Every list of NPS Net Promoter Score questions starts the same way: how likely are you to recommend us, on a 0 to 10 scale. The useful part is everything after it, the follow-up that explains the score, the drivers that make it actionable, and the email that recovers the detractor. This page has the full battery, ready to field.
By the Wavefield Research team · Published Oct 30, 2026
The recommend question traces to Fred Reichheld's 2003 Harvard Business Review article The One Number You Need to Grow, and Bain & Company still runs the Net Promoter System built around it. The standard wording: on a scale of 0 to 10, how likely are you to recommend [company, product, or service] to a friend or colleague? Scores of 9 and 10 are promoters, 7 and 8 are passives, 0 through 6 are detractors, and the NPS score categories matter because the score is promoters minus detractors, nothing else. The discipline is keeping that wording fixed. Net promoter score questions earn their comparability, across industries and across your own waves, from everyone asking the same thing the same way, which is also why the worst edit you can make is an innocent-sounding one like adding highly before recommend.
One structural fact to design around: because NPS is a difference between two percentages, it is roughly twice as noisy as either percentage alone. Whatever nps survey questions you add around the core, size the sample first; the NPS calculator shows the margin of error your response count buys, and the sample size calculator works the same math in reverse.
A good NPS questionnaire is the core question plus a small, deliberate supporting cast. Here is the whole bank; most studies should field five to eight of these, not all of them.
Order matters more than it looks. The recommend question goes first, before anything primes the respondent: a driver battery asked ahead of the score nudges people toward whatever attributes you just made them rate. The open-ended why follows immediately while the reason is fresh, drivers come after, and segmentation questions go last, where survey fatigue does the least damage because they take no thought to answer.
“How likely are you to recommend [company] to a friend or colleague? (0 to 10)”
The standard core question. Keep the 0 to 10 scale and the recommend framing; both are what make your number comparable to anyone else's.
“What is the main reason for your score?”
The single most valuable follow-up. Open-ended, required for detractors if you gate anything, and the input your theme coding runs on.
“What is one thing we could do to improve?”
A sharper alternative to the reason question for passives, because it asks for the missing piece instead of a justification.
“How satisfied are you with [pricing / support / reliability / ease of use]? (1 to 5)”
Three to five of these, one attribute each, turn a bare score into a diagnosis: crosstab drivers against the NPS categories and the weak driver names itself.
“How well does [product] meet your needs today? (1 to 5)”
A fit question that moves earlier than the recommend score, which makes it a leading indicator worth tracking beside NPS.
“Which best describes your role? / How long have you been a customer? / Which plan are you on?”
Only what you cannot pull from your CRM. Every segmentation answer you already know is a question wasted on respondent patience.
“How likely are you to recommend working at [company] to a friend? (0 to 10)”
The eNPS variant for employees. Same arithmetic, different confidentiality stakes: run it anonymous or expect inflated scores.
“Based on your recent [support interaction / purchase / delivery], how likely are you to recommend us? (0 to 10)”
The transactional variant. Keep it in a separate trend line from relationship NPS; blending the two moves the number whenever ticket volume does.
The open-ended why deserves more craft than it usually gets; wording options and coding approaches are covered in open-ended survey questions.
The detractor follow-up is the highest-leverage email in the whole NPS loop and the piece most question lists skip. Two rules before the templates: send it within a day or two while the survey is still remembered, and make it a real message from a named person, because a detractor who just told you the relationship is broken does not want to hear from a no-reply address.
Hi [name],
Thanks for the straight answer on our survey. You scored us a [score] and pointed at [issue], and you are right that it has not been good enough.
Here is what happens next: [one concrete action, with a date]. I will follow up when it ships.
If I misread the problem, reply and correct me. This inbox is mine, not a queue.
[Name, role]
Hi [name],
You gave us a [score] this week, and I wanted to reach out personally rather than let that disappear into a dashboard.
One question, and I promise it is the only one: what is the single thing that most needs to change for us to earn a higher number from you?
Reply in a sentence; I read every one of these.
[Name, role]
What not to do is just as settled: no coupon in the first message, because it reads as buying the score back rather than fixing the problem; no survey link, because they just took one; and no defense of the thing they criticized. The template is a scaffold, and the moment it reads templated it stops working, so rewrite the bracketed parts in your own voice every time.
Change the question and you start a new trend line, whatever the chart implies. If you must change the measure, run old and new wording side by side for a wave so the break is measured instead of guessed at.
An intro listing everything you shipped this quarter, followed by the recommend question, is a push poll with a dashboard. The question stands alone or the score means less.
How satisfied are you with our pricing and support is unanswerable for the customer who loves one and resents the other. One attribute per driver, every time.
An always-open NPS survey samples different moods in different months and calls the blend a trend. Field in defined windows, cap any individual at a couple of asks per year, and watch the response rate as its own metric.
The easiest failure to miss is reading a 4-point move on 200 responses as news. That is inside the margin of error, and a questionnaire can be perfect while the readout ruins it. Why response rates and non-response bias bend the score, and what raising them costs, is covered in survey response rates.
Respondents scoring 9 or 10 are promoters, 7 or 8 are passives, and 0 through 6 are detractors. The score is the percentage of promoters minus the percentage of detractors, which lands between -100 and +100. Passives count in the base but not in either percentage, which is why a pile of 7s and 8s quietly drags a score toward zero.
Anything above zero means more promoters than detractors, and widely cited industry averages run from the teens to the 40s depending on sector. The honest answer is that your own trend beats any benchmark: cross-company comparisons break on survey mode, timing, and sample differences. Compare this wave to your last wave, measured the same way, and read the move against its margin of error.
Percentage of promoters minus percentage of detractors, using all responses including passives in the base. From 200 responses with 90 promoters and 50 detractors: 45% minus 25% gives an NPS of +20. Our free NPS calculator does the arithmetic and, more usefully, prints the margin of error beside the score, which is the number that says whether a change is real.
The core question, the open-ended why, and at most a handful of drivers and segmentation items: five to eight questions total is a sensible ceiling, and two is a legitimate survey. Every added question costs completion rate, and an NPS survey that takes ten minutes selects for your most opinionated customers, which skews the very score you are measuring.
Change the bracketed subject, never the structure. Swapping company for product, or adding a transactional lead-in, keeps the instrument intact. Changing the scale, dropping the recommend framing, or adding qualifiers like highly recommend breaks comparability with your own history and everyone else's. If you run a tracker, lock the wording at wave one and treat any edit as starting a new trend line.
Customer NPS usually should not be, because the follow-up is the point: you want to close the loop with the detractor who wrote the angry paragraph. Say plainly that responses are linked to accounts. Employee NPS is the reverse: anonymity is what buys honesty, so run eNPS unattributed and accept that you close loops with teams rather than individuals.
Related: NPS calculator · NPS survey software · open-ended survey questions · customer feedback surveys
Brief the agent and it programs the whole battery: core question, drivers, logic, and the coded why. $99 per NPS study, $99 per tracker wave.
Build an NPS study free