Speak, Don’t Type: Voice Surveys for Practitioner Research
Put an open text box into an online survey and you will mostly get very little back. A few keywords, half a sentence, often nothing at all. Yet that box is exactly where someone could explain why they ticked a 6 on the scale. Voice surveys start at this point.
A survey with a record button
A voice survey is an online survey in which respondents answer open questions by speaking. Depending on the settings, a recording can run up to ten minutes. Scales, multiple-choice items and filters work as in any other online survey. Anyone who would rather not speak can type.
In the talk I used Voiceform. The path from voice to dataset is short there: you ask a question, the person records an answer on their phone, the tool transcribes it, you export. What comes next, no tool does for you. You get text. The analysis is still research work.
Where the method fits
A questionnaire shows distributions and lets you compare across many people. It rarely tells you why someone ticks a 6. The text box underneath adds a few words to the score and often stays empty. In an interview you can ask follow-up questions and build rapport, and you pay for that in time, one person at a time.
Voice surveys sit in between. They collect reasons and experiences from the field, from many participants at once. What they lack is the follow-up question in the moment.
For practitioner research and action research, that position works well. Both depend on experience that arises close to practice, and both need feedback between cycles, when there is no time for another round of interviews. A teacher who wants to know how her students experienced a new teaching phase gets more from one spoken question on their phones than from a rating scale, with far less effort than twenty conversations.
Two uses that are easy to overlook
The first is the think-aloud protocol. Participants voice their thoughts while doing something else: testing a service, working through a task, or right after a situation in the field. Because the recording is asynchronous, you need no appointment and nobody sitting next to them.
The second addresses the missing follow-up. Voiceform can respond to an answer with an AI-supported follow-up question that you define in advance, for example asking for an example or a reason. That does not replace an interview. It does often get a second sentence out of a short answer.
What the data show
In a randomised study on academic procrastination, we compared the two response modes (Froehlich & Jahn, in review). Allocation ran 3:1 in favour of the text version. 98 people started the survey, 70 in text mode and 28 in voice mode.
Spoken answers averaged 97.5 words, typed ones 21.7. That is about four and a half times as long. Each answer contained 8.07 topics instead of 3.36. And the voice version took less time to complete: 5:08 minutes against 5:57.
Another figure interests me more. Of 277 topics, 66 % appeared in both modes. The students talked about largely the same things. What the response mode changed was mainly how much they said about them. The voice group is small and the study is still under review, so I read these numbers as a clear signal that still needs confirming.
Here is what that looks like in practice, a spoken answer from the study (translated from German):
“If I still have lots of time, I put it off until it’s really urgent. […] Sometimes it’s less about uni stuff and more about appointments and that social side of things. Maybe also because you’re worried something won’t work out.”
In a text box, that would probably have read “time management”.
What to settle before your first real use
Transcripts contain errors, most often with technical terms, names and dialect. Plan time to check them against the recordings, at least in samples.
A voice is personal data. Before real people answer, consent, data processing agreements and deletion periods need to be in place. With Voiceform it pays to read the terms of service closely, especially the passage on the provider’s use of de-identified data. And if a grade, a job or a relationship depends on taking part, you have to take voluntariness seriously.
Always offer a text alternative. It lowers the threshold for everyone who cannot or does not want to speak: in an open-plan office, on the tram, across a language barrier.
And a bad question stays a bad question, spoken or typed.
Free Workbook
The workbook takes you from your own research question to a pilot you can stand behind. It includes the checklist for data protection and consent and space for your first question.