A survey is not a long form. A form processes one person and its answers trigger something for them; a survey measures a population, and a single response means almost nothing on its own. That difference decides everything about how you write one: the wording, the question types, the length, and what you do with the results.
If that distinction is new, survey vs form is the shorter read. This is the practical version.
1. Decide what one number you are tracking
Every survey worth running has a headline measure: overall satisfaction, likelihood to recommend, agreement with one statement. Pick it before you write anything else, phrase it carefully, and then never change the wording again. A rewrite between rounds resets your trend line, and a trend is the entire value of a survey. The absolute number on its own tells you very little.
Everything else on the survey exists to explain movement in that one number.
2. Work out who you are asking, and who will actually answer
You are not measuring your population. You are measuring the subset of it that fills in surveys, and the two differ in predictable ways: happier customers and more engaged employees answer more often. You cannot eliminate that, but you can avoid making it worse. Do not send a satisfaction survey only to the people who contacted support, and do not run it only in the week after a good release.
Note the sample size honestly, too. Fourteen responses is an anecdote with a percentage sign on it.
3. Use question types that aggregate
This is the practical difference between a survey and a form. A survey needs answers that can be added up:
- Linear scales for the tracked measure. Label both ends, keep the range fixed forever.
- A rating grid for a set of statements against one scale: several statements in one question, far quicker to answer than separate ones, and identical between rounds.
- Multiple choice for anything you want to count, and for the one question you branch on.
- Open text, sparingly. One or two, placed after a branch so the prompt is specific: “what would make the biggest difference to workload?” gets a usable answer where “any other comments?” gets silence.
Ten open questions is not a survey, it is a reading assignment. The question types page has the full set, and the survey maker page covers how they behave in practice.
4. Write the questions so they measure one thing each
The failure modes are consistent and easy to avoid:
Double-barrelled questions. “How satisfied are you with the speed and quality of support?”. Someone who found it fast and unhelpful cannot answer. Split it.
Leading wording. “How helpful was our new dashboard?” presumes helpfulness. “How would you rate the new dashboard?” does not.
Absolutes. “Always / never” options force people into a position they do not hold. Give the middle a home.
Vague quantities. “Regularly” means weekly to one person and quarterly to another. Ask for the actual frequency in bands.
Ranges that overlap. 1–5, 5–10, 10–20 puts two answers on every boundary. Check your bands.
5. Branch to keep it short, not to make it clever
Ask one choice question that identifies what someone is really rating or which area concerns them, and route each option to a short section of its own. Everyone answers the tracked question and then a handful of relevant ones, rather than everyone answering everything.
The constraint: branching works on multiple choice, checkbox and dropdown questions. A linear scale cannot route, so if you want to send low scorers somewhere different, put a short choice question next to the scale and branch on that.
6. Be precise about anonymity
Anonymous means you do not collect identity: no name, no email, and no team or tenure question either, unless the groups are large. Combinations identify people: role plus tenure plus team narrows a small organisation to one person even with no name attached.
Decide the answer before you write the survey, say it plainly at the top, and then honour it. Questionful respondents never need an account, so a link survey collects only what you ask for. Also resist filtering results down to groups of two or three afterwards, it undoes the promise you made to get the responses.
7. Keep it under two minutes, and run it on a cycle
Short and repeated beats long and annual, every time. Quarterly or twice a year, identical wording, same link shape, published results. Response rates in round two depend almost entirely on people seeing that round one changed something.
8. Know what your analysis can and cannot do
In Questionful you get a per-question response summary and a CSV export, plus an optional public summary that never publishes open text. That covers most satisfaction, feedback and internal surveys.
What it does not include: cross-tabulation, weighting, significance testing, or a respondent panel to buy a sample from. If your survey is genuine market research that needs those, do the analysis in a spreadsheet from the export, or use a research tool built for it. The SurveyMonkey comparison is honest about where that line falls.
Where to start
Pick the shape closest to your survey and edit rather than starting blank: customer satisfaction, employee satisfaction, market research, or event feedback. Each page carries the question set, the branching, and the reasoning behind both. And if you are running surveys as part of a wider feedback loop, the customer feedback use case covers what to do with the responses once they arrive.