A survey closes, the number comes out at 14 percent, and the question that follows is always the same: is that bad? The search for an average is really a search for permission to stop worrying, or a reason to start. Neither arrives from a single figure, because the published averages for survey response rates are not measuring the same thing as each other, and almost none of them are measuring the same thing as the survey that just closed.
That is not a reason to give up on the number. The response rate is one of the few pieces of evidence available about whether the answers in hand represent the people who were asked. It is worth calculating carefully and worth comparing. The trick is knowing what it can be compared against.
What the rate is counting, and the two numbers it gets mixed up with
A response rate is answers divided by the number of people who were asked, expressed as a percentage. Three different quantities hide inside that sentence, and each one produces a different figure from exactly the same survey.
The denominator is where the disagreement lives. Asked can mean the number of invitations sent, the number of invitations that were delivered, the number of people who opened the invitation, or the number of people known to be eligible. Sending 1,000 emails, having 850 delivered, and collecting 120 answers gives a rate of 12 percent or 14.1 percent depending on which of the first two figures goes underneath. Both are defensible. Only one of them can be compared with last quarter, and only if last quarter used the same one.
The numerator has the same problem in miniature. A partial answer, abandoned at question nine of twelve, is a response by some definitions and not by others. Screened-out respondents who were never eligible sit awkwardly in both places.
Two other rates get called the response rate in casual use:
- Completion rate is the share of people who started the survey and reached the end. It measures the questionnaire, not the invitation.
- Click or open rate on the invitation measures the subject line and the sender reputation. A healthy open rate with a low response rate points at the form. A poor open rate with a high completion rate points at the invitation.
Separating those three tells you where to work. A single blended figure tells you only that something is happening.
The American Association for Public Opinion Research maintains the reference document for this, Standard Definitions, which sets out final disposition codes for every case in a sample and several distinct outcome rates built from them. It exists precisely because researchers kept publishing incomparable numbers. The detail is heavier than most teams need, but the underlying discipline is not: decide what counts as a response and what counts as asked, write it down, and keep using it.
Why there is no single average to check yourself against
Published averages for survey response rates range from single digits to above 80 percent, and the spread is not noise. It reflects four things that change the answer completely.
Who is being asked. A request to an employee from their own organisation, a request to a paying customer about a product they use weekly, and a request to a stranger from an unknown number are three different social transactions. Response rates fall in that order, steeply.
Whether there is an existing relationship. People answer organisations they already deal with. That single factor separates most high published figures from most low ones.
When the request arrives. A question asked immediately after a delivery, a visit or a support ticket catches the moment the experience is still live. The same question sent three weeks later competes with everything else in the inbox.
How the sample was built. A census of a known list of 40 people and a random sample of the general population cannot produce comparable rates, because reaching a named colleague and reaching an unknown household are not comparable tasks.
Two published reference points show the scale of the gap. Pew Research Center reported that response rates for its telephone public opinion polls fell to 7 percent in 2017 and 6 percent in 2018, having held around 9 percent for several years before that, in an analysis of its own survey data. Survicate, which runs customer feedback surveys, recommends aiming for 25 percent on a customer survey campaign and treating that as an initial goal. Neither figure is wrong. They describe different work, and a team that judges its own customer survey against the telephone polling figure will conclude it is doing four times better than it is.
What the rate tends to look like by survey type
Rather than quoting a benchmark that will not match your denominator, it is more useful to know which direction each survey type pulls, and what the rate is most sensitive to. The table below is a guide to what to expect and where to look when the number disappoints.
| Survey type | Who is asked | Usual denominator | What moves the rate most |
|---|---|---|---|
| Employee or internal survey | A known, listed population with an organisational reason to answer | Headcount invited | Whether last year's results visibly led to anything |
| Post-interaction feedback, such as after support or delivery | A customer, moments after the event | Interactions that triggered a request | Delay between the event and the request |
| Customer satisfaction by email | A list of existing customers | Emails delivered | List age and how often that list is asked |
| In-product or website intercept | Whoever happens to arrive | Sessions where the prompt appeared | Placement and how many questions are visible at once |
| Event registration follow-up | People who already chose to attend | Attendees, not registrants | Whether attendance was confirmed |
| Cold or general population research | Strangers | Sample drawn, or contacts attempted | Channel, incentive and callbacks |
| Academic or panel study | Recruited or named participants | Eligible participants | Reminder schedule and incentive design |
Read down the middle column first. Every row has a different denominator by nature, which is why the averages published for these categories cannot be laid side by side. A 30 percent internal survey and a 30 percent intercept survey are not comparable achievements, and a team running both will get more from tracking each against its own history than from arguing about which is healthy.
The right-hand column is where the gains are. In most of these rows, the single biggest lever is not the questionnaire at all. It is timing, list quality, or whether people believe answering leads anywhere.
The two things that matter more than the percentage
A response rate is a proxy. What it is standing in for is whether the people who answered resemble the people who did not, and that question deserves direct attention.
Nonresponse bias is the real risk, and a high rate does not remove it. If the 20 percent who answered are disproportionately the delighted and the furious, the middle is missing and the average sentiment is meaningless no matter how the percentage looks. This is checkable without extra research. Compare the answering group with the full list on attributes already known: plan, tenure, region, purchase volume, how long they have been a customer. Where the distributions diverge, say so when the results are presented. A survey reported with its known skew is more useful than one reported with a flattering percentage and no caveat.
The absolute number of usable answers decides what can be said. Two hundred answers from a list of 10,000 is a 2 percent rate and enough to see a clear pattern in a simple question. Eight answers from a list of 12 is a 67 percent rate and cannot support a percentage at all, because each answer moves the result by 12 points. Small populations should be reported as counts, not shares. Six of eleven managers said the process is slow is a true and usable sentence. 54.5 percent of managers said so invites a confidence nobody has earned.
When both of those are handled, the response rate becomes what it should be: a diagnostic for the invitation process, not a grade for the project.
How to build a benchmark that actually applies
The only average worth measuring against is the one produced by the same survey, to the same list, using the same definition. Two cycles is enough to start.
- Write the definition down once. What counts as a response, what counts as asked, whether partials are included, whether screened-out cases are removed from the denominator. Keep it next to the results.
- Record the denominator at the moment of sending, not afterwards. Delivered counts and list sizes drift, and a number reconstructed a month later will not match.
- Log the conditions alongside the rate. Channel, day, time, subject line, number of questions, whether a reminder went out, whether an incentive was offered. Without these, a change in the rate has no explanation attached.
- Split the rate by segment. An overall figure of 18 percent that is 34 percent among recent customers and 4 percent among those older than two years is two findings, not one, and only the split version suggests an action.
- Compare like periods. Year on year beats month on month for anything with a seasonal shape, and internal surveys almost always have one.
After two or three runs, the question changes from what is the average survey response rate to why did this run differ from the last one, which is answerable and actionable. Response rate reporting only becomes useful at that point.
What to do with the answers once they arrive
There is a quiet reason response rates fall on repeat surveys, and it has nothing to do with survey design. People stop answering organisations that never respond. When a customer writes a paragraph of specific feedback and hears nothing for three weeks, the next invitation gets ignored, and the rate drops for reasons no amount of subject line testing will fix.
That makes the handling of responses part of the response rate problem. A submission that arrives in a shared inbox and a spreadsheet has no owner, no state, and no record of whether anyone replied. Tools built around response management attach an owner and a stage to each individual answer and keep the reply on the same screen as the answer, so open-ended feedback can be acknowledged and closed out rather than aggregated and forgotten. The typical uses where this pays off first are the ones where answers need individual follow-up: applications, support requests, and any feedback form where someone might have asked a direct question.
Acknowledging respondents is also the cheapest response rate intervention available, because it is the one that compounds across every future survey to the same list.
What to change first
Before hunting for a benchmark, fix the definition: decide what counts as a response and what counts as asked, and record both at send time so the next run is comparable. Then check whether the people who answered look like the people who did not, and report the gap rather than the percentage alone. If open-ended answers currently go unanswered, closing that loop does more for the next survey's rate than any redesign, and a tool like Halict that keeps the reply beside the response is what makes it routine.
Q1. Is a 10 percent response rate too low to use?
It depends on the population size and the question. Ten percent of 5,000 is 500 answers, which is plenty for spotting a clear pattern. Ten percent of 60 is six answers, which should be reported as six comments rather than as a percentage. The bigger issue is who those respondents are, so compare them with the full list on the attributes already known before deciding whether the result is usable.
Q2. Should partial answers count as responses?
Either choice is defensible as long as it never changes between survey runs. Counting them raises the rate and mixes complete and incomplete data in the analysis. Excluding them keeps the dataset clean and produces a lower figure. The common approach is to exclude partials from the response rate and report the completion rate separately so both effects stay visible.
Q3. Why does the same survey report two different response rates in two tools?
Because the tools are using different denominators. One may divide by invitations sent, another by invitations successfully delivered, and a third by people who opened the invitation or viewed the first question. Check which denominator each tool uses before treating the figures as a trend, and pick one definition to report.
Q4. Do incentives reliably raise the response rate?
They usually raise participation, and they change who participates. An incentive attracts people motivated by the reward rather than by the subject, which can dilute the feedback quality even while the percentage improves. If an incentive is used, keep it consistent across runs so the rates stay comparable, and watch whether the content of the open-ended answers gets thinner.
Q5. How often can the same list be surveyed before the rate collapses?
There is no fixed interval, but fatigue is driven more by perceived pointlessness than by frequency. A list asked monthly that sees changes attributed to its feedback holds up better than a list asked twice a year that never hears back. Before shortening the interval, make sure the last round produced something respondents were told about.