A rating scale question asks a person to place one thing on a numbered range: how satisfied, how easy, how likely, how good. Everyone answers on the same range, so the answers can be averaged, compared between groups and tracked from one quarter to the next.
That comparability is the reason to use one, and it is also how rating data goes wrong. A scale with unlabeled ends, a different number of points from last year's survey, or a question that asks about two things at once still produces a tidy average. The average is simply measuring something other than what you think it measures.
The craft is small and learnable. Pick the right kind of scale, choose the number of points on purpose, label both ends in plain words, and report the spread of answers instead of a lone mean.
Five kinds of rating scale
Most rating scale survey questions take one of five forms. They are not interchangeable: each suits a different kind of judgment and returns a slightly different shape of data.
| Type | What the respondent sees | Best for | Watch out for |
|---|---|---|---|
| Numeric scale | A row of numbered buttons, such as 1 to 5 | Satisfaction, effort, quality, anything you plan to average | The numbers mean nothing without words at the ends |
| Star rating | A row of stars, usually five | Quick, consumer-facing quality checks | People read stars as a review, so scores run high |
| Slider | A handle dragged along a range with a minimum, maximum and step | 0–10 likelihood, wide or fine-grained ranges | Hard to land on an exact value if the step is too small |
| Labeled scale | A word on every point: Poor, Fair, Good, Very good, Excellent | Audiences who think in words, not numbers | The gaps between words are uneven |
| Agreement (Likert) | Strongly disagree to strongly agree, against a statement | Attitudes, several statements on one topic | People lean toward agreeing with whatever you wrote |
Numeric scales are the workhorse. They are compact, they fit any screen, and the answer is already a number. Their weakness is that a 3 means nothing on its own, so the ends need words, which the section on anchors covers.
Star ratings borrow the grammar of online reviews. That makes them instantly understood and slightly inflated: years of shopping have taught people that four stars is fine and three is a complaint. Use stars for quick quality checks where a friendly format matters more than precision, and never compare a star average with a numeric one.
Sliders earn their place when the range is wide or has to start at zero. A 0–10 likelihood question has eleven points, which is a crowded row of buttons on a phone and a comfortable slider. Set the step to 1 so people land on whole numbers, and show the value as they drag.
Labeled scales put a word on every point instead of a number. Survey methodologists generally find that fully labeled scales produce more consistent answers than scales labeled only at the ends, because nobody has to guess what a 4 means. The cost is space, so they work best at five points. In a one-question-per-screen form, a labeled scale is simply a single-choice question whose options are the labels.
Agreement scales are the Likert format: a statement, then strongly disagree to strongly agree. They have their own rules, covered in the Likert scale guide. The short version: prefer a direct question ("How easy was setup?") over a statement people can nod along to ("Setup was easy"), because the second invites agreement for its own sake.
Sometimes the right rating scale is no rating scale. If every item on your list scores high and you need a winner, a ranking question forces the trade-off a rating never will. The survey question types guide lays out the whole family and when each one fits.
How many points: 1–5 vs 1–7 vs 0–10
There is no universally correct number of points. There is a correct number for the judgment you are asking for and the people answering it. Three rules settle most cases.
Match the points to how finely people can actually tell levels apart. Most people can reliably separate five levels of satisfaction. Ask for ten and the extra points mostly add noise: one person's 7 is another person's 8 for reasons you will never recover. Seven points buys a little extra separation with engaged or expert audiences. Beyond that, the precision lives in the report, not in anybody's head.
Keep the scale you already have. If last year's survey used 1–5, this year's uses 1–5 with the same words at the ends. Changing the number of points breaks every comparison you might want to make, and stretching a five-point mean to fit a ten-point one afterward is arithmetic, not equivalence.
Use 0–10 when the scale is a standard. Net Promoter Score is defined on 0–10, and its promoter and detractor cutoffs only make sense on that exact range. The same goes for any benchmark you want to compare against: use the benchmark's scale, anchors and wording, or the comparison is decoration.
| Scale | Points | Good for | Trade-off |
|---|---|---|---|
| 1–3 | 3 | Very quick pulse checks, kiosks, young respondents | Too coarse to show change over time |
| 1–5 | 5 | Satisfaction, quality, most general-audience questions | Answers bunch at 4 and 5, so the top separates less |
| 1–7 | 7 | Effort, agreement, engaged or expert audiences | Harder to put a plain word on every point |
| 1–10 | 10 | Overall scores people already expect out of ten | No true midpoint, and some people read 1 as zero |
| 0–10 | 11 | Likelihood to recommend (NPS), probability questions | Crowded as buttons on a phone; a slider fits better |
Odd or even: the midpoint question
An odd number of points gives people a neutral middle. An even number forces them to lean one way. Neither is right in general. Keep the midpoint when neutral is a real, honest answer, and it often is: plenty of people genuinely have no feeling about your invoice layout. Remove it only when you have reason to believe people are hiding in the middle to avoid a judgment, and accept that you will turn some genuine neutrals into opinions they do not hold.
What a midpoint must never be is a home for "I don't know" or "not applicable". Those are different answers, and they quietly drag the average toward the center when they share a button with "neither". If a question may not apply to someone, ask a yes or no question first and show the scale only to the people it fits.
2 →
How likely are you to renew when your plan ends?
0 is not at all likely, 10 is extremely likely.
Five points for general audiences, seven for engaged ones, 0–10 only when you are matching a standard. Whatever you choose, use it for every comparable scale in the survey and in every repeat of it.
Anchoring the ends
The words at each end of a scale, called anchors, shape your data more than the number of points does. A 1–5 with no labels leaves every respondent to invent their own meaning for 1 and 5, and they will not all invent the same one.
Label both ends, every time. "1 is very difficult, 5 is very easy." Put it directly under the question, where it is read before anyone taps. A label that only appears after the first answer, or in an intro screen three questions back, does not count.
Make the anchors true opposites. "Very dissatisfied" to "Very satisfied" is balanced. "Not satisfied" to "Extremely satisfied" is not: the top anchor is stronger than the bottom one, and the imbalance pulls answers upward. Read your two anchors side by side and check that they are the same distance from neutral.
Decide unipolar or bipolar. A unipolar scale runs from none to a lot: "Not at all likely" to "Extremely likely". A bipolar scale runs from one pole through neutral to its opposite: "Very dissatisfied" through "Neither" to "Very satisfied". Likelihood, frequency and importance are unipolar. Satisfaction and agreement are bipolar. Mix them up ("Very unlikely" to "Very likely") and the midpoint means something nobody can explain.
Keep the direction consistent. Low on the left, high on the right, for every scale in the survey. Flip one to keep people alert and a share of them will answer it backwards without noticing. Once the answers are in, that mistake is invisible.
3 →
How easy was it to set up your account?
1 is very difficult, 5 is very easy.
30 rating scale examples by use case
Grouped by what you are measuring. Each rates exactly one thing and carries a suggested scale and anchors. Swap in your own product, event or team.
Satisfaction
- How satisfied are you with your experience today? (1–5, very dissatisfied to very satisfied)
- How satisfied are you with the speed of our support replies? (1–5, same anchors)
- Overall, how satisfied are you with the product you received? (1–5)
- How satisfied are you with the value you get for the price? (1–5)
- How satisfied were you with the booking process? (1–5)
A ready-made version, with an overall star rating up front and the follow-up questions already written, is the customer satisfaction survey template.
Effort
- How easy was it to find what you were looking for? (1–7, very difficult to very easy)
- How easy was it to get your issue resolved? (1–7)
- How much effort did it take to place your order? (1–5, very little to a great deal)
- How easy was it to set up your account? (1–5, very difficult to very easy)
- How easy is it to understand your monthly invoice? (1–5)
Likelihood
- How likely are you to recommend us to a colleague? (0–10, not at all likely to extremely likely)
- How likely are you to renew when your plan ends? (0–10)
- How likely are you to attend next year's event? (1–5, not at all likely to extremely likely)
- How likely are you to buy from us again in the next six months? (1–5)
- How likely are you to use this feature again this week? (1–5)
Quality
- How would you rate the quality of the food? (5 stars, poor to excellent)
- How would you rate the clarity of the instructor's explanations? (1–5, poor to excellent)
- How would you rate the condition of the item when it arrived? (5 stars)
- How useful was today's session? (1–5, not at all useful to extremely useful)
- How accurate was the information on our website? (1–5, not at all accurate to completely accurate)
Importance
- How important is offline access to your work? (1–5, not at all important to extremely important)
- How important is price when you choose a provider? (1–5)
- How important is it that we reply within one hour? (1–5)
- How important is a mobile app to you? (1–5)
- How important are integrations with the tools you already use? (1–5)
Confidence and fit
- How confident do you feel using the new reporting screen? (1–5, not at all confident to completely confident)
- How well did the onboarding call prepare you to get started? (1–5, not at all well to extremely well)
- How clearly were the next steps explained? (1–5, not at all clearly to extremely clearly)
- How well does the product fit the way your team works? (1–7, not at all well to extremely well)
- How confident are you that the problem will not happen again? (1–5)
Notice what the importance group is missing: a trade-off. Rate five features for importance and most will land at 4 or 5. If you need to choose between them, rate first and rank second. For wording beyond rating scales, the survey questions guide has 120 examples written for copying.
How to analyze rating scale results
Rating data arrives as one number per person per question, and the instinct is to average it. The average works as a headline. On its own, it is the least informative number you can report.
Look at the distribution first. Count how many people chose each point. Two questions with a mean of 3.0 can mean opposite things: everyone choosing 3, or half choosing 1 and half choosing 5. The first is a shrug. The second is two groups of people who need two different responses from you.
Report top-box share beside the mean. On a five-point scale, the top-two-box share is the percentage who chose 4 or 5. It is easier to explain than a mean ("76% were satisfied") and it moves when the thing you care about moves. Report the bottom-two-box share as well, so a growing unhappy tail cannot hide behind a stable average.
Use the median when answers pile up at one end. Satisfaction answers cluster near the top. A handful of very low scores drags the mean down far more than the median, so report both and label which is which.
Calculate NPS on 0–10 exactly. Promoters answer 9 or 10, passives 7 or 8, detractors 0 to 6. The score is the percentage of promoters minus the percentage of detractors, so it runs from -100 to 100. It only works on the scale it was defined on.
A worked example: 200 customers answered "How satisfied are you with our support?" on a 1–5 scale, in two consecutive quarters.
| Quarter | 1 | 2 | 3 | 4 | 5 | Mean | Top two | Bottom two |
|---|---|---|---|---|---|---|---|---|
| Q2 | 4% | 8% | 17% | 40% | 31% | 3.9 | 71% | 12% |
| Q3 | 10% | 10% | 4% | 30% | 46% | 3.9 | 76% | 20% |
Same mean, same median of 4. A dashboard showing only the average reports a quiet quarter. The distribution tells a different story: the neutral middle emptied out in both directions. More customers are very satisfied, and the share who are unhappy grew from 12% to 20%. Something changed for a subset of customers, perhaps a new support channel that works well for some and badly for others, and the mean hid it completely.
A mean tells you where the middle is. The distribution tells you who is standing there.
Two rules keep comparisons honest. Only compare results from identical scales with identical anchors and wording. And always report how many people answered each question, because an optional rating answered by forty enthusiasts is not the view of your customers.
None of this needs a statistics package. Upform exports every response to CSV on the free plan, or appends each one to a spreadsheet as it arrives through the Google Sheets integration. Each percentage in the table is a COUNTIF divided by the number of answers, and the mean is one AVERAGE. Upform does not calculate scores or NPS for you, so the formulas live in the sheet, where you can see them.
Seven mistakes that ruin rating data
Rating two things at once. "How satisfied are you with the speed and friendliness of our support?" A 3 could mean fast and rude or slow and lovely. Split it into two questions, or decide which one you actually need.
Leaving the ends unlabeled. Covered above, and still the most common flaw in rating scale surveys. Every scale gets words at both ends, in the help text, every time.
Leading wording. "How much did you enjoy the workshop?" assumes enjoyment and has nowhere for someone who did not. Ask "How would you rate the workshop?" and let the anchors carry the direction.
Changing the scale between waves. Moving from 1–5 to 1–10 halfway through a tracking study throws away every earlier result. If you must change, run both versions side by side for one wave so you can see how they relate.
Twenty ratings in a row. A long block of identical scales produces a column of identical answers: people stop reading and tap the same point. Keep ratings to the handful you will act on, and break them up with other formats.
Using the midpoint for "not applicable". It inflates the neutral share and drags the mean toward the center. Ask whether the question applies before showing the scale.
Never asking why. A 2 tells you something is wrong. It does not tell you what. Follow low scores with one optional open question, and read every answer.
The broader discipline, from question order to survey length, is covered in the survey design guide.
Building rating scale questions in Upform
Upform has three rating question types, all on the free plan with unlimited responses. Each maps to one of the scales above.
- Rating shows a row of numbered buttons from 1 to a maximum you choose, anywhere from 3 to 10. It always starts at 1, never 0.
- Star Rating shows between 3 and 10 stars, five by default.
- Slider Scale takes a minimum, a maximum and a step. This is the one for 0–10: minimum 0, maximum 10, step 1. The handle rests at the middle of the range, but a value is only recorded once the person moves it, so an untouched slider is never saved as a neutral answer.
None of the three has separate fields for end labels, so the anchors go in the question's description, which sits directly under the question on every screen size. A labeled scale with a word on every point is a Single Choice question with the labels as options, and an agreement scale is the same with strongly disagree to strongly agree.
Conditional logic can react to the score. For Rating and Star Rating the operators are equals, greater than and less than; Slider Scale adds greater or equal, less or equal and between. The most useful rule is also the simplest: when the score is low, show an open question that asks what went wrong. A field with a Show rule stays hidden until the rule fires, so nobody else ever sees it.
Conditional Logic
1 rule configured
Rule Summary
When Resolution ease less than "3" → Show field: What made it difficult?
For the 0–10 recommend question, the NPS survey template is already built on a Slider Scale, with the anchors written and a different follow-up question for detractors, passives and promoters. And if what you actually need is an order rather than a level, switch to a ranking question: a rating scale is the right tool for how much, and the wrong one for which first.