How to create an online survey
A survey is the easiest thing in the world to get wrong. Too many questions, half of which change nothing, inconsistent scales, and nothing readable at the end. This walkthrough is about avoiding that — and it starts before you write the first question.
Step by step
- 1
Write down what you will do with the answer
One sentence per question: “if most people answer X, we will do Y.” A question you cannot write that sentence for does not belong in the survey. This single step usually deletes a third of them.
- 2
Pick the format by what you can do with the answer
A scale can be totalled and compared over time. An open question tells you why, but somebody has to read it. A good survey has both — a number for the trend and one open question for the explanation.
- 3
Ask about one thing at a time
“How satisfied are you with our speed and communication?” is a trap — when someone answers 3, you do not know which half it was about. Split it in two.
- 4
Keep scales consistent
If 1 is worst in the first question, it must be worst in all of them. A flipped scale in the middle ruins the data and nobody notices.
- 5
Decide on anonymity up front
An anonymous survey gets franker answers. A named one lets you follow up. You cannot have both, so choose by what you intend to do with the answers.
- 6
Skip what does not apply
Conditional logic shortens the survey for each individual without removing questions. Someone who did not say they use a thing never gets asked how they like using it.
- 7
Split a longer survey into steps
Three short screens with a progress bar beat one long one. Partially filled responses are saved, so people can come back.
How many questions is too many
There is no fixed number, because motivation decides. A customer who just filed a complaint will sit through more than a random visitor. A practical guide: write the survey, then delete every question you could not write an action sentence for in step one. What remains is the right length.
When the survey genuinely has to be long, say up front how long it takes — and be honest. An underestimate is worse than no estimate: someone who discovers at minute seven that “two minutes” was untrue will close it.
How to read the results
The average is the most dangerous number in a survey. An average of 3 can mean everyone is lukewarm, or that half your respondents are delighted and half are furious — two completely different situations calling for completely different action. So always look at the distribution, not just the mean.
In Gatherino the summary shows the distribution for every question, with a chart for scales and NPS. Open answers can be browsed in the table, filtered and searched without diacritics. Everything exports to CSV, XLSX or JSON whenever you want to run the data through something of your own.
FAQ
How many questions should a survey have?
As many as you can state an action for. A typical customer survey lands at five to eight.
Should the survey be anonymous?
It depends on the purpose. Anonymity raises candour but rules out following up. Decide before you send it — you cannot change it afterwards.
How do I know the survey is too long?
From the partial responses. When they pile up at the same step, that step is the problem.
Is the average enough?
No. An average hides a split sample. Look at the distribution.
Related
Try Gatherino for free
Free plan: 3 forms and 100 responses a month. No credit card, EU-hosted data.
Get started free →