A ranking question asks respondents to put a short list of items in order against a single criterion: most to least important, first to last choice. It forces a trade-off, because no two items can share a place.
That constraint is the whole point and the whole problem. Put five features on a rating scale and a polite respondent calls four of them "very important", which builds no roadmap. Make them order the same five and somebody has to come last. In exchange you give up intensity: you learn that A beat B, never by how much.
Ranking is also one of the most misused question types in survey writing, almost always in the same two ways: too many items, and no stated criterion.
Ranking vs rating vs multiple choice
These three look interchangeable on a feature list and behave completely differently in a dataset. The question is not which is nicer, but what shape of answer you need on the other side: an order, a level, or a set.
| Ranking | Rating | Multiple choice | |
|---|---|---|---|
| What the respondent does | Puts items in order, 1 to N | Scores each item on the same scale | Ticks every item that applies |
| What you get back | A position per item | A number per item | A yes or no per item |
| Can two items tie | No | Yes | Not applicable |
| Tells you how strongly | No | Yes | No |
| Effort for the respondent | High, every item weighed against every other | Low, one judgment per item | Lowest, one pass down the list |
| Behaves on a phone | Badly, since drag targets are small | Well | Well |
| Reach for it when | Everything sounds important and you need one winner | You want a number to track over time | You need to know which options are in play |
| Typical failure | People place the top two and guess the rest | Everything scores four out of five | Everything gets ticked |
A fourth option gets forgotten and is often the right one: a multiple choice with a maximum. "Select the three that matter most" from a list of eight is far cheaper than ranking eight items, and it tells you which items are in the top group without anybody having to separate sixth place from seventh. The survey question types guide lays out the full set.
When a ranking question is the right tool
Three situations genuinely call for one. They share a shape: a decision is waiting that can only go one way.
You have to choose, and the list is all upside. Four roadmap items, three venues, five benefits you could fund. Nobody calls any of them a bad idea, so "how important is this?" returns a wall of high scores. Ordering surfaces a preference the respondent holds but would never volunteer.
Your rating results have already flattened. If last quarter came back with averages of 4.6, 4.5 and 4.4 out of 5, the scale has stopped discriminating. A ranking re-separates the same items, because people now compare them with each other rather than with an abstract idea of "important".
The order itself is the deliverable. Session scheduling, shortlists, seating, sequencing a rollout. When the output of the research is a list in order, asking for a list in order is not a proxy. It is the thing itself.
Before you write a ranking question, say out loud what you will do differently if item three beats item four. If you cannot, use a rating scale and save everyone the work.
When it is the wrong one
Ranking is expensive, and the cost is charged to the respondent. Three reasons to use something else.
It falls apart past five or six items
Ordering a list is not one decision, it is many. Five items is ten implicit pairwise comparisons; eight items is twenty-eight, and nobody performs twenty-eight comparisons in a survey. What happens instead is that people place their clear first and second choices and fill the rest in roughly the order the options appear on screen. Your top two are real. Positions four through eight are the order you wrote the list in, handed back with a standard deviation attached.
It throws away how much anyone cares
A ranking records order and nothing else. The gap between first and second might be a chasm or a coin flip, and your data cannot tell the difference. It is also relative to your list, so an item finishes first among five options a respondent is lukewarm about and you read it as enthusiasm. If strength of feeling matters, pair the ranking with a rating scale question or a Likert scale.
It is the hardest question type on a phone
Drag-and-drop asks somebody to press, hold, drag and drop inside a list that scrolls under their finger, on targets a few millimetres across, and a mis-drop changes an answer silently. On mobile it is one of the most common places for people to quit a survey outright, a theme the form abandonment guide picks up. If most of your traffic is mobile, break the ranking into one tap per screen or drop the format.
How many items to rank
The honest answer is five. Four if the question sits late in a long survey. Six only when the items are things the respondent already has opinions about: their own team's benefits, say, rather than feature names they met thirty seconds ago.
Three items is usually a ranking that should have been a single choice: ask which is best and you have the same information for a third of the work. If your list has ten items, the list is the problem. Screen it first: "which of these do you use at all?" Then rank the survivors, or take a top three by multiple choice and stop.
Five items, one criterion, one screen. If you cannot get the list down to five, you are asking a question you have not finished thinking about.
3 →
Which of these would make the biggest difference to your week?
Pick one. We will ask for your second choice next.
15 example ranking questions
Grouped by what you are trying to learn. Every one names a criterion and stops at five items, which is the entire craft of writing these. Edit the option lists for your own context.
Product and roadmap
- Rank these four improvements from most to least valuable to your team: faster search, offline mode, better exports, a mobile app.
- Put these five reasons for choosing us in order, starting with the one that mattered most.
- Order these three plans from the one you would buy today to the one you would buy last.
Employee and internal
- Rank these benefits from most to least valuable to you: extra leave, learning budget, home-office allowance, private healthcare.
- Put these four things in order of how much they slow your work down in a normal week.
- Rank the ways you would most like to get feedback: one-to-one, in writing, in a team review, informally.
Events
- Rank these session topics from the one you would most like to attend to the one you would skip.
- Put these three dates in order of preference for next year's conference.
- Rank what mattered most in your decision to come: speakers, location, price, the chance to meet people.
Education and training
- Rank these four formats from most to least helpful: live sessions, seminars, recorded video, group projects.
- Put these assessment types in order of how fairly you think they measure what you actually learned.
- Rank the parts of this course you would keep if only two of them could stay.
Market research
- Rank these five brands from the one you would buy today to the one you would buy last.
- Put these attributes in order of how much they influence what you pay: price, brand, sustainability, warranty, design.
- Rank these three pack sizes from most to least likely to end up in your basket.
For a longer library, including the rating and open-ended questions that pair well with these, the survey questions guide has 120 written for copying.
How to analyze the answers
Ranking data arrives as a position per item per respondent. Two numbers come out of it, and reports usually leave out the second.
Average rank. Score first place 1, second place 2, on to last, then average each item across everyone who ranked it. Lower is better, which trips up every readout ever presented, so label the column and say so underneath. It gives you the order of the list and nothing else.
First-choice share. The percentage of respondents who put the item in position one. This is what tells you whether anybody feels strongly. Add last-place share beside it and you have the shape of the answer, not just its middle.
A worked example: 100 people ranked four improvements.
| Item | Average rank | Ranked first | Ranked last |
|---|---|---|---|
| Faster search | 2.4 | 18% | 12% |
| Offline mode | 2.4 | 41% | 35% |
| Better exports | 2.5 | 20% | 22% |
| A mobile app | 2.7 | 21% | 31% |
Read the first column alone and the list is a four-way tie, so you build whatever was already on the plan. Read the next two and it changes completely. Offline mode is the only item anyone feels anything about: four in ten want it first, more than a third want it last. Faster search has an identical average and no passion behind it in either direction.
Those are not two similar features. They are one feature with two audiences and one feature nobody objects to. The average hid it, because a polarized item and a consistently middling item land on the same number. That is the most common way ranking results get misread, and the fix costs two columns.
An average rank tells you the order. It never tells you whether the order matters.
Two rules keep the rest honest. Never compare average ranks across questions with different item counts. A 2.4 out of four and a 2.4 out of eight mean opposite things. And if people rank only their top three from a longer list, an item chosen by twelve will post a flattering average precisely because the other eighty-eight ignored it, so report how many people ranked each item at all.
None of this needs a statistics package: export the responses to CSV and the three columns above are three spreadsheet formulas. Upform exports on the free plan, and form analytics separately tells you whether the ranking question is where people stopped, which, if you ignored the item limit, it will be.
Writing them well: six rules
Name the criterion in the question. "Rank these features" is not a question. By what? What you would pay for, use daily, miss most? Different criteria produce different orders from the same person, and if you do not choose one, each respondent silently chooses their own.
Say which end is which. Put "1 = most important" in the help text, every time. A good share of bad ranking data is nothing more exotic than people ordering the list backwards, and it is invisible once the answers are in.
Keep the items mutually exclusive. If "price" and "value for money" both appear, they split the same vote and the order is an artifact of your wording. Merge anything a hurried reader would take as the same idea.
Use the respondent's words. "Offline mode" ranks; "Feature 27" does not. Internal names force the reader to translate before they can compare, and translation is where they give up and order the list top to bottom.
Never put "Other" in a ranking. An unnamed item cannot be weighed against named ones, and wherever people park it, it distorts every average on the list. Ask for the open answer separately.
Ask why straight afterwards. One optional question ("what made your first choice first?") is often worth more than the ranking itself. The order tells you what to do; the sentence tells you why.
The form design best practices guide covers the layout side of the same discipline: one question per screen, a named criterion, nothing optional that nobody reads.
Building one in Upform
Being straight about this: Upform does not have a dedicated ranking field. There are 21 question types and a drag-to-order list is not among them. Two approaches get ranked data out of an Upform survey anyway, both on the free plan.
A sequence of single-choice questions. Ask "which of these matters most?" on one screen, then "and which comes second?" on the next, using the same options. Four items become three screens; the fourth is whatever is left. It is keyboard-fast (A, B, C, Enter) and it behaves on a phone, which a drag list does not. The honest cost: nothing stops somebody picking the same option twice, so say "pick a different one" in the help text and check the export for duplicates. Conditional logic cannot remove a used option, but it can jump past the follow-ups once the first answer has settled the branch.
A slider per item. Give each item its own Slider Scale question with the same minimum and maximum, and tell people to use each number once. You get a value per item that sorts into an order, plus the intensity a true ranking discards. The cost is one screen per item, so it suits four or five, not eight.
Approach two, in the results
Four Slider Scale questions sharing one range. The order falls out of the numbers, and the numbers show the gap between first and second.
Which to pick: three or four items, use the choice sequence; five where strength of feeling matters, use sliders; anything longer, go back to how many items to rank and cut the list. And if what you need is the level rather than the order, a rating scale takes one screen and everybody finishes it. Ranking is a good tool used about a tenth as often as it appears in surveys.