Skip to main content

UX and research · Beginner · 13 min

Design a short survey people will finish

By Padding · Updated 18 Nov 2026
Any online form builder and a spreadsheet · November 2026

You will have a survey that takes under five minutes, built from a short list of questions that each feed a decision, piloted with three people, sent, and reported with an honest completion rate.

On this page

Most surveys that nobody finishes were too long before the first question was written. They grew in a shared document where every colleague added "while we are asking", and nobody had to answer for the cost: each extra question loses some respondents and makes the rest answer more carelessly. The fix is not clever wording. It is deciding what the survey is for and refusing everything else.

This tutorial starts one step earlier than most: whether you should run a survey at all. Then you write down what it must answer, cut the question list until each one earns its place, put the survivors in an order people can move through, pilot with three people and send it. You finish with a survey under five minutes and a completion rate you can state without hedging.

The running example is a made-up co-working space, Hollin Works, with about 400 members. The team wants to know whether to open on Sundays and whether to add bookable phone booths. Use your own product as you go; the example is there so each step has something concrete to point at.

Step 1 · Decide whether this should be a survey

A survey is good at one thing: telling you how many people do, have or prefer something, across more people than you could ever talk to. It is poor at why, at how people actually use a product, and at predicting what they will do later. People are bad at forecasting their own behaviour, and a box that says "Would you use Sunday opening?" collects polite yeses.

Write the question your team is asking in one line, then sort it:

  • "How many members work at weekends now, and where?" is a counting question. A survey fits.
  • "Why do members leave after three months?" is a why question. Interviews fit better; a survey would hand you a list of reasons you supplied yourself.
  • "Can members book a desk without help?" is a usability question. Watch people try it instead.
  • "How often do members use meeting rooms?" may already be in your booking data. Check before you ask people to remember it.

If your question is a why or a can-they question, stop here and pick another method. If it mixes both, split it: run the survey for the counting half and a few interviews for the rest.

For Hollin Works, the line becomes: "We need to know how many members already work on Sundays, and how many make calls they would rather take somewhere private."

Check it worked

You have one sentence that starts "We need to know how many" or "how often", and nothing in it asks why or asks people to predict a future choice. If you had to split the question, the other half is written down with its own method next to it.

Step 2 · Write what the survey must answer

Open a new document. Under a heading Decisions, write each decision the results will feed, who makes it, and by when. Under each one, write the specific thing you need to know, numbered N1, N2 and so on. Keep it to three needs or fewer.

For Hollin Works:

  • Decision: open on Sundays from March, or not. Operations lead, by 15 January.
  • N1: How many members worked on at least one Sunday in the past month, and where.
  • Decision: convert two meeting rooms into phone booths. Founder, by 15 January.
  • N2: How many members make calls at their desk that they would rather take privately, and how often.
  • N3: How long those calls usually last, so booths are sized for real use.

Then write who should answer: "Members with an active membership in the last 60 days." This rules out former members and day-pass visitors, whose answers would blur the numbers the decisions depend on.

Add one more line: the number that would change each decision. "If fewer than 1 in 10 members worked on a Sunday last month, we do not open." Writing the threshold now stops anyone reading the results generously later.

Check it worked

Your document has no more than three numbered needs, each sits under a named decision with an owner and a date, and at least one need has a threshold written next to it. There is a single line saying who should answer.

Step 3 · Draft every question, then cut hard

Now let everyone add questions. Put each on its own line in a list headed Candidates. This is the stage where the list gets long, and that is fine, because the next part is cutting it.

Go through the candidates one at a time and tag each with the need it serves (N1, N2, N3). Then apply these cuts in order:

  1. No tag, no question. "How did you hear about us?" may be interesting, but no decision on the page uses it. Delete it.
  2. You already have it. Membership type, sign-up date and floor are in your records. If you need them, join them to responses afterwards by member ID or a per-person link instead of asking, but read the note on anonymity below first. If no decision needs them, as at Hollin Works, drop them.
  3. Demographics you will not split by. Only ask age, role or company size if the decision changes depending on the answer. For Hollin Works it does not.
  4. Two questions for one need. Keep the one that is closer to recent behaviour. "On how many Sundays did you work last month?" beats "Do you usually work weekends?"
  5. Open text. Keep at most one, and make it optional.

Aim for six to eight questions in total. The Hollin Works list ends at six:

  1. In the past month, on how many Sundays did you do any paid work? (None / 1 / 2 / 3 or more) (N1)
  2. If any, where did you work on those Sundays? (Home / A café or library / Another co-working space / Somewhere else) (N1)
  3. In a typical week here, how many calls do you take at your desk? (None / 1–2 / 3–5 / 6 or more) (N2)
  4. Of those, how many would you rather take somewhere private? (None / Some / Most / All) (N2)
  5. How long is a typical one of those calls? (Under 15 minutes / 15–30 minutes / 30–60 minutes / Over an hour) (N3)
  6. Anything else about calls or weekend working you would like us to know? (optional, free text)

The wording of each question and its answer options is a skill of its own; the tutorial linked at the end covers leading, double-barrelled and vague questions. For now, get the list short.

Check it worked

Every remaining question has a need tag next to it, no need has more than two questions, there is no more than one free-text box, and the total is eight or fewer. The deleted candidates are still in the document under a struck-through heading, so you can show colleagues why theirs did not make it.

Step 4 · Put the questions in an order people can move through

Order changes both how many people finish and what they answer. Arrange the list using these rules:

  • Easy and factual first. Start with a question anyone can answer in two seconds without thinking about their feelings. Question 1 for Hollin Works counts Sundays.
  • Group by topic. Keep the Sunday questions together and the call questions together. Jumping between subjects makes people reread.
  • General before specific. Ask how many calls before asking which ones they would rather take privately, so the second question has context.
  • Skip what does not apply. If someone answers "None" to question 1, they should not see question 2. Use your form builder's option to show questions based on an earlier answer; if it has none, add "Does not apply" as an answer instead.
  • Free text and anything personal last. Someone who has answered five questions is more likely to type a sentence than someone who has answered none.

Then write the opening screen. Three short lines are enough: what it is for, how long it takes, and what happens to the answers. For example: "We are deciding whether to open on Sundays and add phone booths. This takes about three minutes. Answers are anonymous and only the Hollin Works team sees them."

Decide now whether that word is true, because three choices in this tutorial each make responses identifiable: joining answers to your records by member ID, sending each person their own link, and sending a reminder only to people who have not responded (your tool can only know who has not answered if it knows who each answer came from). If you do any of these, do not write "anonymous". Write "Answers are confidential; only the [team name] team sees them" and keep that promise. Hollin Works needs none of the three, so it uses one shared link, keeps the word "anonymous", and sends its reminder to everyone.

Build the survey in your form builder in this order, one question per screen or all on one page, whichever the tool does well. Turn off "required" on every question except the ones the decisions cannot do without.

Check it worked

Read the survey in your form builder's preview from start to finish. The first question can be answered without reading it twice, topics do not interleave, the free-text box is last, and choosing "None" on a lead question skips its follow-up. The opening screen states a time in minutes and says who sees the answers.

Step 5 · Pilot it with three people

Before anyone else sees the survey, sit with three people who match your "who should answer" line, in person or on a video call. Two members and one colleague who is not on the project works well. Do not use anyone who helped write it; they already know what each question means.

For each pilot:

  1. Silent, timed run. Send the link, start a timer when they open it, and ask them to fill it in as they normally would, without talking. Stop the timer when they submit. Watch quietly and note anywhere they pause, reread or change an answer.
  2. Think-aloud pass. Then go back to the first question together. Ask them to read each one aloud and say what they think it is asking and why they chose their answer. Note anything they misread, any option they wanted but could not find, and any "it depends". Do not time this pass; talking through each question takes far longer than answering it.

Write the results in a small table: name, silent-run time, and every stumble from either pass with the question number. Afterwards, fix the questions people stumbled on. Common finds are an answer range that misses a real case (someone takes calls but none at their desk), a time frame that is unclear ("this month" when it is the 2nd), and a term only your team uses.

If any silent run takes over five minutes, cut again. Go back to your candidates list and remove the weakest question; do not try to make questions faster to read.

Check it worked

Your pilot table has three rows, each with a silent-run time under five minutes, and every stumble you noted has either a change in the survey or a line saying why you left it. The time you put on the opening screen matches the slowest silent run, rounded up to the next whole minute, not the think-aloud pass.

Step 6 · Send it and record what you sent

Decide three things before you press send, and write them at the top of your spreadsheet:

  • Invited: the exact number of people you sent it to. For Hollin Works, 380 active members on the email list.
  • Channel: how they received it (email, in-app message, a link on the door). Send through one channel if you can, so the invited count is real.
  • Close date: when you stop accepting answers. A week is usually enough for an existing audience.

Send at a time your audience reads messages, not when it suits you. Put the time and the topic in the subject line or first line: "3 minutes: help us decide on Sunday opening". Send one reminder halfway to the close date. If your survey is anonymous, as Hollin Works' is, send it to everyone with a line thanking those who already answered. Sending it only to people who have not responded requires identified responses, so do that only if your opening screen says "confidential" rather than "anonymous".

If your form builder records how many people opened or started the survey as well as how many submitted, note where it shows that number. Not every tool counts starts, and you need to know before you report.

Check it worked

Your spreadsheet has a row with the invited count, the channel, the send date, the reminder date and the close date, all filled in before results come back. The live survey link opens to the same opening screen you piloted.

Step 7 · Report the completion rate honestly

After the close date, fill in three numbers:

  • Invited: from your spreadsheet.
  • Started: people who opened the survey and answered at least one question, if your tool counts them.
  • Completed: people who submitted.

Two rates come from these, and they mean different things, so name the one you are giving:

  • Completion rate is completed ÷ started. It tells you whether the survey itself was easy to finish. A short, piloted survey should lose few people once they begin.
  • Response rate is completed ÷ invited. It tells you how many of the people you asked took part, and depends as much on the channel and timing as on the survey.

For Hollin Works, if 380 were invited, 131 started and 118 completed, the completion rate is 118 ÷ 131, about 90%, and the response rate is 118 ÷ 380, about 31%. Write both on one line at the top of your results: "380 members invited, 118 completed (31% response rate); of 131 who started, 90% finished."

If your tool does not count starts, report only the response rate and say so. If it shows where people stopped, note the question where most left; that question is the first to cut or rewrite next time.

Check it worked

Your results document opens with one line that gives the invited and completed counts, names which rate it reports, and shows the division behind it. If you report a completion rate, a started count is in the spreadsheet to back it up.

Next steps

The six questions here still need careful wording before they are trustworthy. Write unbiased survey questions and answer scales works through flawed questions and how to choose answer options that fit what you measure.

When the answers are in, Read survey results without fooling yourself covers who did not answer, small numbers and comparing groups fairly.

If the survey is one part of a bigger study, put it in a one-page research plan first. If you have no list of people to send it to, start with recruiting participants for user research.

Need this done for you?

Padding does this work for growing companies, from the first diagnosis to a system that keeps working.

Digital Transformation

Get new tutorials

New tutorials, about twice a month.