Back
Employee Surveys: Questions That Get Honest Answers

Matthieu Saussaye

The Short Answer
Employees answer honestly when three things are true: they cannot be identified, they saw something change after the last survey, and the questions ask about facts rather than feelings about management.
Set a minimum reporting size, commonly 5 or more responses, and never break it.
Publish what changed since the previous round before you ask again.
Ask about specific, recent, observable things.
Do not run a survey you have no intention of acting on.
Four Types of Employee Survey
These are different instruments with different cadences, lengths and audiences. Running one of them and calling it a listening program is the most common mistake.
Type | Frequency | Length | Who takes it | What it is for |
|---|---|---|---|---|
Engagement | Once or twice a year | 25 to 40 questions | Everyone | A full diagnostic and a benchmark against your own history |
Pulse | Monthly or quarterly | 3 to 8 questions | Everyone, or a rotating sample | Tracking movement between full surveys and testing whether a fix worked |
Onboarding | At 30, 60 and 90 days | 5 to 10 questions | New joiners only | Catching a bad start while it is still fixable |
Exit | On departure | Interview or 10 to 15 questions | Leavers | Understanding why people go, from the only people who can say freely |
Two notes on the table. Pulse surveys are not miniature engagement surveys: they exist to track a small number of things you are actively working on, and if the questions never change in response to what you are doing, you are collecting a trend line nobody uses.
Exit surveys have a structural honesty advantage and a structural honesty problem. People about to leave have little to lose, but they also want a reference, so they soften. Running exit conversations a few weeks after the departure date, and away from the line manager, produces markedly more candid material.
Why Employees Do Not Answer Honestly
The instinct to protect yourself at work is rational. If you cannot be sure an answer is untraceable, the safe move is to give the answer that costs nothing. Four specific fears drive most of the distortion.
They do not believe the survey is anonymous
And often they are right to doubt it. A survey that greets you by name, sends a personalized reminder that stops the moment you complete it, or asks for department, tenure, level and location in a 12-person team is anonymous in name only. Employees do this arithmetic instantly.
What actually helps is being specific rather than reassuring. State which demographic fields are collected and why. State the minimum group size below which nothing is reported. State who sees the raw text, and confirm that managers receive aggregates only. If a third party processes the data, say so. "Your responses are completely confidential" is a slogan; "no result is shown for any group smaller than five, and your manager sees only group averages and unedited comments with names removed" is a commitment someone can check.
Nothing happened last time
The largest single cause of falling participation and softening scores. If an organization surveys, produces a deck and changes nothing, employees learn that the survey is a ritual. The rational response is to complete it quickly with harmless answers.
The fix is sequencing. Before the next round opens, publish what the last round said, what you decided to do, what you decided not to do and why, and what has actually shipped. The "decided not to do" part carries more credibility than the rest combined, because it proves someone read the results and made a real choice rather than picking the comfortable items.
The questions invite a safe answer
"Do you feel supported by leadership?" is an invitation to say yes. It is abstract, it is about a powerful group, and disagreeing feels like an accusation. "In the past month, has a manager helped you unblock something that was stuck?" asks about an event that either happened or did not.
Anchor questions to specific, recent, observable things. It reduces social desirability, it reduces acquiescence, and it produces answers you can act on, because "no" points at a behavior rather than at a sentiment. The mechanics of both of those biases are covered in our guide to survey bias.
The survey is too long
Forty questions of agree-or-disagree grids produce straightlining by question fifteen. That is not honest data, it is data shaped like data. Cut the instrument until every question has an owner who will act on the answer.
The Confidentiality Threshold
Set a minimum number of responses below which a segment is never reported. A threshold of five is a common convention; some organizations use ten for anything sensitive. The exact number matters less than the fact that it is decided in advance, published, and enforced without exceptions.
Three rules make it real.
Suppress the cell, and check what suppression reveals. If a team of 20 is reported and four of its five sub-teams are shown, the fifth can be reconstructed by subtraction. Suppress the complementary cells too.
Apply it to open text as well as scores. A verbatim can identify someone through phrasing or content even with names stripped. Either withhold comments below threshold or review them before release.
Never make an exception for a senior request. One manager receiving a breakdown they should not have had, once, ends the credibility of the whole program. And employees find out.
Small teams are still covered: their responses roll up into the next level. A manager of three people gets no dashboard of their own, and that is the correct outcome, not a gap to be engineered around.
Question Examples by Survey Type
Engagement
How likely are you to still be working here in 12 months? (0 to 10)
In the past two weeks, how often did you have the information you needed to do your job well?
When you raise a problem, how often does something happen as a result?
How clear are you about what success in your role looks like this quarter?
Thinking about the last decision that affected your work, how well was the reasoning explained?
What is the single biggest thing slowing you down right now? (open)
Pulse
How manageable was your workload over the past two weeks?
Did anything block you for more than a day this month? If so, what? (open)
Since the last survey, have you noticed any change in how meetings are run?
What is one thing we should stop doing? (open)
Onboarding, at 30 and 90 days
Did you have the accounts, access and equipment you needed on day one?
How well does the job match what you were told during the hiring process?
Who have you been able to go to with questions? (open)
What was confusing in your first month that nobody warned you about? (open)
At 90 days: is there anything about the role you would have wanted to know before accepting?
Exit
When did you first start thinking about leaving, and what prompted it? (open)
What could have been done, and by whom, to change your decision? (open)
How does your new role differ from this one in the way that matters most to you? (open)
Would you recommend this company to someone with your skills? Why or why not? (open)
Was there a point where raising something would have made a difference? (open)
Notice how many exit questions are open. By then you want the story, not a score, and the sample is too small for scale points to mean anything. That material only becomes usable if you code it into themes rather than reading it once; how to collect, code and report qualitative data covers the method.
Scales and Wording
Prefer item-specific scales over agree-or-disagree grids. "How manageable was your workload?" running from very unmanageable to very manageable produces cleaner data than asking people to agree with "My workload is manageable", because there is no agreeable direction to default to. Keep the number of scale points and the labels identical across waves, since changing them breaks your trend line more thoroughly than any real event will. Likert scale design goes through the format choices.
Keep demographic questions to the minimum you will actually analyze. Every additional attribute narrows the intersection and makes people feel identifiable, whether or not they are.
Do Not Run One Unless You Intend to Act
A survey is a promise. Asking someone what is wrong signals that you plan to do something about it, and an organization that asks repeatedly and changes nothing is actively teaching its staff that speaking up is pointless. On that measure, no survey at all is better than an annual one that produces a deck.
Before you field, commit to four things in writing.
A named owner for the results, senior enough to change something.
A date when findings will be shared with everyone, not only with the leadership team.
A budget or a slot in the roadmap for whatever comes out. If nothing can move for six months, run the survey in six months.
The scope of what is not on the table, stated up front. If the office policy is fixed for the year, say so before you ask, rather than collecting hundreds of comments about it and ignoring them.
The Loop That Makes the Next Round Honest
Five steps, in order.
Share the results within two weeks, including the uncomfortable ones. Filtered results are noticed and cost you more than the bad numbers would have.
Let teams interpret their own data before any action plan is written for them. The people in the team usually know what the score means; the dashboard does not.
Pick two things and finish them. Ten half-actions read as zero. Two visible completed changes reset the credibility of the entire program.
Say what you are not doing, and why. This is the highest-value paragraph in any results communication.
Reference it when you ask again. Open the next survey by naming what changed since the last one. Participation and candor both respond to this more than to any reminder email.
The measurable signal that this is working is not the engagement score. It is the length and specificity of the open responses. When people write more, and name concrete things, they have concluded that writing is worth their time.
Related Reading
Let people say it in their own words
Scale points tell you a team is struggling. They never tell you what to fix. The answer is always in the open box, which is exactly the part most survey programs collect and never read properly.
SmartInterview runs surveys by voice or text with an AI that asks the follow-up question a good interviewer would have asked, so a short answer becomes a specific one, in whatever language the employee is most comfortable using. The responses come back coded into themes you can count, aggregated so no individual is identifiable, and each theme stays traceable to the underlying comments for anyone who needs to check the reading.
Try it on one pulse question and compare it with what your current form returns.
Frequently Asked Questions
How do you make an employee survey truly anonymous?
Collect only the demographic fields you will genuinely analyze, set a minimum group size below which no result is reported, suppress complementary cells so a hidden group cannot be reconstructed by subtraction, review open comments for identifying detail, and state all of this in plain language before the first question. Vague reassurance does not work; a specific, checkable rule does.
What is the minimum group size for reporting employee survey results?
Five responses is a widely used convention, with a higher threshold for sensitive topics. The precise figure matters less than deciding it before fieldwork, publishing it, and never granting an exception. Smaller teams simply roll up into the next reporting level.
How often should you run employee surveys?
A full engagement survey once or twice a year, with short pulse surveys monthly or quarterly in between, plus event-triggered onboarding surveys at 30, 60 and 90 days and an exit survey on departure. The constraint is not appetite for data, it is your capacity to act between rounds.
What questions should you ask in an engagement survey?
Questions about specific, recent, observable things: whether you had the information you needed in the past two weeks, whether raising a problem led to anything happening, how clearly the reasoning behind the last significant decision was explained. Avoid abstract loyalty statements, which invite agreement and produce nothing actionable.
Why do employees not answer honestly?
Usually because they doubt the anonymity, because the last survey changed nothing, or because the questions are phrased so that a positive answer is the socially safe one. All three are fixable, and the second is the one that does the most long-term damage to participation.
What is the difference between an engagement survey and a pulse survey?
An engagement survey is a long annual or biannual diagnostic across the whole organization. A pulse survey is a handful of questions run monthly or quarterly to track the specific things you are currently working on and to check whether a change actually landed. Pulse surveys are not shortened engagement surveys; their question set should evolve with your priorities.
Should managers see their team's verbatim comments?
Only above your reporting threshold, and only after comments have been checked for identifying detail. Managers of very small teams should receive rolled-up results instead. Withholding a breakdown from one manager is a small cost; a single identifiable comment reaching the wrong person ends the credibility of the program.


