Forms · Glossary

The NPS question: exact wording and scale

The Net Promoter question asks how likely somebody is to recommend a company, product or service to a friend or colleague, answered on a scale from zero to ten where zero is not at all likely and ten is extremely likely. It is almost always followed by one open question asking the reason for the score.

The wording matters more than most single questions, because the value of the metric comes entirely from everybody asking it the same way. Change the scale or the phrasing and the benchmark you were chasing stops applying.

Nuwan Madhusanka · Co-founder

6 min read · Published 12 September 2026

How the score is built
Group	Scores	Counted as
Promoters	9 and 10	Added to the percentage
Passives	7 and 8	Ignored in the calculation
Detractors	0 to 6	Subtracted from the percentage
The score	Promoter percentage minus detractor percentage	A number from minus 100 to 100

Why the wording is fixed

The question was popularised by Fred Reichheld in a two thousand and three article and has been used in a consistent form ever since, which is the only reason cross company comparison is possible at all. The standard form names the thing being recommended, names the audience for the recommendation, and anchors both ends of the scale in words. Substituting would you recommend us, or shortening the scale to five points, or dropping the friend or colleague phrase all produce a number that cannot be compared with anybody else's. If you are not using the score for external comparison, none of that matters much and a simpler satisfaction question may serve better. If you are, the wording is part of the instrument rather than a style choice.

The follow up is where the value is

A score on its own tells you the direction and nothing about the cause. The open follow up is what makes the exercise actionable, and its wording changes what comes back. What is the main reason for your score produces justification of the number. What is the one thing that would have made this better produces a specific, actionable suggestion. Many teams ask both, which halves the number of people who answer either. Keep it to one box, make it optional, and size the box for two or three sentences. Reading a hundred of these is a couple of hours well spent and it is the part of the exercise most organisations skip, leaving them with a number that moves and no explanation.

Sampling and timing decide whether the number means anything

Net Promoter is a relationship measure, so it should be asked periodically rather than after every interaction. Send it after a support call and you will mostly measure the support call. Twice a year to any one customer is a sensible ceiling, and the sample should be drawn to represent the customer base rather than the people most willing to answer. Small samples produce a score that swings on a handful of responses, so a movement of a few points should not be treated as a change. Report the number of responses alongside the score, always, and be sceptical of any internal dashboard that shows the score without it.

The grouping throws away information

Collapsing eleven points into three groups and subtracting one from the other discards most of what was collected. A move from six to eight is a real improvement for that customer and does not change the score at all, while a move from six to seven turns a detractor into a passive and shifts the score for a smaller real change. That is why the raw distribution deserves as much attention as the headline number. Plot how many people chose each point and watch the shape rather than the summary. Two organisations with the same score can have entirely different distributions, and the one with a long tail at the bottom has a different problem from the one with a wall of sevens.

Building the question in a form

The nought to ten question is built as a radio field with eleven options, with both ends labelled in words so the scale is anchored. Radio, dropdown and checkbox fields can be laid out in one, two or three columns, and eleven options usually read best in a single row on desktop and stacked on a phone, so test it on a small screen. The follow up is a long text field, left optional. Answers export one column per field, with radio fields writing the option labels rather than the stored values, so the score column arrives ready to group.

Closing the loop on a low score

A detractor who writes a paragraph and hears nothing back is worse off than one who was never asked. Deciding in advance who reads low scores, how quickly they respond and what they are empowered to offer turns the survey from a measurement exercise into a retention one. Set a threshold, route anything at or below it to a named person within a stated number of working days, and keep the reply human rather than templated. Two cautions. Do not promise a response to everybody unless the volume genuinely allows it, since an unkept promise costs more than no promise. And keep the follow up separate from the score itself, because a customer who is contacted and satisfied should not have their original answer edited. Record the outcome alongside the response instead, so the pattern of what fixes a low score becomes visible over time.

Questions people ask

Should the scale be zero to ten or one to ten?

Zero to ten, which gives eleven points. Dropping the zero shortens the detractor range and produces scores that are not comparable with anybody else's. If comparability does not matter to you, the scale length is less important than being consistent with yourself across every wave you run.

Can I ask NPS and a satisfaction question together?

You can, and the usual result is that the second answer inherits the mood of the first. Two scoring questions in a row also train people to answer both the same way. If both metrics are needed, ask them on different occasions with different triggers rather than stacking them on one form.

What is a good score?

It depends entirely on the industry and the market, and published benchmarks vary so widely that they are easy to misuse. The comparison worth making is with your own score over time, using the same question, the same scale and a comparable sample. That is the only version where the confounds are held still enough to read a change.

How many responses do I need?

Enough that a handful of answers cannot move the number. A few hundred responses gives a reasonably stable score for most purposes; a few dozen does not. If your customer base is small, report the distribution and the count rather than the score, because a score calculated from thirty responses will swing alarmingly and mean nothing.

Should the question be anonymous?

Identified responses let you follow up on a low score, which is often worth more than the metric. Anonymous responses get slightly more candour. What fails is implying anonymity while collecting an identifier, because respondents notice and trust in the next survey drops. Say plainly which it is and what you will do with the answer.

Does the recommendation phrasing work for every business?

It works poorly where recommending is not something customers would ever do, such as a regulated monopoly or a service nobody discusses socially. In those cases the question collects a general sentiment reading rather than a recommendation intention, and a well worded satisfaction or effort question will be easier for people to answer honestly.

Make one with forms

The button opens the generator with this use case already described. Change the wording to match your own.

Create a form with OneCraft

Sources

Written and checked by the OneCraft team. Last checked 12 September 2026.