How to make a survey: 10 steps from decision to results

Most surveys that disappoint were in trouble before anyone wrote a question. They began as a list of things someone was curious about and produced a spreadsheet nobody acted on. A survey is a tool for making a decision with less guessing, and it works best when you build it backward from that decision.

This guide walks through how to create a survey in the order the work happens, with one running example: a neighborhood yoga studio with about 500 active members, deciding whether its one new class should run on weekdays at 6:30am or on Saturdays at 3pm. Every number here is standard sampling arithmetic or a labeled illustration.

How to make a survey: the 10 steps at a glance

1. Name the decision the survey will inform, in one sentence.
2. Choose who to ask and how you will reach them.
3. Work out how many responses you need for the precision the decision requires.
4. Write the questions, starting with the one the decision depends on.
5. Pick a question type for each answer you need.
6. Put the questions in order and build the survey.
7. Add logic so people skip questions that do not apply to them.
8. Pilot it with five people from the real audience.
9. Publish and share it by link, embed or email.
10. Analyze the responses and make the decision you planned.

Step 1: Name the decision the survey informs

Before any question, finish this sentence: "We will use the answers to decide..." If you cannot finish it, you are not ready to write a survey yet.

The studio's sentence is "We will use the answers to decide whether the new class runs on weekdays at 6:30am or on Saturdays at 3pm." That one line does three jobs:

• It names the audience. Current members, since they are the people who would attend. Not the public, and not everyone who ever took a trial class.
• It names the must-have question. Which of the two times would you actually attend?
• It gives you a delete test. For every other question, ask: if this answer came back tomorrow, would it change the decision or how we carry it out? If not, the question goes.

Then write down, before any answers arrive, what result would make you choose each option: "If one time is clearly ahead on 'would attend at least twice a month', we run it for a term." Setting the rule in advance stops you from reading the results the way you hoped they would come out.

Step 2: Choose who to ask and how you will reach them

The group you want conclusions about is your population. For the studio, that is active members, defined as anyone who attended a class in the last three months. Write the definition down, because it decides whose answers count.

Next, decide how the survey will reach them. Every channel reaches a different slice of people:

• Email reaches everyone on the list, but only people who read your emails will answer.
• A link or embed on your website reaches people who visit it, who lean toward your most engaged customers.
• A QR code at the front desk reaches whoever is in the building that week, which for the studio means people the current schedule already suits.
• A post in a group chat or community reaches the most active members and few of the quiet ones.

If the population is small, invite everyone. If it is large, a random sample you invite directly usually tells you more than a public link anyone can find, because you know who was asked. Some bias remains either way: people with strong opinions answer more readily. You can shrink it by keeping the survey short, saying how long it takes, explaining what the answers will change, and sending one reminder.

Collect names or emails only if the decision needs them. Data you never collect is data you never have to protect.

Step 3: Work out how many responses you need

How many responses are enough depends on how precise the answer has to be, and precision is described by the margin of error. If 60% of respondents say they would attend the Saturday class and the margin of error is ±5 points, the true share across all members is likely somewhere between 55% and 65%.

"Likely" has a specific meaning. Surveys usually work at a 95% confidence level: run the same survey many times on fresh random samples, and about 95 of every 100 ranges calculated this way would contain the true figure. A handy rule of thumb is that the margin of error is roughly 1 divided by the square root of the number of responses:

• 100 responses: 1 ÷ 10, about ±10 points.
• 400 responses: 1 ÷ 20, about ±5 points.
• 1,100 responses: 1 ÷ 33, about ±3 points.

Small populations need fewer responses than that, because each answer covers a bigger share of the group. These are the standard figures at 95% confidence, assuming the most cautious case, where answers split 50/50:

Completed responses needed at a 95% confidence level with a 50/50 split, from the standard sample size formula with the finite population correction, rounded up.

The studio has 500 members, so ±5 points would take 218 responses, nearly half the membership. Its decision does not need that precision. It only needs to know whether one time is clearly ahead, and at ±10 points, which takes 81 responses, a 60 to 35 split is unmistakable. A 48 to 45 result would be a finding too: the options are close enough to decide on other grounds, such as which teacher is free.

Then work backward to invitations: divide the responses you need by the share of your audience that usually answers. As an illustration, if about one member in three replies to studio emails, 81 responses means inviting at least 243 people. With 500 members, the studio invites everyone.

One exception: when a question exists to find reasons rather than measure shares ("why did you cancel?"), keep collecting until new answers stop adding reasons you have not seen.

Step 4: Write the questions

Write the must-have question first, the one from Step 1. Then add only the questions that change how you read it. The studio also asks how long each person has been a member and how many classes they take, because a class that newer, twice-a-week members want is a different bet from one only long-time members want. The survey questions guide has 120 examples to copy. Whatever you borrow, check it against these five mistakes:

Use the words your respondents use, and give every list of options a way out, such as "None of these" or "Other", so nobody is pushed into a wrong answer.

Mark a question required only if your analysis breaks without it, which for the studio means the class-time question and nothing else. Required questions that feel optional make people leave or type "n/a" to get past them, one of the causes in the form abandonment guide. Then run the delete test on every question. The studio's survey ends at six questions and a thank-you screen.

Step 5: Pick the right question type for each answer

The question type decides what the answer looks like in your data, so choose it from the analysis backward. If you will count the answers, give people fixed options. If you want reasons in their own words, leave it open. The survey question types guide covers every type with examples. This table is the short version:

For scales, keep the same direction and the same number of points throughout the survey, and label both ends so a 4 means the same thing to everyone.

Two familiar formats need a plan. Upform has no matrix or grid question, so a block of statements rated on one scale becomes one Rating question per statement. With one question per screen, that is easier to answer on a phone than a grid, but each statement adds a screen, so keep only the ones you will act on. There is no ranking type either. To find priorities, use Multiple Choice with a cap ("pick up to three") or Single Choice for the top pick, as the ranking questions guide explains.

Step 6: Put the questions in order, then build the survey online

Order changes answers. An early question can color a later one, and a hard question at the start costs you people who would have answered everything after it. A dependable order:

• Open with something easy and on topic. "How long have you been a member?" takes a second and says what the survey is about.
• Go from general to specific. Ask how well the current schedule suits people before showing them new times, or the new times will color the rating.
• Put the decision question early, while attention is highest. The studio asks it third.
• Save open text and personal questions for the end, when people have a reason to answer them.
• Close with a thank-you screen that says what happens next, and when.

Now build it. In Upform a survey is a form: start from a blank form or one of the survey templates, add each question from the Add Question dialog, and drag the rows into the order you chose. The builder has three columns: the question list on the left with the Thank You screen pinned last, a 1:1 live preview in the middle with desktop, tablet and mobile views, and the settings for the selected question on the right, including its Required switch.

Saves are manual, and the builder warns you before you leave with unsaved changes, so save after each round of edits.

Step 7: Add logic so people skip what does not apply

Every question someone answers that does not apply to them costs attention and adds noise to your data. Conditional logic removes those questions for the people they do not fit. In Upform, conditional logic is a list of rules. Each rule joins one or more conditions with AND or OR, using 20 operators, and takes one of four actions: show a question, hide it, jump to another question, or end the form.

The studio needs two rules:

Write each rule as a sentence first ("when X is Y, show Z") and check it says what you mean. Plan around the operators each type offers: Rating and Star Rating support equals, greater than and less than, so "show a follow-up when the rating is less than 3" works. And remember that a question shown to a fifth of respondents collects about a fifth as many answers. That is the rule working, not people leaving.

Step 8: Pilot it with five people

A pilot finds the question everyone reads differently before it reaches your whole audience. Five is not a magic number. It is a practical one: enough people to surface the questions they stumble on, few enough that you can watch every one.

Pick them from the real audience (the studio asks five members, not its teachers) and run it like this:

1. Ask each person to take the survey on their own phone while you watch, and to think aloud as they go.
2. Note every place they hesitate, reread a question or ask what something means. Do not explain. Real respondents will not have you there.
3. Time them. Whatever the invitation promises, such as "about two minutes", has to be true.
4. Afterward, ask what they thought the key question was asking. A wrong answer here is the most useful thing a pilot can produce.

Then take the survey yourself once for every logic path, and check that each test answer landed in the right column where you will analyze it. In Upform, run the pilot on a copy: duplicate the form from the dashboard, publish the copy and send its link to your testers, so test answers never reach your real data. Once a form is published, the Preview button in the builder's top bar opens it in a new tab exactly as respondents see it. If the pilot changed a lot, try it on two or three new people.

Step 9: Publish and share the survey

A survey stays private until you publish it. In Upform, the button at the right of the builder's top bar reads Unpublished until you click it, and the Share tab then gives you three ways to reach people:

• A link, for emails, messages and social posts, or printed as a QR code for the front desk.
• An embed, an iframe snippet that puts the survey inside a page on your website. The how to embed a form guide covers where to place it.
• An email button. Most email clients do not support iframes, so the survey cannot live inside the email itself. Instead you paste in a button that opens it.

<iframe src="https://app.upform.com/preview/member-schedule-survey" width="100%" height="600" frameborder="0" style="border:none;border-radius:8px;"></iframe>Copy

The invitation matters as much as the survey. Say who is asking, why this person, how long it takes, what the answers will change, and when they will hear the result. Keep it short:

Set a closing date, send one reminder a few days before it, and avoid sharing the link somewhere new halfway through: a new channel brings in a different slice of people.

Step 10: Analyze the responses and make the decision

Start with the count. Compare the number of completed responses with your target from Step 3 before you read a single percentage. If you are short, the margin is wider than you planned, and your report should say so.

In Upform, every response lands in the Responses table, which you can search, filter by form and date, and export as CSV. Clicking a row opens every answer from that person in a side drawer. For pivot tables and charts, connect Google Sheets before you publish, and each response is appended to your sheet as it arrives.

Then work through the answers in this order:

1. Answer the decision question first. Count how many people would attend each option. People could pick both times, so the shares will not add up to 100%.
2. Split by one factor that matters, such as length of membership, only where each group has enough responses.
3. Tag every open answer with a short theme ("works late", "childcare on Saturdays"), count the themes, and quote a few word for word.
4. Report the result with its sample size and margin.
5. Make the decision you committed to in Step 1, then tell respondents what you decided. It is the best reason you can give anyone to answer the next survey.

For the studio, an illustrative result reads like this: 142 of 500 members responded. 58% said they would attend the Saturday 3pm class at least twice a month and 31% the weekday 6:30am class, with a margin of about ±7 points. The gap is far wider than the margin, so the Saturday class runs next term, and the front desk says so.

The Results tab adds views, starts, completion rate and average completion time. Its question funnel counts submitted responses, so a question with fewer answers than the one before it was optional or skipped by logic, a quick check that your rules fired.

A checklist before you send any survey

• You can finish the sentence "We will use the answers to decide..."
• The population is written down, and the channel you picked actually reaches it.
• You know how many responses you need, and how many people to invite to get them.
• Every question passes the delete test, and none of them asks two things at once.
• Every list of options is complete, has no overlaps, and offers a way out.
• Only the questions your analysis cannot do without are required.
• Five people from the real audience have taken it, and you have tested every logic path yourself.
• The invitation states how long the survey takes and when people will hear the result.

If a box stays unticked, fix it before you send. A survey that goes out a week late can still inform the decision. One that goes out with a question people read in different ways cannot.