How to create a survey, from the decision back to the questions

To create a survey, work in seven steps: decide which decision the answers will inform, choose who must answer, write only the questions that decision needs, order them from easy to hard, choose a one page or one question at a time format, pilot it with a handful of people, then share the link with a closing date. Most of the work is in the first three steps.

Step 1: start from the decision

A survey is worth sending when its answers will change something: which product to keep, which session to drop, whether staff workload is sustainable. Write that decision down in one sentence before writing any question, because every later choice follows from it: who to ask, what to ask and in what order. Pew Research Center’s notes on questionnaire design list question wording, open against closed questions and question order as the choices its staff weigh, and each one is easier with the purpose fixed first.

Most of the 30 survey examples on this site say on their welcome screen what the answers are for, such as deciding where a gym’s equipment budget goes or informing a hybrid work policy review. That one line also does the work of an invitation.

Step 2: choose the audience

Decide who must answer and who must not. Customers who bought in the last month, every member of staff, students in one course. If the decision is about a group, you need answers from that group, not whoever happens to click. The sample size page and the free sample size calculator help when you need enough answers to trust a percentage.

Step 3: write the shortest question list

List every question you could ask, then strike each one whose answer would not change the decision. What remains is the survey. The 30 survey examples here ask between 7 and 14 questions, with a median of 8.5. SurveyMonkey’s analysis of about 100,000 surveys found the first question takes about 75 seconds and later ones about 30, so a ten question survey takes around five minutes. The survey question count post breaks length down by survey type, and how to write survey questions covers wording.

Step 4: put them in order

Easy, general and closed questions go first; open, sensitive and demographic questions go last. 20 of the 30 survey examples open with a single choice question and 20 end on a text question, usually the one thing we could do better. Group questions about one topic together, and use the same scale for questions you want to compare. How to write a questionnaire covers block order in detail.

Step 5: choose the format

A short survey sent as a link works best one question at a time, because it feels quick on a phone; all 30 survey examples here are conversational. A long survey filled in at a desk works better on one page with sections, so people can see how much is left. GOV.UK’s guidance on form structure applies to both: only add a question if you know you need the answer, and start by splitting a form into pages that each hold one thing.

Step 6: pilot it

Send the draft to five people like your audience and watch them answer, or ask them afterwards what each question meant to them. The US Census Bureau requires pretesting for every new questionnaire and names cognitive interviews, usability testing and respondent debriefings among the accepted methods. A small business does not need a lab, only a few honest testers. How to pilot a survey sets out a five person test.

Step 7: share it and close it

Decide where the link goes (email, a receipt, a QR code at the counter), when it closes and whether one reminder goes out. Put the honest time on the welcome screen; the survey examples here promise a median of three minutes. After it closes, read the answers against the decision from step 1, which is what how to analyse survey results walks through.

Survey or form

A survey asks people what they think; a form collects what someone needs to give you, such as a booking or an application. The difference matters for anonymity and length. The form against survey page sets out when each fits, and the customer feedback survey shows a six question survey built this way.

A worked example

Saltbush Cycles, a fictional bike shop, wants to know whether to open on Sunday mornings. The decision is a yes or no about one trading slot, so the survey only needs a few questions: how often the customer visits, which days and times suit them, whether they would come on a Sunday morning and for what (a service, parts, a test ride), and one open question for anything else. That is five questions, not twenty.

The audience is customers who have bought something in the last six months, not everyone on the mailing list, because people who never visit cannot tell the shop much about opening hours. The order runs from easy to specific: visit frequency first, then current times, then the Sunday question, then the reason, then the open box. The format is one question at a time, because it goes out by text message after a service. Five customers try it in the shop first, and one points out that “Sunday morning” means 7 am to a road cyclist and 10 am to a family, so the option becomes “Sunday 8 am to 12 pm”. The link closes after two weeks, and the answers are read against the single decision written down at the start.

In the product

A survey can be generated from a description of up to 10,000 characters in 32 suggested languages, as a classic or conversational form, and edited in the builder afterwards. Conversational forms can skip ahead with simple jump rules on an answer; there is no question randomisation, scoring or calculation. Responses arrive in a table and export to CSV with one column per question. The conversational form builder page shows the one question at a time format, and the guide to creating a form with AI walks through a prompt.