The 12 types at a glance
Every survey format in common use is one of the twelve below, or a costume worn by one of them. Picture choice is single choice with images. A slider is a rating scale you drag. A semantic differential is a Likert scale with words at both ends. Learn the twelve and nothing in a survey tool will surprise you.
| Type | What the respondent does | What you get back |
|---|---|---|
| Open-ended | Writes a few sentences | Language, reasons, surprises |
| Short text | Types one line | A fact: name, title, city |
| Single choice | Picks one option | Clean segments that sum to 100% |
| Multiple choice | Picks several options | Share of people per option |
| Dropdown | Opens a long list, picks one | Standardized values you can filter on |
| Likert scale | Agrees or disagrees with a statement | Strength of an attitude |
| Rating scale | Scores something 1 to 5 or 1 to 10 | A number you can track over time |
| Star rating | Taps a star | An average out of five, fast |
| NPS / 0-10 | Answers one 0-10 question | A benchmarkable score from -100 to +100 |
| Ranking | Puts a short list in order | Forced trade-offs between good options |
| Matrix | Applies one scale to several rows | Comparable scores across items |
| Yes / No | Answers a binary | A percentage and a branch |
1. Open-ended (long text)
What it is. A free-text box with no options. It is the only type that can tell you something you did not think to ask.
Use it when you do not yet know the answer space, or when you need the reason behind a score. Do not use it when you already know the five plausible answers, because you are charging people typing for something they could have clicked. Never stack two in a row.
Example: "What almost stopped you from signing up?"
The analysis it gives you: no chart. You read the answers, tag them into themes and count the themes. It also carries the highest abandonment cost of any type, so place it late and leave it optional.
2. Short text
What it is. A single-line field: name, job title, city, order number, website.
Use it when the answer is short, factual and impossible to list in advance. There are millions of job titles. Do not use it when you will group by the answer later. "New York", "NYC" and "new york" are three values in your spreadsheet.
Example: "What is your job title?"
The analysis it gives you: counts, but only after cleanup. Use the typed variants where they exist, such as email, number, phone and date: validating the format at entry is what saves you that cleanup.
3. Single choice
What it is. Pick exactly one from a visible list. Radio buttons on a page-style form; a keyboard-selectable list on a conversational one.
Use it when the options are genuinely exclusive and you want clean segments: role, plan, main reason, primary channel. Do not use it when more than one answer can be true, or when the list runs past about seven options, where a dropdown reads better.
Example: "What is the main reason you are canceling?"
The analysis it gives you: counts and percentages that sum to 100, which makes this the cleanest type for cross-tabs and the safest to branch on. The failure mode is overlapping options: if "Too expensive", "Not enough value" and "Budget cut" all read as money, the split between them means nothing.
4. Multiple choice (select all)
What it is. A list where more than one option can be chosen, optionally capped at a maximum number of selections.
Use it when several answers are true at once: which channels someone uses, which features they rely on. Do not use it when you need a single winner. Percentages add up past 100 and no option looks dominant.
Example: "Which of these have you used in the last 30 days? Select all that apply."
The analysis it gives you: the share of respondents who chose each option, not the share of selections. Always include "None of these" and "Other", or people pick the least-wrong option and the data overstates every item.
5. Dropdown
What it is. A collapsed list that opens on click, usually searchable: country, industry, company size.
Use it when the list is long, standard and known in advance. Do not use it when there are four options, which belong on the screen where people can see them, and never for a scale, where seeing every point at once is the whole point.
Example: "Which country are you based in?"
The analysis it gives you: the same counts and shares as single choice. These answers are worth standardizing above all others, because they become the filters everyone else slices by.
6. Likert scale
What it is. A statement plus a symmetric agreement scale, usually five or seven points from strongly disagree to strongly agree.
Use it when you want to test a specific claim rather than a vague feeling, which is why it dominates employee and attitudinal research. Do not use it when you want a quality judgment, which is a rating scale, and never with an unbalanced set of points: if the options run from "Good" to "Excellent", everyone scores well and you learn nothing.
Example: "I have the tools I need to do my job well." Strongly disagree to strongly agree.
The analysis it gives you: a distribution and a top-two-box number, the share who agree or strongly agree. The Likert scale guide covers point counts and the neutral-midpoint argument.
7. Rating scale
What it is. A direct judgment on a numbered range, typically 1 to 5 or 1 to 10, sometimes drawn as a slider.
Use it when you want a number you can average and watch move: satisfaction, ease, quality. It is faster than a Likert scale and easier on a phone. Do not use it when you are tempted to change the range between waves. A 1 to 5 in March and a 1 to 10 in June are not comparable.
Example: "How easy was it to get set up? 1 is very difficult, 5 is very easy."
The analysis it gives you: a mean, a distribution and a trend line. Label both ends, or people invent their own meaning for the middle. See the rating scale questions guide.
8. Star rating
What it is. A rating scale drawn as stars, almost always five of them.
Use it when the subject is something people are used to rating, such as a purchase or a stay, and you want the fastest answer one thumb can give. Do not use it when you need nuance. Stars skew high and pile up at five and one, so the middle is thin and small changes look dramatic.
Example: "How would you rate the workshop?"
The analysis it gives you: an average out of five and the share of four and five-star answers. Put one open-ended question behind it, because a star never tells you what to fix.
9. NPS and 0-10 scales
What it is. One question on a 0 to 10 scale in fixed wording, with respondents grouped into promoters (9-10), passives (7-8) and detractors (0-6). The score is the percentage of promoters minus the percentage of detractors, from -100 to +100.
Use it when you want a single number to track by quarter against a published benchmark, paired with a follow-up asking why. Do not use it when your sample is small: the score swings below a few hundred responses, and a two-point move sends a team chasing nothing.
Example: "How likely are you to recommend us to a friend or colleague?" Leave the wording alone; rewriting it breaks the benchmark you are buying.
The analysis it gives you: the score, plus three segments worth reading separately. Start with the detractor comments. The NPS survey template has the standard wording set up.
10. Ranking
What it is. The respondent puts a short list in order of importance or preference.
Use it when everything on the list sounds important in isolation and you need a forced trade-off, such as which of five roadmap items matters most. Do not use it when the list runs past about five items: people rank the top two, guess the rest, and positions four through eight are noise. It is also the heaviest type on a phone.
Example: "Rank these four improvements from most to least valuable to you."
The analysis it gives you: an average rank per item and the share who put each item first. A capped multiple-choice question gets you most of that for a fraction of the effort, which the ranking questions guide goes into.
4 →
Which three of these would make the biggest difference to you?
Choose up to three. Press A to D, or click.
A ranking question without the ranking: a multiple choice question capped at three selections. It answers the same trade-off, and nobody has to drag anything on a phone. In Upform that is a Multiple Choice question with a maximum of three.
11. Matrix
What it is. One scale applied to several rows at once, such as rating six aspects of a conference on the same 1 to 5 range.
Use it when the rows genuinely share a scale and your audience is mostly on a desktop. Do not use it when people will answer on a phone, or when the grid runs past about six rows. Wide grids cause straight-lining, where someone picks one column and runs it down the page to escape.
Example: "Rate each of these from 1 to 5: venue, schedule, sessions, catering, networking."
The analysis it gives you: a mean per row and a heatmap of rows against scale points, the fastest way to see which item drags the average down. The matrix questions guide explains how to split one across separate screens without losing comparability.
12. Yes / No
What it is. A binary. Two options, one keystroke, no thinking.
Use it when you need a gate before a branch, or a plain fact: did you contact support, are you a current customer. It is the cheapest question in a survey and the best opener, because an easy first answer raises the odds someone finishes. Do not use it when the thing is a matter of degree. "Were you satisfied?" throws away everything interesting between yes and no.
Example: "Did you contact support about this?"
The analysis it gives you: a percentage and a fork in the road. Most well-built surveys are a spine of yes/no questions with the real questions hanging off them.
Closed vs open: the trade you are making
Eleven of the twelve types are closed: the respondent picks from answers you wrote. One is open. That single distinction decides more about your survey than any other choice.
Closed questions are countable the moment they arrive, so a hundred responses give you a chart the same afternoon. The cost is that the results can only contain what you already imagined: if the real reason people churn is not on your list of five, the data still looks confident and is still wrong. Open questions have the opposite shape. They cost the respondent effort and you an hour of reading, and they cannot be averaged, but they can change your mind.
So: open early in a project, closed once you know the answer space. Inside any one survey, keep roughly one open question per five closed, and put the open one after a closed one. Someone who has already answered something easy writes you three sentences; someone who meets a blank box on screen one writes nothing and leaves, which our guide to form abandonment covers in more detail.
Nominal, ordinal, interval and ratio, in plain English
Each question type produces one of four kinds of data, and the kind decides which math is legitimate. This is the layer most articles on question types skip, and it is why so many survey decks are quietly nonsense.
The level that catches everyone is ordinal. On a Likert scale, the distance from disagree to neutral is not guaranteed to match the distance from agree to strongly agree, so a mean of 4.2 rests on an assumption nobody can check. Report the median or the share in the top two boxes, and show the distribution next to it. The same caution applies to star ratings and to ranked positions.
| Level | What it means | Types that produce it | What you can legitimately calculate |
|---|---|---|---|
| Nominal | Labels, no order | Single choice, multiple choice, dropdown, yes/no, picture choice | Counts, percentages, the mode, cross-tabs. Never an average. |
| Ordinal | Ordered, gaps not equal | Likert scale, ranking, star rating, satisfaction scales | Median, mode, top-two-box share. Means are common and contested. |
| Interval | Ordered, even gaps, no true zero | Rating scales read as numbers, NPS 0-10 | Mean, standard deviation, differences between waves. Not ratios. |
| Ratio | Even gaps and a real zero | Number fields: headcount, revenue, visits, minutes | Everything, ratios and growth rates included. |
One practical consequence: if a question can be answered with a number, ask for the number. "How many people work at your company?" gives you ratio data you can band any way you like later. "Small, medium or large?" gives you three buckets you can never unpick.
How many of each to use
A good eight-question feedback survey usually looks like this: one yes/no gate at the top, two or three choice questions, one scale you intend to track for the next two years, one open-ended question, and any demographics last, where they feel like paperwork instead of an interrogation.
Four working limits. At most one matrix per survey, and only if desktop is where people will answer. At most one ranking, never longer than five items. One open-ended question per five closed ones. And no more than three or four distinct types in one survey: every new format costs the respondent a moment of working out how to answer it, and a survey that changes shape on every screen feels longer than it is.
Order matters as much as mix: easy and concrete first, scales in the middle, open text and demographics last. The survey questions guide has 120 examples grouped by what you are trying to learn, and how to make a survey walks through building one end to end. Then let a per-question drop-off funnel settle the argument: it names the exact question where people leave, which is nearly always the matrix, the third open-ended question, or a required field nobody had an honest answer to.
Which Upform question types map to which
Upform ships 21 question types, and they cover these twelve formats with three honest gaps. Open-ended is Long Text; short text is Short Text plus validated variants for email, number, phone, date and address. Single choice is Single Choice with A/B/C shortcuts, or Picture Choice for an image grid. Multiple Choice takes an optional maximum number of selections. Dropdown is a custom control, not a native browser one. Star Rating and Yes/No have their own fields, the second with Y and N shortcuts.
The gaps are worth naming. There is no dedicated Likert field: build one as a Single Choice question with five or seven labeled points, which is what a Likert scale is anyway. There is no ranking or matrix field either, so use a capped Multiple Choice for the first and separate screens for the second. For a 0-10 NPS question, use Slider Scale with a minimum of 0, a maximum of 10 and a step of 1. Upform will not compute the score, since there is no calculation engine, so export to Google Sheets and work it out there in one formula.
Survey question type FAQs
What are the 12 survey question types?
The 12 formats that cover almost every survey are open-ended (long text), short text, single choice, multiple choice, dropdown, Likert scale, rating scale, star rating, NPS or 0-10, ranking, matrix, and yes/no. Everything else is a variation on one of these, such as picture choice, which is single choice with images, or a slider, which is a rating scale you drag. Most surveys need four or five of them, not all twelve.
What is the difference between open-ended and closed-ended questions?
A closed-ended question gives the respondent a fixed set of answers to choose from, so the results are countable the moment they arrive. An open-ended question leaves a blank box, so answers come back in the respondent's own words and have to be read and tagged before they mean anything. Closed questions tell you how many; open questions tell you why. A good survey is mostly closed with one or two open.
Which survey question type gives the most reliable data?
Single choice with genuinely exclusive options, because there is one way to answer it and one way to read the result. Reliability falls as interpretation creeps in: rating scales depend on how you label the ends, Likert scales depend on a balanced set of points, and matrices lose accuracy on phones where people straight-line down a column. Open-ended questions are the richest and the least comparable.
What is the difference between nominal, ordinal, interval and ratio data?
Nominal data is labels with no order, like country or plan, so you can count it but never average it. Ordinal data has an order with uneven gaps, like a Likert scale, so the median and the top-two-box share are safer than the mean. Interval data has even gaps but no true zero, like a 1 to 10 rating, so differences are meaningful and ratios are not. Ratio data has even gaps and a real zero, like headcount or revenue, and every calculation is allowed.
Can you average a Likert scale?
Strictly, no. The distance from disagree to neutral is not guaranteed to match the distance from agree to strongly agree, so a mean rests on an assumption you cannot check. In practice many teams do average it and watch the trend, which is defensible as long as the scale never changes. The safer report is the percentage who agree or strongly agree, shown next to the full distribution.
How many question types should one survey use?
Three or four in a survey of about eight questions. Variety holds attention, but every new format costs the respondent a moment of working out how to answer it, and a survey that changes shape on every screen feels longer than it is. A reliable mix is one yes/no gate, two or three closed questions, one scale you track over time, and one open-ended question at the end.