Declining survey response rates: why surveys still matter
Declining survey response rates are real and decades old. Why fewer customers answer, why surveys still matter, and smarter, shorter ways to ask them.
Table of contents
- Key takeaways
- The evidence on declining survey response rates
- What causes declining survey response rates
- Why customer surveys are still worth running
- Response rate vs nonresponse bias: what each tells you
- How to ask smarter: eight ways to get better answers
- A worked example: every ticket vs a sampled single question
- Where to start
- FAQ
The survey report has a line nobody likes to read aloud: the response rate is down again. It was down last quarter, and the year before. Someone suggests a reminder email, someone else a prize draw, and a third person asks the real question: with declining survey response rates, is any of this still worth doing?
A survey response rate is the share of the people you invited who completed the survey. It has been falling for decades in the most carefully run surveys there are, the ones fielded by pollsters and government statisticians, and customer surveys are not exempt.
Surveys are still worth running. The answer is neither more surveys nor none: it is smarter ways to ask, surveys that take less effort to finish, and one habit most teams skip, checking whether the people who answered look like those who did not.
Key takeaways
- Response rates have been falling for decades, even in government surveys, so a falling rate on your own survey does not mean you broke something.
- Much of the decline in customer surveys is self-inflicted: too many requests, too many questions, forms that fight a phone, and no visible result from answering.
- Surveys remain the only scalable way to hear why customers behave as they do, including those who never complain, and to track a comparable number.
- A low response rate does not by itself mean biased results; bias depends on whether the people who answer differ from those who do not on the thing you measure.
- The smarter survey asks fewer people fewer questions, in the moment and channel of the experience, and checks respondents against non-respondents.
The evidence on declining survey response rates
The longest public record comes from telephone polling. Pew Research Center reported that its phone surveys had a response rate of 37 percent in 1996 and 9 percent in 2014. After a few stable years the slide resumed, to 7 percent in 2017 and 6 percent in 2018.
Government surveys show the same curve. In the US Bureau of Labor Statistics’ household survey response rates, the Current Population Survey, the monthly survey the unemployment rate comes from, went from 86.4 percent in February 2016 to 65.9 percent in February 2026. The interview part of the Consumer Expenditure Survey fell from 63.8 percent to 42.6 percent between February 2016 and January 2024.
Those surveys have trained interviewers and the budget to knock on doors. If they cannot hold their response rates, a post-purchase survey will not buck the trend with a new button color. Some of the decline is the climate you work in. A good part is your own doing, and that part can change.
What causes declining survey response rates
Pew named, among other reasons, concerns about intrusions on people’s time and privacy and a general lack of interest in taking surveys. In customer programs, six causes do most of the damage.
Every interaction ends in a survey. A survey costs almost nothing to send, so every team adds one: after the delivery, the support ticket, the app update, the renewal. Together they teach the customer that feedback requests are background noise, the same arithmetic that makes cheap mass marketing expensive in goodwill.
Long questionnaires about things the customer did not do. A support survey that also asks about the website and the brand is three surveys stapled together. Customers notice, and they stop at the grid.
No visible effect of answering. A customer who reported a broken return label last spring and got the same broken label this spring has learned something about your survey.
Filters and channel fatigue. Pew pointed to the surge in robocalls, an estimated 3.4 billion a month at the time, and to technology that wrongly flags survey calls as spam. Email has its own version: invitations look like marketing, get filtered like marketing, and sit among a dozen other “how did we do?” messages.
Forms that fight a phone. Invitations are often opened one-handed, in a queue. A grid of fifteen attributes, a dropdown of every country, or a login screen before the first question each lose people who were willing to help.
Trust and privacy. People are rightly wary of links in unexpected messages. If the invitation does not say who is asking, why, and what happens to the answer, the careful customer ignores it.
Most of that list is made of design decisions, which is the good news.
Why customer surveys are still worth running
When response rates fall, someone proposes replacing surveys with behavioral data. It is excellent data, and it still cannot do three things a survey does.
It shows what, not why. Analytics show that people abandon the checkout at the delivery page. They cannot say whether it was the price, the dates on offer, or a form that wiped the address. One well-placed question can.
It cannot hear from the quiet. Complaints come from customers who bother to complain. Many unhappy customers simply drift off, and customers who go silent leave nothing in the complaint log. A survey that reaches a fair sample of them is the only scalable way to hear their reasons in time.
It gives you a comparable number. The same question, asked the same way of a similar sample every quarter, is how you know whether the fix to the returns process did anything.
Interviews are richer but do not produce a trend line across thousands of customers. Surveys are the cheapest way to ask a representative question at scale, which is exactly why they get overused.
Response rate vs nonresponse bias: what each tells you
Here is the reframe. What you care about is not the response rate but whether the answers you got are close to what everyone you asked would have said.
The gap between those two is nonresponse bias. It appears only when the people who answer differ from the people who do not on the thing you measure. If the customers who reply to your satisfaction survey are happier than the ones who ignore it, your score is too high at any response rate. If they are not, a small response still gives you an honest number.
The research is more reassuring than most teams expect. In a 2006 paper, the survey methodologist Robert Groves pulled together 30 studies that had estimated nonresponse bias. The correlation between nonresponse rate and the size of the bias was 0.33, so the rate accounted for only about a tenth of the variation in bias. Most of the variation sat between different questions in the same survey, not between surveys. A later meta-analysis by Groves and Emilia Peytcheva of 59 studies found very little correlation between nonresponse rates and bias.
Pew’s testing shows what that looks like. Its low-response phone polls tracked high-response in-person surveys well on party affiliation and religion, but not on civic engagement: 46 percent of phone respondents reported working with neighbors, against 8 percent in the Current Population Survey. One likely reason is that agreeing to a survey is itself a helpful act, so the people who agree are also more likely to volunteer. Nor does a higher response rate cure this kind of bias: phone polls overstated voter registration by about as much in 1996, at 37 percent response, as in 2014, at 9 percent.
The customer version is easy to picture: if the people who answer a post-ticket survey are mostly the delighted and the furious, the average hides the lukewarm middle, and the response rate cannot show you that. It is also why the usual rescue, more reminders and a reward, can raise the rate without improving the answer. A National Academies review noted that incentives reduce refusals, but little is known about their effect on nonresponse bias.
| Question | Response rate | Nonresponse bias |
|---|---|---|
| What does it measure? | Share of invited people who finished | Distance between respondents’ answers and everyone’s |
| What moves it? | Frequency, length, channel, timing, reminders, incentives | Whether the urge to answer is linked to what you measure |
| Is it on the dashboard? | Always | Almost never |
| What does it tell you? | How much you ask of customers | Whether to trust a particular number |
| How do you check it? | Completes divided by eligible invitations | Compare respondents with non-respondents on data you hold |
None of this makes a low rate harmless. Pew’s view is that low rates do not make a poll inaccurate on their own but do signal a higher risk of error. Treat the rate as a gauge of goodwill used up, and bias as the question of whether a number is right. It is the same stance as acting on customer data that will never be perfect: good enough for the decision beats perfect and late.
How to ask smarter: eight ways to get better answers
Each of these makes the survey easier to finish, fairer to the customer, or more honest about who answered.
1. Ask fewer questions
Next to every question, write the decision it informs and who owns that decision. Any question with a blank beside it goes. What usually survives is one rating the team tracks, one open box asking why, and perhaps one question about what just happened.
2. Ask in the moment, in the channel where it happened
A question about the checkout belongs at the end of the checkout, not in an email three days later. A chat or messaging conversation gets its question in the same thread. When the experience ends in an email, put the first question inside the email itself, so tapping a number is the answer and the rest is optional. None of this helps if you cannot reach customers by email at all, which is worth fixing first.
3. Sample, and set a contact-frequency rule
A random sample, sized for the precision you need, tells you how a process is working, and it leaves everyone else fresh. Pair it with one rule every team follows: no customer is asked more than once in a set period, whatever the trigger. That takes a shared record of who was asked when, and someone with the authority to refuse the eleventh survey.
4. Design for a thumb on a phone
One question per screen. Large tap targets. No grids, no dropdowns where three buttons would do, no login. Before anything goes out, fill it in yourself on an old phone over a slow connection, standing up.
5. Do not ask what you already know
If the customer came from a link you sent, you already know the product, the order and how long they have been a customer. Attach that to the response instead of asking again, and say in the invitation what you attach, which also helps with trust.
6. Close the loop so customers see what changed
In my experience, people answer the next survey when answering the last one visibly did something. Reply personally to the customer who reported a specific problem. Publish a short, dated note of what changed because of feedback, and link to it from the next invitation.
7. Mix methods instead of adding questions
When someone wants to add a question, ask whether another method would answer it better. Five short conversations explain a problem more fully than a new scale sent to thousands, and conversations with your best customers are a good place to begin. Support transcripts already hold what customers said unprompted, and pairing scores with comments is how a number turns into a reason.
8. Check nonresponse bias with data you already hold
This is the step almost nobody takes. You know things about the people who did not answer: tenure, spend, plan, region, recent contacts. Compare them with respondents. If the two groups match closely, a low rate matters less. If respondents skew toward long-standing customers with few support contacts, you know which way the score leans and can weight it or say so on the slide.
Better still, look at what non-respondents did next. If they churned faster, your satisfaction score is describing the survivors. The logic is the same as measuring with a control group: the people you did not hear from are the comparison.
A worked example: every ticket vs a sampled single question
The numbers here are illustrative and round. Suppose a support team closes 10,000 tickets a month. In the tired setup, every closed ticket triggers a twelve-question email survey the next day, so a customer with three tickets gets three surveys. Suppose 5 percent respond: 500 completed surveys and 9,500 unanswered invitations.
In the smarter setup, one ticket in four is sampled at random, skipping anyone surveyed in the last ninety days. The resolution email carries one rating question, answered with a tap, plus an optional comment. Suppose 20 percent respond: again 500 responses, from a quarter of the invitations.
| Tired way | Smarter way | |
|---|---|---|
| Who is asked | Everyone, after every ticket | A random quarter of tickets, with a ninety-day rest rule |
| Invitations a month | 10,000 | 2,500 or fewer |
| Questions | Twelve, in a separate survey | One, inside the resolution email, plus a comment |
| Responses at the assumed rate | 500 at 5 percent | 500 at 20 percent |
| Bias check | None | Respondents compared with non-respondents on tenure and ticket type |
The response count is the same, but at least 7,500 fewer invitations go out, the answers arrive while the ticket is fresh, and the team knows whether the 500 resemble the 2,500. Product, category and handling time come from the ticketing system; the reasons come from the comments.
Where to start
- Count the surveys one customer can receive. List every survey any team sends, with its trigger, and work out the most a busy customer could get in a quarter.
- Cut one questionnaire in half with the decision-and-owner test.
- Move the first question into the invitation for your highest-volume survey.
- Agree a contact-frequency rule across teams, and name who enforces it.
- Run one bias check on last quarter’s respondents and non-respondents, comparing tenure, spend and recent contacts.
- Publish one “what changed” note and link it from the next invitation, so the people who answered can see that someone read what they wrote.
FAQ
Why are survey response rates declining?
People are asked for feedback after almost every purchase, delivery and support contact, often in long surveys about things they did not do. Spam filtering, call blocking and wariness about unexpected links cut how many invitations are even seen, and customers who saw nothing change last time have little reason to answer again.
What is a good survey response rate?
It depends on the channel, the relationship and the moment, so there is no single benchmark worth copying. A better test is whether respondents resemble the customers you invited on attributes you already know, such as tenure, spend and region. Track your own rate over time as a measure of fatigue.
Does a low response rate mean my results are biased?
Not necessarily. Bias appears when the people who answer differ from those who do not on the thing being measured, and research across many surveys has found the response rate to be a weak predictor of how large that bias is. A low rate does raise the risk, so check it by comparing respondents with non-respondents.
How can I increase survey response rates?
Ask fewer customers fewer questions, at the moment and in the channel where the experience happened, with the first question inside the invitation. Make the survey easy to finish on a phone, stop asking for information you already hold, and show customers what changed because of earlier answers.
Should I use incentives for customer surveys?
Incentives tend to raise participation, mainly by reducing refusals, and research on mail surveys found that paying up front works better than promising a reward. Much less is known about whether they reduce bias. Fix length, timing and relevance first, and if you add an incentive, check whether it changed who answered.