Survey Response Rates: Check What Was Counted

A survey can have a 30% response rate and a 40% response rate without a single answer changing. Read two vendor guides and you can arrive at both figures, using the same people and the same study.

SurveyMonkey’s July 28 guide to survey response-rate benchmarks defines response rate as people who started a survey divided by people who received the invitation. It separately defines completion rate as people who finished divided by people who started. That gives you two measures: how many recipients began, and how many starters reached the end. But Typeform’s April 9, 2026 guide defines response rate as completed surveys divided by people who received the survey. The same label refers to a different calculation.

I think a response-rate comparison is only usable when it includes the counts, the formula and who was invited through which channel.

One survey, three percentages

Take a clearly hypothetical study. You have 100 invitation recipients. Of those people, 40 start the survey and 30 finish it.

Following the definitions in those guides, you get:

• Typeform’s response rate: 30 finishes ÷ 100 recipients × 100 = 30%.
• SurveyMonkey’s response rate: 40 starts ÷ 100 recipients × 100 = 40%.
• SurveyMonkey’s completion rate: 30 finishes ÷ 40 starts × 100 = 75%.

All three calculations describe something real about this hypothetical study. The trouble starts when someone copies just “response rate” and a percentage into a report.

Suppose you use the Typeform calculation this month and the SurveyMonkey calculation next month. You could report an increase from 30% to 40% even if the underlying counts were identical. A colleague might credit a better subject line. You might decide to keep an incentive that made no difference. The apparent improvement came entirely from changing what counted as a response.

I can see the appeal of a single percentage. It fits neatly into a weekly update. But I would rather read a slightly untidy sentence that tells me what happened than a clean number that leaves me guessing.

Where I agree with SurveyMonkey

Separating starts from finishes is useful. If invitation recipients rarely begin, I would look at who received the invitation, whether they had a reason to answer and how the invitation reached them. If people begin but leave before finishing, I would inspect the survey itself, including its length, questions and technical behaviour.

Those are places to investigate, not diagnoses. A low start rate does not prove that your subject line failed. People leaving halfway through does not prove that the survey was too long.

Still, the distinction gives you a better starting point than one combined figure. On that, I agree with SurveyMonkey’s approach.

Where I part company is with treating any vendor’s shorthand as a definition that travels safely between tools. SurveyMonkey’s starts-based calculation and Typeform’s finishes-based calculation should be named explicitly when you use them. Neither guide’s shorthand should stand in for a universal methodological standard.

For a founder, the practical consequence is straightforward: before judging your survey against a benchmark, check that both figures describe the same event. I would not use a headline benchmark as a target until I could see its calculation and the circumstances behind it.

What to put beside the percentage

For a small customer study, you do not need a sprawling reporting system. You need enough detail that someone else can reconstruct the calculation.

I would keep a short methods note with:

• Audience and recruitment channel. Who qualified, and how did the invitation reach them?
• Counts. Invitation recipients, starts and finishes, where those counts are known.
• Explicit formulas. Name the numerator and denominator beside each reported percentage.
• Counting rules. Explain what counts as a start or finish, and how you handled screening exits, duplicates and excluded responses.

That last point can seem fussy until you try to compare studies. Opening a survey link and answering the first question may be recorded as different events. Someone who fails a screener has a different reason for leaving than someone who abandons an eligible survey halfway through. Preserve those distinctions where you can.

Also check whether “received” actually means received. An email-send count may include messages that bounced. If all you know is how many invitations you sent, label the denominator as invitations sent. Do not quietly rename it recipients.

A public link introduces another limit. You might know how many people started and finished without knowing how many people saw an invitation. In that case, report the counts you have and the finish-to-start calculation. Leave the invitation-based rate unreported rather than filling in a denominator you cannot observe.

Audience and recruitment channel

Even matching formulas do not make every comparison useful.

A survey emailed to recent repeat buyers reaches people with an existing relationship to your business. A recruited panel study may reach eligible category shoppers who have never heard of you. I would expect those groups to have different reasons for participating, so I would not treat a difference in their response rates as evidence that one survey was better designed.

Record the recruitment context before comparing results. Include eligibility rules, any incentive and how long the study stayed open. If those changed, acknowledge the change alongside the percentage.

At Probe, you can send a study to your own audience, with own-audience studies free up to 10 questions and 25 responses. Or we can recruit participants from our recruitment partner’s vetted panel, with one all-in study price shown before launch. Those routes serve different recruitment needs. They do not make the resulting audiences interchangeable.

Who did not answer?

Once the arithmetic is clear, there is a separate question: could the people who did not answer differ in ways that affect your decision?

The Census Bureau’s September 15 post on evaluating nonresponse bias in the 2026 CPS ASEC discusses using administrative data to investigate nonresponse bias. I would take that as a prompt to ask about missing respondents, not as evidence for a particular size or direction of bias in your customer survey.

Imagine you are researching reasons for cancellation, but your invitation only reaches customers who still open your emails. You have a reason to ask whose experience is missing. A correctly calculated response rate cannot answer that by itself.

Where you have appropriate customer records, compare respondents with the invited audience on relevant characteristics you already know, such as recent purchase activity or subscription status. That will not tell you everything about people who stayed silent, but it can reveal gaps you should investigate before acting on the answers.

Before your next survey goes out, write the reporting sentence with blanks: “We invited ___ eligible people through ___; ___ started and ___ finished. We calculated this percentage as ___ divided by ___.” If you cannot fill in one of those fields, decide now whether you can collect it or need to report the limit.