Voice of Customer Surveys: Hear What Customers Actually Mean, Not Just What They Type

Most voice of customer programmes contain no voices. They contain scores, typed comments, star ratings and support tickets, all of it text. A voice of customer survey, taken literally, asks customers to speak, and the evidence says they then say about 2.5 times as much as they type.

Your VoC data is probably text all the way down. That is a fixable gap. Below we cover the four standard VoC methods, what each one hears and misses, where a spoken answer adds signal that text cannot carry, and a question bank you can paste into your next customer feedback form.

Voice of customer started as listening, then became typing

The phrase has an origin worth knowing. In 1993, Abbie Griffin and John Hauser published "The Voice of the Customer" in Marketing Science. They studied the first stage of Quality Function Deployment, the process product teams used to turn customer needs into engineering targets. Their questions were practical: how many customers to interview, how many analysts should read the transcripts, how many needs a team misses, and whether focus groups beat one-on-one interviews.

In that world, the voice of the customer was literal. Teams sat with customers, recorded what they said, and read the transcripts closely. The method was slow and expensive, and it produced depth.

Scale pushed VoC toward scores and text boxes

Then customer feedback moved online, and the cheapest data won. A 0 to 10 score costs a respondent two seconds. A typed comment box costs nothing to add. Over time, most VoC dashboards became a score trend with a word cloud underneath.

We think that trade went too far. A score tells you something moved. It rarely tells you why, and the typed comment under it is usually one line. We cover that empty-box problem in why the "any other comments" question stays empty.

The founder of NPS asked for reasons too

Even the most famous single-number metric was never meant to stand alone. In The Ultimate Question 2.0, Fred Reichheld and Rob Markey wrote that companies should ask at least one follow-up after the recommend score: "What is the primary reason for your score?"Reichheld and Markey, The Ultimate Question 2.0, Bain and Company advance excerpt, 2011, p. 4.. They described it as the way to hear customers' reasons in their own words. We go deeper on that question in designing the NPS follow-up question.

The 4 voice of customer methods, and what each one misses

Every VoC method is a trade between depth, scale and effort. None is complete on its own, and the honest comparison includes where the typed and scored options win.

The typed open question deserves credit. It is easy to search, cheap to analyse, and nobody has to find a quiet room to answer it. If your team will never listen to a recording, a text box is the better choice. We would still argue most teams underuse the bottom row.

What a spoken answer carries that a typed one cannot

Length is the obvious difference. The randomised experiment by Höhne, Gavras and Claassen found spoken answers averaged 54.7 words against 21.5 typed, and covered more topicsHöhne, Gavras and Claassen, Social Science Computer Review, 2024. N = 1,001 smartphone respondents, four open questions on sensitive topics.. We analyse that study in full in our post on spoken answers, including its large drop-off cost.

The same paper reports a detail VoC teams should notice. It found no differences in sentiment between voice and text answers, once both were in text form. So the extra signal in a spoken answer is not hidden in more emotional words. It lives in the length, the extra topics, and the sound itself.

Tone carries emotion that listeners can decode

The science of vocal emotion is old and well replicated. Juslin and Laukka reviewed 104 studies of vocal expressionJuslin and Laukka, Psychological Bulletin, 2003, vol. 129, pp. 770 to 814. 104 studies of vocal expression and 41 of music performance. and found that discrete emotions were communicated to listeners through emotion-specific patterns of acoustic cues. In plain terms: how something is said carries information about how the speaker feels, and people can hear it.

For VoC, that matters in small, practical ways. "It's fine" said flatly and "it's fine" said with a laugh are different answers. A typed survey records them as the same string.

The same words, read flatly, lose the signal

The sharpest evidence comes from Michael Kraus at Yale. Across five experiments with 1,772 participantsKraus, American Psychologist, 2017, vol. 72, pp. 644 to 654. Five experiments, N = 1,772 in total., people judged others' emotions more accurately from voice alone than from video with sound, or from video alone.

His fifth experiment is the one we would show any VoC team. Listeners heard recorded conversations between friends. One group heard the real voices. Another heard a monotone synthetic voice read the exact transcript. Same words, no human tone. That group judged the speakers' emotions least accurately of all four conditions.

Two caveats belong next to that chart. The participants were online panel workers judging friends teasing each other, not customers talking about a product. And a 2018 comment by Rossiter in the same journal, "Voice-only does not meaningfully improve detection of emotions", argued the voice-only advantage is too small to matter in practice. The synthetic-voice gap was the largest in experiment 5, but treat the direction as the lesson, not the exact size.

Hesitation and confidence are audible

Listeners also read certainty from speech. Brennan and Williams studied prosody and filled pauses as cues to a speaker's state of knowing, the "um" before a guess. In customer research, the pause before "yes, I'd probably pay for that" is often the most honest part of the answer.

Hearing also changes how much credit listeners give the speaker. Schroeder and Epley found that evaluators rated job candidates as more competent and intelligent when they heard a pitch than when they read the same pitchSchroeder and Epley, Psychological Science, 2015, vol. 26, pp. 877 to 891. Hypothetical employers and professional recruiters, several experiments.. Their setting was hiring, not feedback. The lesson transfers: a customer heard is a customer taken more seriously in the room where decisions get made.

How to design a voice of customer survey

The design job is to collect stories without losing people. Spoken questions cost more effort than taps, so you spend them only where reasons matter.

On step 5: in our own builder, the text fallback on a voice question is a switch, and it is off by default. Turn it on for any customer audience. The reason is the drop-off evidence: in the Höhne experiment, about half the people assigned to voice-only answers broke off before the questions began.

Voice of customer survey questions: a question bank

Good VoC questions ask about an episode. "How do you feel about our onboarding?" invites a verdict. "Tell me about your first day using us" invites a story, and stories contain the details you can act on.

Ask about one episode, not an average

People are bad at averaging their own experience and good at telling one story. "Usually" questions get vague answers. "Last time" questions get names, steps and feelings. If an answer surprises you, the episode framing also makes it easy to check against your logs.

Keep scores and choices as taps

Do not ask people to speak a number. Ratings, plans and yes or no answers are faster to tap and easier to analyse. Voice is for reasons. A form that mixes the two well feels short, even when it collects a lot. For multilingual customers, invite them to answer in whichever language they think in; we cover that in collecting answers in Hindi, Tamil or any language.

Turning recordings into decisions

A folder of audio is not insight. Plan the analysis before you send the survey, because the cost is real and it lands on someone's calendar.

Under about 100 answers, listen to all of them. It takes an afternoon, and you hear what transcripts drop. Above that, sample: listen to 20 to 30 end to end, write themes, then skim the rest for those themes. Tag each answer with its theme next to the respondent's score, so you can see which reasons cluster among unhappy customers.

Be clear about what our own product does here. Would You Rather stores each recording next to the respondent's other answers, and can sync everything to a spreadsheet with audio links; see sending form responses to Google Sheets. It does not transcribe answers or score their sentiment. If you need transcripts for search, export the audio links and run them through a speech-to-text engine you trust.

Our view is that the listening is the point. A team that hears 20 customers explain a churn reason will act faster than a team that reads a bar chart of the same reason. That is a belief, not a measured result, and you can test it on your own team.

Where voice adds little, and where the evidence is weak

Voice is not always worth the cost. There are clear cases where a typed or scored question is the better tool, and the evidence base has gaps you should know about.

First, drop-off. Voice-only questions lose many respondents, and the people who leave may differ from those who stay. If your customers are mostly at desks in open offices, expect more typing than talking. Second, questions with short factual answers gain nothing from voice. "Which plan are you on?" should never be a recording.

Third, the research on tone comes mostly from lab studies of emotion between friends or strangers, not from product feedback. We are extrapolating from it, and you should treat the tone argument as plausible rather than proven for VoC. Fourth, analysis effort scales with volume. Ten thousand recordings with nobody to listen are worse than ten thousand typed lines someone can search.

Finally, privacy. A voice is identifying in a way a typed comment is not. Tell customers how recordings are stored and who hears them, and do not ask for voice where anonymity is the point.

How to run your first voice of customer survey this week

Pick one moment that matters this quarter: cancellations, first week, or renewals. Write one episode question for it, add two taps before it, and offer voice beside a text box. Send it to the next 50 customers who hit that moment.

Then hold one meeting where the team listens to 10 recordings together before anyone opens a dashboard. Note which themes come up and which surprised you. If you already run NPS, add the spoken reason question to it instead; it is the smallest possible change, and we explain why in our NPS follow-up guide. For the typing side of the argument, see why typed open answers come back short.

References

1. Griffin, A. and Hauser, J. R., The Voice of the Customer, Marketing Science, 1993. Used for the origin of the term and the questions the method addressed (abstract only).
2. 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 answer length, topics, the absence of sentiment differences, and break-off.
3. Kraus, M. W., Voice-Only Communication Enhances Empathic Accuracy, American Psychologist, 2017. Used for the five experiments and the experiment 5 synthetic voice results.
4. Rossiter, J. R., Voice-only does not meaningfully improve detection of emotions: Comment on Kraus (2017), American Psychologist, 2018. Used as the critique of the effect size in reference 3.
5. Juslin, P. N. and Laukka, P., Communication of emotions in vocal expression and music performance, Psychological Bulletin, 2003. Used for the review of 104 vocal expression studies.
6. Brennan, S. E. and Williams, M., The Feeling of Another's Knowing: Prosody and Filled Pauses as Cues to Listeners about the Metacognitive States of Speakers, Journal of Memory and Language, 1995. Used for the point that hesitation signals certainty (title and scope only).
7. Schroeder, J. and Epley, N., The Sound of Intellect, Psychological Science, 2015. Used for heard versus read evaluations.
8. Reichheld, F. and Markey, R., The Ultimate Question 2.0, advance excerpt, Bain and Company, 2011. Used for the original recommend question and its follow-up.

The weakest point in this source base: none of the tone and emotion studies used customer feedback, so the case that tone improves VoC decisions is an inference from lab research, not a measured result.