Picking the right question type is half of survey design — the other half is wording it well (see our guide to writing better questions). There are more types of survey questions than most people reach for by default, and picking the wrong one is one of the most common ways a survey quietly produces bad data. Here's a practical rundown of every major type, what each is actually good for, an example of how to word it, and where people misuse it.
Multiple choice
What it's for: when you already know the likely set of answers and want fast, clean, easy-to-analyze data.
Example: "Which of these features do you use most?" — Dashboard, Reports, Integrations, Mobile app, Other.
Watch out for: forcing respondents into an option that doesn't fit. Always consider whether you need an "Other" free-text fallback, and don't offer so many options that the question becomes a wall of text. If someone can only pick one answer but the real world allows more than one to be true, you're not looking at a multiple-choice question — you need a checkbox question instead (below).
Checkbox (select multiple)
What it's for: the same situation as multiple choice, except more than one answer can genuinely be true at once. "Which of these features have you used?" is a checkbox question; "Which of these features do you use most?" is multiple choice — the wording itself signals whether one answer or several are expected.
Example: "Which of these channels have you used to contact support?" — Email, Live chat, Phone, Help center articles.
Watch out for: treating checkbox and multiple choice as interchangeable in your own head while writing the survey. Respondents pick up on the mismatch even if they can't name it — a "select all that apply" instruction on a question that only lets them click one option is a confusing, broken experience.
Star rating
What it's for: quick, familiar, low-effort satisfaction snapshots — most people have rated something with stars before, so there's no learning curve. Works well right after a specific interaction (a purchase, a support ticket).
Example: "How would you rate your experience with this order?" — 1 to 5 stars.
Watch out for: stars alone don't tell you why. Pair a star rating with a short optional follow-up ("What's the main reason for your rating?") if you want anything actionable.
Likert scale
What it's for: measuring agreement or intensity — "strongly disagree" to "strongly agree" — across a set of statements. This is the workhorse for attitude and satisfaction measurement (employee engagement, product sentiment) because it lets you compare responses across many statements on the same scale. We go deeper on this one in our Likert scale guide, including how many points to use — see our comparison of 5-point vs. 7-point vs. 10-point scales if you're unsure.
Example: "My manager gives me clear direction" — Strongly disagree, Disagree, Neutral, Agree, Strongly agree.
Watch out for: unbalanced scales (more positive options than negative) and unlabeled midpoints — both quietly bias results toward agreement.
NPS scale (0–10)
What it's for: one specific, standardized question — "how likely are you to recommend us?" — used to track overall loyalty over time. It's a relationship metric, not a per-interaction one; don't ask it after every support ticket.
Example: "On a scale of 0–10, how likely are you to recommend us to a friend or colleague?"
Watch out for: using NPS as your only metric. It tells you that someone is unhappy, not why — always pair it with a reason-for-score follow-up. We cover exactly when NPS beats other loyalty/satisfaction metrics in NPS vs. CSAT vs. CES.
Slider / numeric scale
What it's for: a continuous numeric answer where a discrete scale would feel artificially chunky — budgets, quantities, confidence levels expressed as a percentage. A slider is a different tool from a Likert scale even though both produce a number: Likert measures agreement with a statement across a small number of labeled steps, while a slider captures a genuinely continuous value.
Example: "How confident are you that you'll renew next year?" — a 0–100% slider.
Watch out for: using a slider where a labeled scale would actually be clearer. Without visible tick marks or labels, respondents often can't tell what a "70" means relative to a "75," which adds noise instead of precision — the opposite of what a numeric format is supposed to buy you.
Matrix / grid
What it's for: rating several related items on the same scale in one compact view, without repeating the same question five separate times. This is distinct from ranking (below) — a matrix asks for an independent rating per row, not a relative order across rows.
Example: "Rate each of the following on a 1–5 scale: Product quality, Customer support, Pricing, Ease of use."
Watch out for: long grids. A matrix with more than about six or seven rows starts to produce "straight-lining" — respondents clicking the same column down the whole grid just to finish faster, which quietly destroys the data's validity.
Ranking
What it's for: when relative priority matters more than absolute scores — "rank these five features by importance" tells you something a set of five independent ratings won't, because people rate things they don't actually prioritize highly as "important" when asked in isolation. Ranking forces a real trade-off; a matrix doesn't.
Example: "Drag to rank these features in order of importance to you: Speed, Price, Support, Integrations, Design."
Watch out for: long lists. Ranking more than five or six items reliably gets harder and less reliable the longer the list gets — respondents start guessing at the order of the items in the middle they don't feel strongly about.
Open text / textarea
What it's for: anything you can't anticipate. Open text is the only question type that surfaces answers outside your assumptions, which makes it essential for early-stage research and root-cause questions.
Example: "What's the main reason you're canceling?"
Watch out for: overusing it. Free text takes longer to answer and longer to analyze, and it has the highest abandonment cost of any question type. Reserve it for the one or two questions per survey where you genuinely can't predict the answer, and keep it optional where you can.
Yes/no
What it's for: binary facts and screening questions, often used to branch respondents into different follow-up paths.
Example: "Have you used our product in the last 30 days?"
Watch out for: using yes/no where a scale would actually capture more signal. "Are you satisfied?" throws away everything between "barely satisfied" and "extremely satisfied" that a 5-point scale would have kept.
Date / time
What it's for: collecting a specific date, time, or combined date-and-time value where the exact moment matters for follow-up or scheduling — event registrations, appointment preferences, "when did this happen" questions.
Example: "What date did you first contact support about this issue?"
Watch out for: asking for precision you don't need. If you only care about the month, don't force a full date picker — the extra friction isn't worth the resolution you're not going to use.
Image / media choice
What it's for: letting respondents pick between visual options where words would be a poor substitute — logo preference, product photo A/B testing, design mockup feedback.
Example: "Which packaging design do you prefer?" — shown as a set of selectable images rather than text labels.
Watch out for: image order bias. Respondents tend to favor whichever option they see first; randomize image order when the images themselves are the thing being tested, not just illustrative decoration.
File upload
What it's for: collecting an actual document, photo, or attachment rather than a description of one — receipts for an expense-report survey, screenshots for a bug report, portfolio samples for an application form.
Example: "Upload a photo of the damaged item."
Watch out for: making file upload required on a general-audience survey. It's the single highest-friction question type — respondents on mobile in particular will often abandon rather than dig up a file — so reserve it for surveys where the upload is the actual point.
Hidden fields
What it's for: capturing context the respondent never sees or types themselves — a UTM parameter, a referral code, an account ID passed in via the survey link — so you can segment responses later without adding a visible question and burning respondent attention on it.
Example: a utm_source value silently captured from the URL a respondent arrived from, later usable to filter "did NPS scores differ between the email campaign and the in-app link."
Watch out for: trying to use a hidden field for anything the respondent needs to see or confirm — by definition, they never will.
Understanding response types (the data format, not the question format)
"Question type" and "response type" get used interchangeably, but it's worth separating them once you get to analysis: the response type is the shape of the data a question produces, and several different question types can produce the same response type.
| Response type | What it means | Produced by |
|---|---|---|
| Categorical (nominal) | Unordered named categories — no category is "more" than another | Multiple choice, checkbox, yes/no |
| Ordinal | Ordered categories, but the gaps between them aren't guaranteed equal | Likert scale, star rating, ranking |
| Numeric (interval/ratio) | A true number where the gaps are meaningful and consistent | NPS, slider, matrix scores |
| Open text | Unstructured natural language | Open text / textarea |
| Media | A file, image, or attachment rather than typed data | File upload, image choice, signature |
This matters because the response type determines what analysis is even valid — averaging ordinal data (like Likert scores) is a common, generally-accepted approximation, but averaging genuinely categorical data (like which support channel someone picked) doesn't mean anything at all. Know which type of response you're collecting before you decide how you're going to summarize it.
Comparison table
| Question type | Best for | Avoid when | Example use case |
|---|---|---|---|
| Multiple choice | A known, finite set of answers, pick one | The real answer could be more than one thing | Feature usage |
| Checkbox | A known set of answers, pick any that apply | Only one answer is ever true | Channels used |
| Star rating | Fast satisfaction snapshot after an interaction | You need to know why, not just how much | Post-purchase rating |
| Likert scale | Agreement/intensity across many statements | You only have one statement to ask | Employee engagement |
| NPS | Overall loyalty, tracked over time | You want feedback on one specific interaction | Quarterly relationship check |
| Slider | A continuous numeric value | A labeled discrete scale would be clearer | Confidence percentage |
| Matrix | Rating several related items on one scale | The list is long (6+ rows) | Rating multiple product aspects |
| Ranking | Relative priority between options | The list is long (6+ items) | Feature prioritization |
| Open text | Anything you can't anticipate | You need clean, comparable data at scale | Reason for canceling |
| Yes/no | Binary facts, branching logic | A scale would capture more signal | Recent usage check |
| Date/time | A specific moment matters | You only need rough recency | Event registration |
| Image choice | Visual preference | Text would communicate the same choice fine | Design A/B test |
| File upload | An actual document/photo is the point | The survey is meant to be low-friction | Expense receipts |
| Hidden field | Silent context capture (UTM, referral code) | The respondent needs to see or confirm it | Campaign attribution |
Putting it together
Most well-designed surveys mix two or three types rather than relying on one throughout: a scale question for the trackable number, one open-text follow-up for the "why," and multiple choice or checkbox for anything with a known, finite set of answers. As a rough guideline, most effective surveys use somewhere between three and eight questions total, drawn from two or three different types — a survey that's entirely one question type (fifteen Likert statements in a row, say) tends to fatigue respondents faster than the same content mixed with a lighter format in between.
FAQ
What are the main types of survey questions? The most common are multiple choice, checkbox, star rating, Likert scale, NPS, slider/numeric, matrix, ranking, open text, yes/no, date/time, image choice, file upload, and hidden fields — each suited to a different combination of how many answers are possible and how precise the data needs to be.
When should I use open-ended vs. closed-ended questions? Use closed-ended types (multiple choice, scales, yes/no) when you already know the realistic range of answers and want data you can compare and count at scale. Use open-ended text when you genuinely don't know what the answer will be, or when the "why" behind a number matters more than the number itself — early-stage research and root-cause questions are the classic case.
How many question types should one survey use? Two or three is typical for a focused survey. Mixing formats — one trackable scale question, one open-text follow-up, one or two closed-ended screening questions — usually produces better completion rates than either an all-scale or all-open-text survey, because it varies the cognitive effort required question to question instead of demanding the same kind of thinking repeatedly.
What's the difference between a matrix question and a ranking question? A matrix asks for an independent rating of each item on the same scale (so two items can both score a 5). A ranking question forces a relative order across the items (so only one item can be first). Use a matrix when absolute scores matter; use ranking when you specifically need to know what beats what.
Surveyee supports every type above, and the template gallery has pre-built combinations for common goals — Customer Satisfaction pairs star rating with NPS and open text, Employee Feedback leans on Likert scales, and Product Feedback mixes star rating, multiple choice, and NPS in one flow.