Voice vs Text Questions: Which Gets Skipped More? The Bigger Leak Was Lost Recordings
Optional typed questions were left blank 6.5% to 9.8% of the time on our forms. Required voice answers went missing 11.6% of the time, until we fixed uploads.
The short version
- On our largest open-question form, optional questions were left blank 6.5% (a tap question, 59 of 904) and 9.8% (a typed transcription task, 87 of 890) of the time.
- Required voice answers still went missing: before 1 July 2026, 74 of 636 completed applications arrived without at least one required recording, and 153 without the required resume file.
- After we uploaded recordings as they were made and blocked Next until each one saved, missing required answers fell to 0 of 282.
- Published research says voluntary voice questions are skipped far more than text: about 35% against under 2% in one 2024 experiment, and Pew reports open text questions average 18% non-response.
The answer you've heard is that voice questions get skipped more, and in published experiments that is true. In our own data, the bigger problem was different and more fixable. Before 1 July 2026, 11.6% of completed applications on our busiest form arrived missing a recording the respondent was required to give. The voice vs text skip rate debate hides a third category nobody talks about: answers that were given and never arrived.
Skipped, missing and quit are three different failures, and they need three different fixes. We separate them below using read-only queries on Would You Rather's production database, with published research for the comparison our data cannot make on its own.
When skipping is allowed, voice questions are skipped far more than text ones: about 35% against under 2% in a 2024 experiment. On our forms, voice questions were required, so the real leak was technical. Required recordings went missing in 11.6% of completed applications until we uploaded each one as it was made.
Three ways an answer goes missing
Most discussions lump every blank cell together. That hides the cause, and the cause decides the fix. We use three terms throughout this post.
| Failure | What happens | Who causes it | Typical fix |
|---|---|---|---|
| Skipped (item non-response) | Respondent finishes but leaves a question blank | The respondent, by choice | Better wording, optional typed fallback |
| Missing | Respondent answers but the answer never reaches you | The tool or the network | Upload early, confirm each save |
| Quit (break-off) | Respondent leaves the form partway | The respondent, reacting to burden | Order, effort, time limits |
| Where text still wins | Short facts are rarely skipped in any format | n/a | Keep facts as taps or short text |
This post covers the first two. Break-off has its own analysis in where respondents quit a form by question type, and the two are linked. A required voice question can't be skipped, so people who would have skipped it quit instead.
What we measured, and on which forms
All first-party figures come from Would You Rather's production database, queried on 8 October 2026, covering answers from 18 April to that date. We publish aggregates only. We excluded 4 test forms, single-response forms titled as tests or left untitled.
The skip analysis rests almost entirely on one form. It is an application form for language freelancers that 918 people completed. It mixes contact fields, tap questions, short typed answers, a resume upload, two required spoken questions and one optional typed transcription task. One account owns the forms behind 1,656 of our 1,660 responses, so read every number here as one heavy user's experience.
Our voice questions were required
This is the main limit on the comparison. Every spoken question on that form was required. Respondents could not skip them through the normal flow, so we cannot measure how often people would skip voice by choice. What we can measure is how often a required answer went missing anyway, and how often optional questions were left blank.
Optional questions: what people chose to leave blank
Two questions on the form were optional. One was a single-choice tap question about notice period, which 59 of 904 completed applicants left blank, or 6.5%. The other was a typed task: listen to a short clip in your chosen language and transcribe it. 87 of 890 eligible applicants left the transcription blank, or 9.8%.
That is a heavy typed task, a median of 89 words, sitting at the very end of a long form. A skip rate under 10% is low for that. We think the reason is motivation: applicants wanted the work, and the task looked like part of the test. Customers answering a feedback survey would almost certainly skip more.
Required questions: answers that never arrived
Here is the finding that surprised us. Before 1 July 2026, 636 people completed the application form. 74 of them, 11.6%, arrived without at least one of the two required recordings, and 60 were missing both. The required resume upload was worse: 153 completed applications, 24.1%, arrived without the file.
Those were not choices. The questions were required, and the respondents went on to finish the form. The pattern, both recordings or the file missing together, points at the plumbing. At the time, recordings and files uploaded only when the respondent pressed submit, and the server did not re-check required questions.
- Both recordings arrived562
- One recording missing14
- Both recordings missing60
The fix was moving the upload earlier
On 1 July 2026 we changed two things. Each recording now uploads as soon as the respondent stops recording, and the Next button waits until the answer is confirmed saved. If the save fails, the respondent stays on the question and sees a message asking them to tap Next again. The server also re-checks required questions before accepting a submit.
Since then, none of 282 completed applications has arrived without a required recording or the resume file. We can't prove the change caused the whole drop, because other things changed over the same months. The before and after is stark enough that we believe it did most of the work.
We would not trust any voice-form vendor's completion numbers without asking this question first: when your tool says a response is complete, is every recording actually stored? A form that accepts a submit before the audio lands will look fine in the dashboard and fail in the spreadsheet.
What research says about voluntary voice skips
Because our voice questions were required, the voluntary comparison has to come from published work. The strongest is a 2024 experiment by Höhne, Gavras and Claassen in Social Science Computer Review. They randomly assigned 1,001 German smartphone respondents to answer open questions by typing or speaking.
Among people who stayed to the questions, roughly 35% of the voice group gave no answer to a given question. For text, the figure was under 2%. The voice group also broke off far more before reaching the questions, 51% against 24%. Voice is easy to skip when skipping is free.
Typed open questions are skipped too
Text is not immune. Pew Research Center analysed 92 open-ended questions across more than 40 of its panel surveys from 2018 to 2021. Typed open questions averaged around 18% non-response, against 1% to 2% for closed questions, with individual questions ranging from 3% to just over 50%.
Pew's analysis found three things that push skips up. Non-response was higher on mobile than desktop. It was higher for questions asking for several sentences than for one word. And it was highest for questions with a heavy thinking burden. Those are design levers, and they apply to voice questions as well.
Speech-data forms: people don't skip, they leave
Our script-reading forms behave differently again. These forms ask people to read sentences aloud for a speech dataset, often many in a row. Every recording is required, because the recording is the product. We looked at all 31 such forms with at least one response, 116 responses in total.
Of 85 completed responses, 83 had a recording for every single voice question. The two exceptions were very long forms, with several hundred sentences each, where some sentences were left unrecorded. Among people who finished, missing recordings were close to zero.
The loss happens before the first recording
The failure shows up in the abandoned responses instead. 31 responses on script-reading forms were never completed. In 22 of them, not one recording was saved, even though the respondent had already filled in their contact details. They reached the microphone and left.
That is a small sample, and these forms are often sent to a handful of speakers each, so treat it as a pattern to watch rather than a rate. It matches what the voice experiment found: the biggest cost of a voice question is paid at the moment people first see it. When skipping isn't allowed, the people who would have skipped become people who quit.
Why that matters more than a blank cell
A skipped question costs you one answer from one person. A quit costs you every answer that person would have given afterwards, including the easy ones. If you must choose, a form that lets people skip the microphone is usually better than a form that loses them at it. We think the exception is speech data, where a response without audio is worth nothing anyway.
What a skip does to your results
Skips matter for more than volume. If the people who skip a question differ from those who answer, the answers you keep are biased. Pew's analysis found non-response varied with device and with the effort a question demanded. That means the skippers are not random.
For voice, the plausible skippers are people in public places, people who dislike hearing their own voice, and people less confident speaking the survey's language. Each of those groups can hold a different view. A voice-only question that 35% of respondents skip may tell you mostly about the 65% who were comfortable talking. A typed fallback keeps more of the room in the data.
How to tell a skip from a lost answer
You can usually tell the two apart from the data alone. Skips are scattered: different people leave different optional questions blank, and the rest of their response looks normal. Lost answers cluster: the same response is missing several uploads at once, often every file in it, and often within the same days.
In our pre-July data, 60 of the 74 affected applications were missing both recordings, and 70 of the 74 were also missing the resume file. That clustering is what made us look at the upload code rather than the question wording. Run the same check on your own export before you rewrite a question that people may never have had the chance to answer.
The limits of what our data can settle
Our numbers come from applicants, who had a reason to finish. They also come mostly from one form and one account, so the optional-question skip rates may not travel to customer surveys. Treat the 6.5% and 9.8% as one data point each, not as benchmarks.
What we could not separate
We also can't separate every cause of the July improvement. The form itself changed little, but the respondent mix and the season did. A cleaner test would run the old and new upload code side by side, which we did not do. Finally, we have no voluntary voice skip rate of our own at all, because no busy form in our data made a voice question optional.
How to keep voice answers from going missing
If you run voice questions, deal with the missing problem before the skip problem. It is cheaper to fix and it hides inside "completed" responses.
- 1Submit a test response on a bad connectionUse your phone on mobile data in a weak spot. Then open your results and play every recording. A blank cell here is a lost answer later.
- 2Confirm uploads happen per questionThe recording should save when the respondent finishes it, not when they press submit at the end.
- 3Decide required or optional on purposeRequire voice only when the recording is the deliverable. Otherwise make it optional or add a typed box beside the microphone.
- 4Put the spoken question after the easy onesContact details and taps first. People who have invested a few answers are less likely to abandon at the microphone.
- 5Count blanks per question after launchCheck your first 50 responses for blank voice cells and blank optional fields. Investigate any required question with blanks.
In Would You Rather, the typed alternative is a per-question switch on the voice field, labelled "or type your answer" for respondents, and it is off by default. Turn it on for any audience that might be on a bus. For the wording side, see why the 'any other comments' box stays empty and what happens when open-ended answers must be typed. For the bigger case for voice, start with our guide to voice surveys.
References
- Höhne, J. K., Gavras, K. and Claassen, J., Typing or Speaking? Comparing Text and Voice Answers to Open Questions on Sensitive Topics in Smartphone Surveys, Social Science Computer Review, 2024. Used for voice and text item non-response and break-off.
- Asare-Marfo, D., Why do some open-ended survey questions result in higher item nonresponse rates than others?, Pew Research Center Decoded, 14 October 2021. Used for open and closed question non-response and its drivers.
- Would You Rather, production database, read-only queries, 8 October 2026, and the product's commit history for the 1 July 2026 upload change. Used for every first-party figure. Aggregated and anonymised; not independently audited.
The weakest point in this source base: our skip and missing rates come from one application form whose respondents had a strong reason to finish, and the voluntary voice figure rests on one German experiment.
Questions people ask
Do people skip voice questions more than text questions?
When skipping is allowed, yes. A 2024 randomised experiment found about 35% of voice respondents gave no answer to a given open question, against under 2% for text. Pew Research Center reports that typed open-ended questions average around 18% non-response on its panel, against 1% to 2% for closed questions. Making the voice question required, with a typed alternative, changes the picture.
What is item nonresponse in a survey?
Item nonresponse is when a respondent completes a survey but leaves a particular question unanswered. It differs from unit nonresponse, where someone never responds at all, and from break-off, where they quit partway. Item nonresponse matters because the people who skip may differ from those who answer, which can bias the results for that question.
Should voice questions be required?
Make a voice question required only if the recording is the point of the form, as in speech-data collection or a speaking test. For feedback or research, make it optional or offer a typed box beside the microphone. A required voice question with no alternative pushes people who can't talk right now to quit the whole form rather than skip one question.
Why are some voice answers missing even when the question is required?
Usually the cause is technical, not behavioural. The recording may fail to upload on a weak connection, the browser may block the microphone, or the form may accept a submit before the upload finishes. Check that your tool uploads each recording as soon as it is made and confirms the save before letting the respondent move on.
How do you reduce skipped open-ended questions?
Ask fewer of them, make each one easy to answer from memory, and place them after easy tap questions. Pew found that questions asking for several sentences and questions with a high thinking burden had the highest skip rates. Offering voice as an option, rather than a requirement, can help people who find typing a long answer on a phone tiring.
Drafted with AI from Would You Rather's own aggregated, anonymised product data and published research, then edited and approved by Shivam Singh.
Would you rather hear the answer?
Make a voice form in a minute. It’s free.
Create a form