How to create an online survey (free, step by step)

Making a survey is easy. Making one people finish, that tells you something you did not already know, is the hard part — and it is mostly decided before you write a single question. This guide walks through the whole thing, from working out what you are actually trying to learn to getting responses back.

1. Decide what decision the survey will inform

Start at the end. Write down the decision you intend to make once the answers are in — which feature to build next, whether to change a price, why people cancel. If a question would not change that decision whichever way it was answered, cut it. This single test removes more bad questions than any other, and every question it removes buys you completions.

Surveys that begin with "let's find out what customers think" produce data nobody uses, because there was never a decision waiting on it. Surveys that begin with a specific choice produce short, pointed questionnaires.

2. Keep it short — genuinely short

Completion falls off sharply past about ten questions, and the drop is not gentle. A five-question survey sent to a thousand people will usually out-collect a twenty-question survey sent to the same list, in absolute numbers, not just percentages. Length is the single biggest lever you control.

If you cannot get under ten, split the survey: ask the essentials now, and follow up with the people who answered. A respondent who finished once is far more likely to answer again than a stranger is to finish something long.

3. Choose question types deliberately

Open text boxes feel thorough and are the most common mistake. They are effortful to answer and slow to analyse, so people skip them and you end up with less data, not more. Use closed questions for the things you will count, and reserve one open question for the "why" behind the most important answer.

• Multiple choice — one answer from a known set. Add an "Other" box unless your options are genuinely exhaustive.
• Checkboxes — several answers. Say how many people may pick; "all that apply" and "top three" give different data.
• Rating or scale — strength of feeling. Keep one scale across the whole survey so answers stay comparable.
• Ranking — forces trade-offs that ratings hide, when everything comes back "very important".
• Long text — the reason behind a number. Ask it once, straight after the rating it explains.

4. Write questions that do not lead the answer

"How much did you enjoy the new dashboard?" assumes enjoyment. "What did you think of the new dashboard?" does not. Leading questions are comfortable to write and produce flattering, useless results — you learn what you hoped rather than what is true.

Watch for double-barrelled questions too. "Was the support fast and helpful?" cannot be answered by someone who got a quick, unhelpful reply. Ask one thing at a time.

5. Order questions so momentum builds

Open with something easy and clearly relevant. Personal details — email, job title, company — belong at the end, once the respondent has invested effort and can see the survey was worth it. Asking for contact details first is the most reliable way to lose people on question one.

6. Test it on a phone before you send it

Most responses will arrive on a phone. Matrix questions in particular go from compact on a laptop to unusable on a 390-pixel screen. Fill your own survey out on a handset before sending — it takes two minutes and catches problems no amount of previewing on a desktop will.

7. Share it where people already are

A link in an email people expect beats a link posted somewhere they might see. If you embed the survey on a page, embed it inline rather than behind a popup — inline forms are visible to anyone who scrolls, and are not dismissed reflexively.

8. Watch drop-off, not just responses

A completion percentage tells you something is wrong. Per-question drop-off tells you where. Almost always it is one specific question — a required field someone cannot answer honestly, a request for personal details too early, or a matrix on mobile. Fix that one question and the rest of the survey usually recovers on its own.