How Long Do People Talk When a Form Asks a Question? A Median of 38 Seconds
We measured 396 voice answers: a median of 38 seconds for an English open question, 62 in people's own language, and 10 for reading a sentence aloud.
The short version
- In a random sample of 138 recordings, people answering an English open question in a form spoke for a median of 38.2 seconds, and half of answers fell between 29 and 53 seconds.
- Asked a different question in their own language with a longer cap, they spoke for a median of 61.7 seconds, with a long tail: 1 in 10 answers ran past 157 seconds.
- Reading a set sentence aloud took a median of 10.3 seconds across 120 sampled speech-data recordings, which is why mixed averages of 'voice answers' mislead.
- Only 9 of 138 English answers ended within 2 seconds of the 2-minute cap, so a 2-minute limit rarely cut anyone off on that question.
Thirty-eight seconds. That is the median length of a spoken answer to an open English question in a form, measured across a random sample of 138 real recordings on Would You Rather. Ask people to explain something in their own language with a longer limit and the median rises to 62 seconds. Ask them to read a sentence aloud and it falls to 10.
Voice answer length depends on the task far more than on the format. Below are the distributions, what they mean for setting a time limit, and the caveats that keep these numbers from being universal benchmarks.
In our data, people spoke for a median of 38 seconds when a form asked an open question in English, with half of answers between 29 and 53 seconds. In their own language, with a longer limit, the median was 62 seconds. Reading a set sentence took about 10. Set your cap at roughly twice the answer you expect.
How we measured 396 recordings
Would You Rather stores voice answers as audio files and does not transcribe them, so length here means seconds, not words. On 8 October 2026 we drew a random sample of 400 recordings from the production database. We downloaded each one with a short-lived read-only link, decoded the full audio to measure its true duration, and deleted the file straight away. No recording was kept or listened to for this post.
We decoded instead of reading the file header because browser recordings often carry unreliable duration metadata. Four files would not decode and were dropped, leaving 396. The sampling window runs from the first stored answer on 18 April 2026 to 8 October 2026.
We report medians and percentiles rather than averages throughout. A handful of very long recordings pull the mean up: on the English question, the mean was 45.5 seconds against a median of 38.2. The median is closer to what a typical respondent did, and it is the number you should plan around.
Each recording in a group had the same chance of selection. Because each applicant answers each question once, the open-question samples are close to one recording per person. The script sample is different: a single speaker can record hundreds of sentences, so heavy contributors are more likely to appear in it.
Three groups, sampled separately
We split the sample by task, because mixing them would produce a meaningless average. Two groups come from one application form that 918 people completed. The third comes from 31 speech-data forms.
| Group | Sampled | Valid | What the respondent did | Allowed range |
|---|---|---|---|---|
| English open question | 140 | 138 | Explain a technical idea in 3 to 5 sentences | 12 to 120 seconds |
| Own-language open question | 140 | 138 | Explain an everyday process step by step | 12 to 500 seconds |
| Script reading | 120 | 120 | Read or spell a given sentence aloud | Varied by form, 30 to 180 seconds |
Almost every response in our database belongs to forms owned by one account, which runs recruiting and speech-data projects. Respondents to the application form were job applicants. Keep that in mind: people trying to impress an employer may talk longer than customers giving feedback.
The answer to an English open question: 38 seconds
The English question asked applicants to explain a concept to someone who had never heard of it. The median answer ran 38.2 seconds, half of answers fell between 28.9 and 53.3 seconds, and 1 in 10 ran past 82.5 seconds.
The shape is a hump with a short right tail. Most people said what the question asked and stopped. Very few rambled toward the limit, and very few gave a token answer.
The 2-minute cap rarely bit
Only 9 of 138 answers ended within 2 seconds of the 120-second limit. That is about 1 in 15. We read that as a cap doing its job: it signals how much effort is expected without cutting most people short. The 12-second minimum also barely mattered, with only 3 answers ending close to it.
Self-rated fluency didn't change length
The form asked applicants to rate their English from 1 to 10 before the voice question. Answer length was flat across ratings: a median of 38.9 seconds for ratings of 6 or below, 37.9 for 7 to 8, and 38.6 for 9 to 10. The low-rating group held only 14 recordings, so we would not lean on it. It does suggest the question, not confidence, set the length.
In their own language: 62 seconds and a long tail
The second spoken question asked people to explain a step-by-step everyday process in their own regional language. It allowed up to 500 seconds, a little over 8 minutes. The median answer was 61.7 seconds, the middle half ran from 40.2 to 111.8 seconds, and 1 in 10 ran past 157 seconds.
This distribution is much wider. 30 of the 138 answers ran past 2 minutes and 8 ran past 3. A process question invites detail, and a generous cap lets the detail-minded keep going. The mean, 85.5 seconds, sits well above the median because of that tail.
- 25th percentile
- Median
- 75th percentile
Is it the language or the question?
It is tempting to say people talk longer in their own language. We can't claim that from this data. The two questions differed in topic, in the kind of answer they invited and in their cap. Any of those could explain the gap. By language chosen, Hindi answers had a median of 57.4 seconds (67 recordings) and Tamil 68.9 seconds (27). The other languages had too few recordings to compare.
We still think the own-language result is the more useful number for anyone designing a multilingual form. It shows what happens when you ask a story-shaped question and get out of the way. Our guide to collecting answers in Hindi, Tamil or any language covers the wording side.
Reading a sentence: 10 seconds
Speech-data forms ask people to read set sentences aloud. Those recordings had a median of 10.3 seconds, and 9 in 10 were under 23 seconds. 89 of the 120 were under 15 seconds.
The length here is set by the sentence, not the speaker. That is why we report it separately. Most of the voice answers on our platform are this kind, and blending them into an "average voice answer" would drag the figure toward reading length and tell you nothing about open questions. We explain that composition problem in what 6,274 voice answers on our platform actually were.
The cap didn't set the reading length
The script forms used three different caps: 30, 120 and 180 seconds. Median reading time rose with the cap, from 6.4 seconds on 30-second forms (44 recordings) to 11.2 on 120-second forms (27) and 15.4 on 180-second forms (49). That looks like a cap effect, but the longest reading in the whole sample was 35.4 seconds, far below the larger caps. Nobody was filling the time.
The likelier explanation is that the form creators set longer caps for longer sentences. The cap followed the script, not the other way round. For speech-data work, we think the cap should simply be the longest sentence's comfortable reading time plus a margin for one restart. Anything more only invites background noise at the end of the clip.
What length means in words, roughly
We did not transcribe these recordings, so we can't give measured word counts. Two published figures frame the question. In a randomised smartphone experiment by Höhne, Gavras and Claassen (2024), spoken answers to open questions averaged 54.7 words against 21.5 typed. A lab study found English speech input ran at 153 words per minute.
Open answers include thinking time, pauses and restarts, so they run slower than dictation. We would not multiply our seconds by a dictation rate and call the result a word count. If you need words, transcribe a sample of your own recordings and count. For which engine to use, see our voice survey guide.
How to set the time limit on a voice question
Use the data above as a starting point, then adjust for your audience. The cap does two jobs: it stops runaway recordings, and it tells the respondent how much you are asking for. A cap that is too long can feel like homework.
| Question type | Expect a median of | Suggested cap | Why |
|---|---|---|---|
| Read a sentence aloud | About 10 seconds | 30 seconds | Leaves room for a retake without inviting chatter |
| Focused open question | About 40 seconds | 90 to 120 seconds | Our 2-minute cap cut off about 1 in 15 answers |
| Explain a process or tell a story | About 60 seconds, long tail | 3 to 5 minutes | Long answers are the point, but say so up front |
| Where typing wins | A fact, a name, a number | No voice question | A tap or short text is faster for everyone |
Say the limit in the question
"Up to 2 minutes" in the question text costs four words and removes a guess. Pew Research Center found typed open questions asking for several sentences were skipped more often than one-word questions. A visible limit is how you tell people the effort is bounded. In Would You Rather, each voice question has its own minimum and maximum seconds, and the recorder shows the limit before recording starts, with a ring that fills as time runs.
Keep minimums short
A minimum stops a stray tap from submitting silence. Our largest form used 12 seconds and almost nobody stopped near it. We would not go much higher. A long minimum forces padding from people who have already finished, and that is worse data, not more.
What these numbers can't tell you
Our biggest limitation is who answered. The open-question recordings come from job applicants on one form owned by one account. Applicants have a strong reason to finish and to sound thorough. Customers, students or survey panellists may speak for less time, or more.
Second, duration includes silence. A long pause before speaking, or background noise left running, counts as length. Third, these are samples of 120 to 140 recordings per group, not every file, so the percentiles carry some sampling error. Fourth, we excluded nothing for quality: a recording of mostly silence counts the same as a fluent answer.
Finally, the comparison between English and own-language answers is confounded, as covered above. Treat it as a reason to test, not a result.
Where to start
Take your next form with a spoken question. Write the question so it asks for one thing, set a cap at about twice the length you expect, and state the cap in the question. After the first 30 answers, look at the durations in your export. If more than 1 in 10 hit the cap, raise it. If almost nobody gets past a third of it, shorten it or make the question more specific.
If you collect speech data, plan per sentence rather than per respondent, and read our guide to building a speech dataset for ASR. If you are screening candidates, our list of voice screening questions for hiring pairs each prompt with a sensible limit.
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 spoken vs typed words per answer.
- Ruan, S., Wobbrock, J. O., Liou, K., Ng, A. and Landay, J. A., Speech Is 3x Faster than Typing for English and Mandarin Text Entry on Mobile Devices, arXiv, 2016. Used for speech and keyboard entry rates.
- Asare-Marfo, D., Why do some open-ended survey questions result in higher item nonresponse rates than others?, Pew Research Center Decoded, 2021. Used for the effect of requested answer length on skipping.
- Would You Rather, random sample of 400 voice recordings, decoded for duration and deleted, 8 October 2026. Used for every duration figure. Aggregated and anonymised; not independently audited.
The weakest point in this source base: the open-question durations come from job applicants on a single form, and no published study we found reports spoken answer durations in seconds for web forms.
Questions people ask
How long should a voice answer be in a survey?
For a focused open question, plan for 30 to 60 seconds. In our sample of 138 English answers, the median was 38 seconds and half fell between 29 and 53 seconds. Set the cap at about twice the answer you expect, so it rarely cuts anyone off, and state the limit in the question so people know what effort you are asking for.
What time limit should I set on a voice question?
Use 60 to 120 seconds for most open questions and say the limit in the question text. A 2-minute cap on an English explanation question cut off only 9 of 138 sampled answers in our data. Use longer limits only when you want a story or a process explained, and expect a long tail of answers past 2 minutes.
Do people talk longer in their own language?
Possibly, but our data can't prove it. On one form, people spoke for a median of 62 seconds when asked to explain a process in their own regional language, against 38 seconds for an English question. The two questions also asked different things and had different caps, so the gap may reflect the task rather than the language.
How many words is a 1-minute voice answer?
It depends on speaking rate and pauses, and we did not transcribe our recordings, so we can't give a measured figure. In a lab study, people dictating short English phrases into a phone reached 153 words per minute, but open answers include thinking time. Transcribe a sample of your own recordings if you need a reliable words-per-minute figure.
Should a voice question have a minimum length?
A short minimum, such as 5 to 15 seconds, stops accidental taps from submitting empty audio. Our largest open form used a 12-second minimum, and only 3 of 138 sampled English answers ended within 2 seconds of it. Avoid long minimums: they push people to pad answers or quit, and they add effort to a step that already loses respondents.
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