The net promoter question is one sentence and the scale is eleven points, so writing it takes a minute. What takes judgement is everything around it: who you ask, when, what you ask second, and what you do with the people at the bottom. A score collected without those decisions is a number that moves for reasons nobody can explain, which is how a great many companies end up watching one for years without it changing anything.
The question, and the one that follows it
Ask how likely somebody is to recommend you on a nought to ten scale, then ask why in an open box. The second question is where all the usable content is: the number tells you the size of the problem and the sentence tells you what it is. Without the follow-up, a fallen score sends everyone into a meeting to speculate about causes nobody recorded.
Who to ask, and when not to
Recommendation is a question about the relationship, so ask people who have had one: several interactions, or enough time to form a view. Asking a first-time buyer minutes after checkout produces a rating of the checkout, which is a satisfaction question wearing the wrong scale. Rotate the list so the same customers are not asked every quarter.
Reading the three groups honestly
The scale splits into promoters, passives and detractors, and the headline number is the promoter share minus the detractor share. Two things follow. Movements of a couple of points on small samples are noise, and a score can rise because detractors left rather than because anybody got happier, which is why the count of responses belongs beside the score every time it is shown.
The follow-up is the whole return
The detractors are a list of people who have told you something is wrong and are still talking to you. Working that list within a day or two is where the money in this exercise actually is; the free worksheet on this site sizes it, because a hundred detractors a quarter is a real amount of somebody's week.
Questions people ask about net promoter score survey questions
What is a good NPS?
This site does not publish one. A benchmark is a measurement of other people's customers; your own score last quarter, with the response count next to it, is the comparison that means something.
How many responses do we need for the score to be stable?
Enough that a few answers cannot swing it. Show the response count beside the score always, and treat small movements on small samples as noise.
Should we ask NPS after every interaction?
No. Recommendation is a relationship question. For a single interaction, ask a satisfaction question instead.
Can we run NPS for free?
Yes. The question is one sentence, the scale is standard, and the worksheet here sizes the round. What costs money is keeping the answers and working the follow-up list, which is what Pro does.