Key takeaways
- →Write the decision the survey must inform before you write a single question, then delete anything that does not move it.
- →Most abandonment happens at three cliffs: an intimidating first screen, the first matrix grid, and the first mandatory free-text box.
- →Cap a cold survey at around ten to twelve questions and a four-minute completion time, and say the real number up front.
- →Order questions so effort rises to a single open-text peak and then falls back to easy taps before the finish.
- →Anonymity buys you honesty on attitudes and costs you the ability to follow up, so choose it per survey rather than by default.
- →Publishing the results and one change you made because of them is what gets the next survey answered.
To design a survey people finish, work backwards from the one decision the answers have to inform, cap it at around ten questions, and put the cheapest questions first. Completion is a design problem rather than a motivation problem: people abandon forms because the questions cost more than the sender has earned, not because they do not care about the subject.
Abandonment happens in three predictable places. The first screen, where a long scroll or a demand for demographics signals that this will take a while. The first matrix grid, where twelve rows of radio buttons appear at once. And the first mandatory free-text box, where the effort jumps from tapping to composing with no warning and no way to skip.
This guide covers which question types earn their place, how to word questions that do not answer themselves, how to order them so effort rises and then falls, how long is genuinely too long, when anonymity helps and when it costs you, how to run a survey live inside a session instead of emailing it into the void, and what to do with the results so the next one gets answered too.
Why people abandon a survey halfway through
Every question is a small transaction. The respondent pays attention and candour; you promise that something will happen as a result. Most surveys run a deficit from question one - they ask for ten minutes of considered thought and offer nothing more concrete than a line about how much feedback matters to the organisation.
The drop-off curve is not smooth. It has cliffs, and the cliffs sit wherever the perceived remaining cost suddenly jumps: an unexpected page two, a grid of agreement rows, an open box marked required. Someone who has answered six questions is invested and will usually push through to the end. The whole job is getting them past the first two without a cliff.
There is a second, quieter failure that looks like success. People finish, but they satisfice: straight-lining down a matrix column, picking the midpoint on everything, typing a full stop into the required comment box. That is worse than abandonment, because it arrives as data and gets counted. Once a form asks for more than someone will hold in working memory, careless answering is the rational response, and the resulting averages are confidently wrong.
The numbers to design against
These are design constraints, not research findings - they are the budget you give yourself before the first draft. Constraints are what force the useful argument, which is not whether a question is interesting but whether it is worth one of your ten slots.
The four-minute figure is worth defending hardest. Every additional minute you ask for has to be justified to someone who is answering between two meetings on a phone, and almost no survey is improved by the questions that push it from four minutes to nine.
Start from the decision, not from the question list
Write the decision at the top of the page before you write a single question. 'We will choose one of these three onboarding formats by 5 September.' 'We will decide whether the Thursday all-hands continues in its current form.' Then ask what you would need to know to make that call, and cut everything that does not move it.
This kills the most common survey pathology: the omnibus form built by committee. Four teams each add 'just two quick questions', nobody owns the result, and a thirty-item questionnaire goes out with no decision attached. The response rate drops, the data nobody asked for goes unread, and the organisation learns that surveys are theatre - which makes the next genuinely important one much harder to run.
A useful test for each question: write the sentence you would say if the answers came back overwhelmingly one way. 'If 80 per cent say the handbook was not useful in week one, we rewrite it before the September intake.' If you cannot write that sentence, or if the sentence is just 'that would be interesting', delete the question. Curiosity is not a decision, and it is not worth someone else's four minutes.
Keep a separate parking list for the questions you cut. It is far easier to say no to a colleague's addition when the answer is 'not this survey, it is on the list for the next one' rather than a flat refusal.
Pick the question type that matches the job
The matrix grid is the one to be ruthless about. It looks efficient because it fits fifteen items on one screen, but it presents fifteen decisions simultaneously and invites the respondent to resolve them with one vertical swipe. If you genuinely need fifteen ratings, you probably need a different research method, not a tighter grid.
Ranking is underused by comparison. When you are choosing between five candidate priorities, an agreement scale will tell you that all five are important, because they are. A forced ranking or a small pot of points to allocate produces an answer you can act on, because it makes the respondent do the trade-off you were going to have to do anyway.
| Question type | Best for | Effort to answer | Watch out for |
|---|---|---|---|
| Single-choice multiple choice | A decision where the options are already known | One tap | A missing option forces a false answer - always offer 'none of these' |
| Agreement scale (1-5) | Attitudes you intend to track over several rounds | One tap | An unlabelled midpoint absorbs everyone who has not thought about it |
| 0-10 rating | A single headline number for a trend line | One tap | Close to meaningless without the follow-up asking why |
| Ranking or point allocation | Forcing real trade-offs between competing priorities | Moderate | Above seven items people rank their top two and guess the rest |
| Open text | Reasons, surprises, and language you did not anticipate | High | Keep it optional and allow at most two per survey |
| Matrix grid | Comparing many items against one scale, in theory | High and deceptive | Straight-lining down a column - split it or drop it entirely |
| Yes/no screening question | Routing people past sections that do not apply to them | One tap | Only earns its slot if the branch genuinely removes questions |
Write questions that do not do the answering for you
Top Tips
- Ask about one thing per question. 'Was the training useful and well paced?' cannot be answered by someone who found it useful but rushed, and you will never know which half a low score refers to.
- Strip out the adjectives you would like to be true. 'How helpful was our excellent new handbook?' is a leading question. 'How useful was the handbook during your first week?' is not, and it will get you a worse answer, which is the point.
- Anchor scales in behaviour rather than feeling wherever you can. 'How many times did you open it last week?' beats 'how often do you use it?', because people convert frequency into words like 'sometimes' very inconsistently.
- Label every point on a scale, not just the two ends. A 1-5 where only the extremes carry words is interpreted differently by every respondent, and the middle three points collect noise you cannot interpret later.
- Use an even number of points when you want people to lean one way, and an odd number when neutral is a genuinely valid position. Forcing a side on a question people are truly neutral about just adds noise in a different direction.
- Give every closed question an escape hatch: 'none of these', 'not applicable', 'I do not know'. Without one, people who do not fit your options answer at random, and you cannot separate them from real responses afterwards.
- Read each question aloud to someone outside the project. If they ask a clarifying question, that clarification belongs in the wording - you have just watched the question fail on its first respondent.
Order the questions so effort rises and then falls
People judge an experience disproportionately by its most intense moment and its ending, which is why the shape matters more than the average difficulty. A survey that ends on a demanding open box feels long even when it was short; the same questions in the reverse order feel brisk.
Show a real progress indicator and do not lie with it. A bar that sits at 60 per cent for four screens is worse than no bar at all, because it converts a mild irritation into a sense of being deceived. If your tool cannot show honest progress, say 'four questions left' in the text instead.
Open with one instant question
A single tap that anyone can answer without thinking - which team they are in, or how the last session felt in one word. The first answer converts a visitor into a respondent.
Move into the decision questions
Questions three to six carry the actual decision. Attention is highest here and people are already invested, so this is where the ranking or the trade-off belongs.
Place the one open box at the peak
Put your single free-text question after the closed ones, framed narrowly: 'what is the one thing you would change first?' Never mandatory, never first.
Drop back to easy taps
Follow the open box with two or three low-effort questions. Effort that keeps climbing to the last screen is what produces abandonment at 80 per cent complete.
End with optionals and a promise
Demographics last and optional, then one sentence on what happens next and by when. That sentence is the only payment most respondents get.
How long is genuinely too long
The honest ceiling depends on what you have earned. A two-question pulse check sent to a team that saw last quarter's results acted on will get answered by nearly everyone. A twenty-question engagement survey sent by a department nobody has heard from will not, however well written the questions are.
Say the real length up front and be specific: 'nine questions, about three minutes, closes Friday'. Vague reassurance - 'this will only take a few moments' - is disbelieved by anyone who has been burned before, and it should be, because it usually is not true.
If you cannot get under the ceiling, split the survey rather than compressing it. Two five-question rounds a fortnight apart will both get answered; one ten-question round sent cold may get neither. Splitting also lets you use the first round's answers to sharpen the second, which is a better use of the questions you were going to guess at anyway.
Watch out for the branch that promises to shorten things and does not. Conditional logic is worth it when a yes/no genuinely removes four questions for half the audience. It is not worth it when every path leads to the same twelve questions in a different order, and it makes testing far harder.
Anonymous, confidential, or named
These are three different things and people can tell them apart. Anonymous means you genuinely cannot connect a response to a person. Confidential means you can but you promise not to. Named means responses carry identity and everyone knows it. Say which one you are running, in plain words, in the first line - not in a policy link nobody opens.
Anonymity buys candour on anything with a career cost attached: how the reorganisation is landing, whether the deadline is realistic, whether a manager is listening. It costs you the ability to follow up on the interesting answer, and it removes any check on the person who wants to settle a score. Both of those are real prices, so pay them deliberately rather than defaulting to anonymous because it feels safer.
The credibility trap is worth naming. In a team of six, an anonymous survey that also asks for department, tenure and role is not anonymous, and everyone answering knows it. Either drop the demographics or drop the anonymity claim; keeping both is how organisations quietly teach people never to answer honestly again.
For a mid-session pulse check, anonymous is almost always right. The point there is a fast read of the room, and lowering the cost of the first honest answer matters more than knowing who said what.
Run it inside the session, not after it
The email survey has one genuine advantage: it reaches people who did not attend. If that group matters to your decision - and for anything about the session itself it usually does not - run both, and keep the two data sets apart.
Put the survey in front of people while they are still in the room
A survey answered in the last four minutes of a session gets a completion rate an emailed link will never approach. People are present, the experience is fresh, and answering is the thing everyone is visibly doing rather than a task competing with a full inbox.
Use the same join code as the rest of the session
A second sign-in step at the end costs you a third of the room. Session Flo runs the closing survey from the room code people already joined with, so answering is a tap rather than a fresh login on a phone that has locked itself.
Give it a hard, visible time box
Put ninety seconds on the clock and say so. A countdown that everyone can see turns an open-ended chore into a short shared task, and it stops the three people who write essays from setting the pace for everyone else.
Read two or three responses out loud immediately
Showing aggregated results on screen while people are still sitting there proves the answers went somewhere. It is the single cheapest thing you can do to raise the response rate of every survey you send that group afterwards.
Leave it open for a short tail
Keep it open until the next morning for the people who were pulled out early or joined late, then close it firmly. An indefinitely open survey never gets analysed, because there is never a moment when the data is final.
Separate the in-room and follow-up responses
If you also send the link afterwards, tag the two groups. People answering in the room are warmer and slightly more positive, and knowing which is which stops you comparing this quarter's in-room scores against last quarter's emailed ones.
Close the loop or the next one goes unanswered
The last two items are the ones that decide your response rate next time, and they are the two most often skipped. Publishing a summary is not enough on its own - people can tell the difference between a chart of their own answers and evidence that something moved. Name one change, even a small one, and name the thing you decided not to change and why.
Where you disagree with the majority answer, say so openly. 'Sixty per cent want the all-hands monthly rather than fortnightly, and we are keeping it fortnightly for the next quarter because of the launch - we will revisit in October.' That reads as respect. Silence reads as the survey having been a formality, and formalities do not get answered twice.
Keep the summary in the same place every time and make it short: what you asked, how many answered, the three things the answers said, and what happens next. A recap that lands in the same channel within two days does more for the next survey than any amount of rewording the questions.
- ✓The decision the survey served is written down before it goes out
- ✓Question count and real completion time stated in the invitation
- ✓Anonymity level named explicitly in the first line
- ✓Every closed question has a 'none of these' or 'not applicable'
- ✓At most two open-text boxes, both optional
- ✓Tested on a phone by someone outside the project
- ○Results summary published within 48 hours of closing
- ○One change made because of the answers, named publicly
The five ways a good survey still dies
First, no owner. A survey sent by a distribution list is answered as though it came from nobody, because it did. Put a name on it and put that person's name on the results too.
Second, bad timing. A survey landing the morning after a redundancy announcement, or at the end of quarter, measures the news rather than the topic. Move it a week; the data will be worth more than the schedule you were protecting.
Third, question drift between rounds. If you want a trend line, the wording and the scale must stay identical, even when you have thought of a better phrasing. Improve the questions in the round after you have retired the trend, not in the middle of it.
Fourth, over-analysis of tiny numbers. With eighteen responses, a shift from 3.4 to 3.7 is not a finding. Report the counts, quote the open text, and resist the decimal places that make eighteen people look like a study.
Fifth, no follow-up question. A 0-10 rating on its own tells you where you are and nothing about why. One short 'what would move that number up by one?' is worth more than the score it accompanies, and it is the question people actually enjoy answering.
If you cannot say what you will do differently depending on the answer, you are not running a survey - you are running a formality that costs everyone four minutes.
Frequently asked questions
How many questions should a survey have?
Ten to twelve is the ceiling for anything sent cold by email, and five or fewer for a recurring pulse check you want answered every time. The number that matters is not the count but the honest completion time - aim for under four minutes, timed by someone who was not involved in writing it. If you cannot cut below the ceiling, split the survey into two shorter rounds a fortnight apart rather than compressing the wording of a long one.
Should surveys be anonymous by default?
No - choose per survey. Anonymity is worth it whenever an honest answer carries a career cost: how a reorganisation is landing, whether a deadline is realistic, whether people feel able to raise problems. For anything about ownership, availability or follow-up, named responses are more useful and people rarely object. Whichever you pick, say it in plain words in the first line, and never claim anonymity while collecting team, tenure and role from a group of six.
What is the best way to increase survey completion rates?
Run it while people are still in the session rather than emailing a link afterwards, state the real length up front, and publish the results with one concrete change within two days of closing. Those three do more than any amount of question rewriting. Cutting the survey in half helps more than adding an incentive, and showing aggregated results on screen at the moment people finish is the cheapest trust-builder available.
Are rating scales better than open text questions?
They answer different questions and a good survey uses both, in proportion. Closed scales give you something comparable across rounds and cost one tap; open text gives you reasons and the language people actually use, and costs real effort. Keep the ratio to at most two open boxes in a twelve-question survey, always optional, and frame them narrowly - 'what is the one thing you would change first?' collects far more usable text than 'any other comments'.
How soon should I share the results of a survey?
Within 48 hours of closing, while people still remember answering. The summary needs four things and nothing more: what you asked, how many replied, the three clearest signals, and what happens next. Include the answers you disagree with and explain what you are doing instead. A group that has seen its answers used once will answer the next survey at a noticeably higher rate; a group that has been met with silence will not.
Can I use the same survey every quarter to track a trend?
Yes, and that is one of the strongest reasons to keep a survey short. Trend data requires identical wording, identical scales and a similar audience, so lock the core questions and resist improving them mid-series - a rephrased question breaks the comparison you have been building. Add up to two rotating questions each round for whatever is topical, keep the core five untouched, and note any change in how you distributed it, because an in-room survey and an emailed one are not directly comparable.
Keep reading
- whether a live poll or a survey fits your session — Choosing between an instant read of the room and a considered set of answers.
- running anonymous feedback without losing trust — What changes when responses cannot be traced back to a person.
- open questions versus multiple choice in live sessions — The same trade-off applied to questions asked live rather than in a form.
- measuring whether your meeting actually worked — What to do with the closing survey once you have the responses.
- run your closing survey inside the session with Session Flo — Surveys, polls and recaps from the room code people already joined.