a what's missing that no option list could anticipate. That third example — the follow-up probe after a rating — is the highest-value open-end in most questionnaires, because it attaches an explanation to a number you already have.
The discipline is restraint. Every open-end adds typing effort that measurably raises abandonment, so professional questionnaires carry one to three, placed late, and almost never required. The classic workflow is qualitative, then quantitative: run open-ends to discover the categories, then convert them to closed questions in the next wave. And the traditional cost of open-ends — days of manual coding — is now automated: AI-assisted coding groups responses into themes you can count and crosstab like any closed question.
The selection rule is to work backwards from the analysis: a chart of shares wants single-select, a driver analysis wants scales, a forced priority wants ranking, a quote for the report wants an open-end. Then the type has to be worded cleanly — these four mistakes account for most biased survey data:
“Was the staff friendly and knowledgeable?” asks two things with one answer box. A respondent who found them friendly but clueless has no honest option. Split it — one question per attribute.
“How much did you enjoy our award-winning service?” presumes enjoyment and telegraphs the wanted answer. Neutral framing — “How would you rate the service?” — is the entire difference between measurement and flattery.
Age bands of 25–35 and 35–45 put a 35-year-old in both; a list with no “None of these” forces false positives. Options must be mutually exclusive and collectively exhaustive.
“Excellent / Very good / Good / Fair” offers three positive points and one negative — the average is inflated before anyone answers. Scales need symmetric positive and negative sides around a real midpoint.
On Wavefield, type selection and wording hygiene are the agent's job: it drafts the questionnaire from your brief using these conventions — balanced scales, split-out attributes, randomized options, open-ends placed late — and you review the result instead of writing from scratch. How that works end to end is covered in survey programming, explained.
Two families. Closed-ended questions offer fixed options — multiple choice (single or multi select), dichotomous (two options), Likert scales, rating scales, ranking, and matrix grids — and produce numbers you can chart and test. Open-ended questions collect free text in the respondent's own words. Most professional questionnaires are mostly closed-ended with a small number of deliberate open-ends.
A question with exactly two possible answers: Yes/No, True/False, Aware/Not aware. Dichotomous questions are ideal for screening (“Do you own a car?”) and branching logic, because the split is unambiguous. Their weakness is the flip side: two options can't capture degree, so attitude and satisfaction measurement belongs to scales instead.
A Likert scale measures agreement with a statement using labeled points (Strongly disagree to Strongly agree). A rating scale asks for a direct numeric judgment — satisfaction from 1 to 5, likelihood from 0 to 10. NPS is a rating scale with a fixed formula on top. In practice the analysis is similar; the choice is about whether a statement or a direct question reads more naturally.
One to three, placed late in the questionnaire, and almost never required. Each open-end adds typing effort that measurably increases abandonment, especially on phones. Use them where free text earns its cost — the “why” behind a rating, or discovery of answers you didn't think to list — and let AI-assisted coding turn the text into countable themes afterward.
Question types are where a questionnaire starts — survey programming covers turning them into a fielded instrument, and how to analyze survey results covers what happens after. Related: sample size calculator
Upload a brief and Wavefield's agent drafts the questionnaire — right types, balanced scales, clean wording — ready for your review in minutes.
Start a project