Employee Engagement Surveys: What the Questionnaire Never Shows
The survey closed with 78 responses from 96 people. The overall score came in at 7.1 — same as last semester. “Communication” fell 0.4, “recognition” sat exactly where it always sits. The committee meets for forty minutes, looks at the chart, and walks out with the sentence that was already written in the previous minutes: we need to improve communication.
Nobody on the committee can say what exactly needs improving. And it isn’t a lack of data: the employee engagement survey answers “how much” with precision and “why” with silence. The company takes everyone’s temperature and examines no one.
This article is about the missing part: listening. How to run climate interviews that explain the number, how to record without turning listening into surveillance, and how to get from five hours of audio to five themes and three decisions — instead of another forty-slide report.
The score arrived; the reason did not
The global data helps show the size of the problem. In its State of the Global Workplace 2026 report, Gallup finds that 20% of workers worldwide were engaged in 2025 — the lowest level since 2020 — and estimates that low engagement cost roughly US$10 trillion in lost productivity, around 9% of global GDP.
The item that matters here is Q07 on the Q12 questionnaire: “at work, my opinions seem to count.” Globally, one in four people strongly agree with that statement. In the United States, the figure is 26%.
Now the number that closes the argument: according to Gallup itself, only 8% of employees strongly agree that their organization acts on survey results. Eight percent. In other words, most people fill out a form about their own working life and see nothing change — and the next form is answered with that memory in hand.
And when the company does act, the effect is measurable. The 2021 systematic review of survey follow-up studies brings together two cases from the same organization: Church and Oliver (2006) showed that respondents who said results had been used for action rated job satisfaction more favorably; Church and colleagues (2012), using longitudinal data, found that this group stayed consistently more favorable across every item and every year.
The reading is direct: the number does not change behavior. The consequence of the number does. And consequence requires explanation — which is precisely what the chart does not deliver.
Why the open-ended field doesn’t solve it
The obvious fix would be “leave an open field and ask for comments.” The problem is what people don’t write in it.
The classic study by Milliken, Morrison and Hewlin (2003), published in the Journal of Management Studies, interviewed 40 professionals. The result: 85% reported having been in a situation where they could not raise an issue they considered important. Only 51% said they felt comfortable, in general, raising problems with their boss — and 23% answered “no, not at all.”
What stays out is specific, and it is exactly the useful content:
| What people don’t raise | % who reported it |
|---|---|
| Concern about a colleague’s or boss’s competence | 37.5% |
| Problems with processes, performance, or a suggestion for improvement | 35% |
| Pay and pay equity | 27.5% |
| Disagreement with company policies or decisions | 22.5% |
| Ethical or fairness issues | 20% |
| Harassment or abuse | 17.5% |
The most cited reason for silence was fear of being labeled — as a troublemaker, a snitch, a complainer — followed by fear of damaging working relationships.
Morrison’s 2014 review adds more numbers in the same direction: Ryan and Oestreich (1991) heard from 260 employees at 22 organizations, and 70% said they were afraid to speak up about certain topics; Detert and colleagues (2010), with 439 respondents, found 42% who had withheld information when they felt they had nothing to gain; and Souba and colleagues (2011), listening to 254 department chairs in medicine and surgery, found 69% saying that not speaking up about important problems was common or widespread in their own organization.
Look at how the open field is designed: it asks, in writing, under the respondent’s name, inside a form whose results the area manager will read, for precisely the topics the research says people hide. This is not a problem of employee honesty. It is a problem of collection design.
The kind of listening that explains the number
Qualitative listening does not replace the survey. It is the second move: the survey shows which dimension hurt, the conversation shows which process produced the pain.
The rule that makes the difference is trading opinion for incident. “What do you think of communication?” produces an adjective. “Tell me about the last time you found out about a decision too late” produces a date, a process name, and a cost. An adjective does not become an action plan; an incident does.
| Dimension that dropped | Incident question | What it reveals |
|---|---|---|
| Communication | ”What was the last decision that reached you too late? What did you do while you didn’t know?” | Where information gets stuck and how much rework it generates |
| Recognition | ”When was the last time your work went unnoticed?” | Whether the problem is frequency, criteria, or visibility |
| Workload | ”Last week, what did you leave undone because there wasn’t time?” | Which task gets sacrificed — and who absorbs the damage |
| Leadership | ”The last time you asked for help, what happened?” | Whether leadership responds with resources, advice, or pressure |
How many conversations? Between 8 and 12 per group, split by area and by tenure. Fewer than that and you hear a handful of isolated cases; much more than that and you start doing an audit instead of listening. The practical stopping rule: when the same theme shows up in six out of ten conversations with different people, it is not one person’s bad day — it is a pattern.
The five-question script
Twenty-five to thirty minutes, five open questions, none of them about “climate.” The script exists so you don’t improvise and so you can compare groups.
- “Tell me how your last workweek went — what took up your time?” Opens without steering the topic and shows the gap between the prescribed task and the real task.
- “When was the last time you had to handle something that wasn’t your job?” Reveals processes with no owner, the kind of work nobody measures.
- “When you disagreed with a decision, what happened next?” Measures the safety to disagree with a fact, not with a score from 1 to 5.
- “What have you given up asking for here?” This is the question that captures structured silence — the agenda item that died before the meeting.
- “If I could fix one thing in 30 days, what would you choose?” Closes with a stated priority rather than a venting session.
At the end, a sixth: “What should I have asked that I didn’t?” It usually produces the best line of the conversation.
Three probes do the rest of the work: “what happened next?”, “who else saw that?”, and “how long does that take?”. Avoid a bare “why?” — it sounds like an interrogation and produces justification instead of a narrative.
And don’t ask “how’s the climate?”. The question is abstract, invites the obvious answer, and hands the interviewee the job of diagnosing the company. You are the one doing the diagnosis.
Recording without becoming surveillance
Recording is what lets you actually listen instead of taking notes during the conversation. But recorded listening is personal data processing, and LGPD applies even when the intention is good.
The three principles that matter here are in article 6: purpose (legitimate, specific, explicit purposes disclosed to the data subject), adequacy (compatibility with what was disclosed), and necessity (limitation to the minimum necessary). Article 16 adds that data must be deleted at the end of processing, except for specific legal cases. In practice: if you can’t say in one sentence what the audio is for, don’t record it.
A fifteen-second consent script covers most of it:
“I’m going to record this conversation so I don’t have to rely on my memory when I write the report. The transcript stays with [responsible area], your name won’t appear in the final material, and you can ask me to delete the audio later. Is that okay?”
Five rules that handle the rest:
- Purpose in writing. Before the first conversation, define what will be done with the material, who reads it, how long it is kept, and what will be published.
- Out of scope. Do not record individual conversations about performance, health, harassment, or whistleblowing. Those topics have their own formal channel; mixing the two destroys the channel and creates legal risk.
- Minimum grouping. No excerpt is reported for a group with fewer than five respondents. With three people, one quote identifies who spoke.
- Audio is temporary. Transcribe it, check the excerpts you will use, and delete the audio. What remains is aggregated text, not a voice.
- Third-party names come out. If someone names a colleague in a complaint, the name does not go into the shared transcript — it becomes “a colleague on the same team.”
Watch for a point that slips by: a line like “I haven’t slept since the shift change” can reveal health data, which LGPD treats as sensitive (article 11). That excerpt does not go into the report with a name or a direct quote; it goes in as an aggregated theme — “impact of shift scheduling on rest” — and, if there is a concrete case, it is routed through the proper channel, not through the survey.
From five hours of audio to five themes
Do the math before you promise the process: 12 conversations of 25 minutes are 5 hours of audio, somewhere between 45 and 60 pages of transcript. Nobody codes 50 pages at six o’clock on a Friday. This is why qualitative listening dies in most companies: not for lack of method, but for lack of time to transcribe and organize.
This is where the tool stops being a convenience and becomes a condition of viability. In Sintesy the flow is direct: you upload the interview audio (or record the conversation from the dashboard), and you get a transcript and a summary for each recording. The mind map helps you see the groupings that repeat across conversations; chat with sources lets you query the whole set — “where do people talk about overload?” — without rereading everything.
Two practical plan details, so there are no surprises: the free plan covers 60 minutes of audio per day and 10 syntheses per day, with files up to 20 MB, and long-audio transcripts come truncated at 2,500 characters. For five hours of interviews, the path is the paid plan (Pro), with files up to 2 GB. The action checklist is generated when the content is prescriptive — if the answers are purely descriptive, what comes back is a roadmap, not a task list.
On the analysis method: use the six phases of Braun and Clarke’s thematic analysis (2006) — familiarizing yourself with the material, generating initial codes, searching for themes, reviewing themes, defining and naming, writing up. The 2022 revision renamed the phases, but the logic is the same, and it remains recursive: you go back to the material.
The automatic summary is a starting map, not the analysis. It groups by proximity of subject; you are the one who decides what is a theme, what is an exception, and what is noise. If you skip the coding phase and go straight to the summary, what you have at the end is minutes — not a diagnosis.
One habit that changes the quality: read the full transcripts, not just the summaries. The sentence that explains the number is almost never in the main theme. It is usually in the third answer of a conversation about something else.
The feedback loop is what moves the number
Listening becomes theater when nothing comes back. And coming back is not sending the report: it is saying three things publicly.
- What will be done, with an owner and a deadline. Three items, not thirty.
- What will not be done, and why. Declining with an explanation is what separates a decision from a promise.
- When we talk again. Two weeks, not the next annual cycle.
Gallup’s 8% does not measure a lack of good intentions; it measures a lack of feedback. And feedback has a useful side effect: people answer the questionnaire with less cynicism, because they have evidence that the previous answer was used.
It is worth remembering that a listening conversation is not the same thing as an exit interview: the exit interview arrives too late by definition and exists to learn from people who have already decided to leave. Climate listening exists to act before that.
Five mistakes that cancel out the listening
| Mistake | What happens next | Fix |
|---|---|---|
| Listening and not reporting back | In the next round, nobody says anything relevant | Feedback within two weeks, with three commitments |
| Using a quote as evidence against someone | The source dries up and the area learns to protect itself | Aggregated report, no attribution, no traceable quote |
| Recording without clear consent | Broken trust and legal exposure | Consent script + purpose in writing |
| Generalizing from three people | An expensive decision that looks like data | At least five respondents per group before reporting |
| Comparing the mood of the day with climate | An invented trend | Same script, same groups, same window of the year |
Two smaller but common mistakes: turning the listening into a venting session with no record (nothing that was said becomes a theme) and inviting only the people who already speak in meetings (you end up hearing yourself with four different accents).
What to do this month
Pick the dimension with the worst score. Convene eight people from two areas, with different tenure. Run the five-question script, record with consent, transcribe the same day. Read the full transcripts, mark what repeats, choose three themes. Publish three commitments with an owner and a deadline, and say what will not be done.
Twenty-five minutes of conversation explain more about the 0.4 drop in “communication” than the entire chart. The engagement survey is still useful — just don’t ask it for what it doesn’t do: say why.


