An online survey builder, and how to write a survey worth sending
The builder is the easy part. What separates a survey that changes a decision from one that fills a spreadsheet is question design — scale choice, wording, order, and length. This page covers both: the survey design rules that hold regardless of tool, and what Zunoform gives you free to run them.
Write the decision before you write the questions
The single most common survey failure is not a bad question. It is a survey nobody wrote a purpose for. Before you draft anything, finish this sentence: if the answer comes back X we will do A, and if it comes back Y we will do B. If you cannot finish it, the survey has no job, and every question in it is optional by definition.
That sentence does two useful things. It tells you which questions to cut — any question whose answer would not change A or B — and it tells you what your sample needs to look like. A survey that will justify a pricing change needs responses from people who pay you, not from everyone on the list.
Length is the tax you pay for each question you keep. Drop-off rises with every screen, and the people who abandon are not a random sample of your audience — they are the busiest and often the most valuable. Assume anything past roughly ten questions costs you response quality, and cut accordingly.
Three moves that fix most surveys
Do these before you send, in this order. Each one is cheap and each one catches a different class of error.
Cut to the questions that change something
Go through the draft and, for each question, write down what you would do differently depending on the answer. Delete every question where the honest answer is nothing. This routinely removes a third to a half of a first draft, and the response rate improvement is worth more than the data you cut.
Read every question aloud, looking for the answer you want
Leading questions are hard to see on the page and obvious out loud. "How helpful was our support team?" presupposes helpful. "How would you rate our support?" does not. Also strip double-barrelled questions — anything asking about speed and quality at once produces an answer that means neither.
Pilot it on five people who match your audience
Five is enough to catch every question that reads ambiguously, and it takes an afternoon. Watch where they hesitate rather than what they answer. A question someone has to read twice is a question that will be answered inconsistently at scale, and you will never see that in the results.
Choosing a scale, and how each one misleads you
Every response scale has a characteristic failure mode. Knowing them in advance is most of the skill.
Agree/disagree (Likert) scales
Prone to acquiescence bias — a meaningful share of respondents agree with statements by default, regardless of content. If you use them, mix in reverse-worded items so agreement is not always the positive direction, and be suspicious of any result where every statement scored well. Where you can, replace them with a construct-specific scale: instead of "the checkout was easy: agree/disagree", ask "how difficult was checkout?" on a difficulty scale.
0–10 and 1–5 rating scales
Cleaner than agree/disagree, but only comparable to themselves. A 7.8 average means nothing in isolation and everything as a trend line. Pick one scale and never change it, because changing the number of points mid-programme silently breaks every comparison you had. Odd-numbered scales offer a neutral midpoint, which is honest but attracts people who want to be finished; even-numbered scales force a lean.
Matrix and grid questions
Efficient on desktop, hostile on mobile, and the main driver of straightlining — respondents picking the same column down the whole grid to get out. If you keep a matrix, keep it to five rows or fewer, and check your data for identical response patterns before you trust it. On a conversational one-question-at-a-time layout the problem largely disappears, at the cost of a longer perceived survey.
Open text
The highest-value and highest-cost question type. One well-placed open text field usually teaches you more than ten rating questions, but only if someone actually reads the answers. Place it after the rating it explains ("what drove that score?"), keep it optional, and never open with it — an empty text box as question one is the fastest way to lose a respondent.
Ranking
People are bad at ranking more than about five items and the cognitive load shows up as drop-off. If you have twelve features to prioritise, do not ask for a ranking of twelve. Ask people to pick their top three, which is easier to answer and, in aggregate, gives you the same ordering.
Survey hygiene, as a pre-send checklist
Run this list against the draft before it goes out. Most of these are unrecoverable after send.
- Every question would change a decision — no idle curiosity questions
- No question asks about two things at once
- No question presupposes the answer you're hoping for
- One scale type, used consistently, and not changed between waves
- Demographic questions sit at the end, not the start
- Anything sensitive is optional, and says why you're asking
- The intro states how long it takes, honestly, and what you'll do with the answers
- You've tested it on a phone, not just a laptop
- Someone outside the project has read it cold
What running a survey costs across tools
Survey work has a specific shape — moderate volume, a need for branching, and a hard requirement that the form not look like it belongs to somebody else.
| What surveys need | Elsewhere | Zunoform |
|---|---|---|
| Enough free responses for a real sample | Typeform free stops at 10 a month and closes the form; Jotform free is 100 a month with a 500 storage cap that deletes the oldest | 500 a month free, 5,000 on Pro |
| Skipping questions that don't apply | Google Forms can only jump between sections, so follow-ups mean restructuring the survey | Free, per-question show and hide |
| One question at a time on mobile | Typeform does this well, within a 10-response free cap | Free, conversational layout |
| Your brand, not the tool's | Never on Google Forms; $56/mo tier at Typeform; from $39/mo at Jotform | Pro, $14.99/mo |
| Results out to a spreadsheet automatically | Usually free, sometimes tier-gated | Free, Google Sheets plus Zapier and webhooks |
| Segmenting respondents as they answer | Typically a manual pass over the export afterwards | Free, tag rules and scoring evaluated at submission |
Competitor details verified August 2026.
Scoring and tagging, for surveys that sort themselves
Most survey tools hand you a spreadsheet and leave the segmentation to you. Because scoring here runs server-side at ingest rather than in the browser, a survey can classify its own respondents as they submit: assign points per option, and the total is computed and stored on the response itself.
That turns a research survey into a routing mechanism. A readiness assessment can put every respondent into a band and show them the band on the ending page. A screener can score for fit and let you filter the export by number instead of reading three hundred rows. Tag rules do the categorical version of the same job — label a response based on what was answered, with a fallback tag when nothing matches.
The score is interpolatable anywhere in the form, so you can show someone their own result as part of the thank-you. That single change tends to lift completion, because the survey now gives something back rather than only taking.
Survey templates to start from
Edit the questions, keep the structure. All free to use.
A five-question satisfaction check after any customer interaction.
eNPS, satisfaction score, what's working, and an optional follow-up opt-in.
Buying habits, priority ranking, and feature importance for a market study.
A Likert matrix, usage questions, and an optional follow-up opt-in.
Questions people ask
How many responses do I need for a survey to be meaningful?
It depends on what you're deciding, not on a magic number. For directional internal decisions, thirty to fifty responses from the right people usually beats three hundred from the wrong ones. For anything you'll quote publicly or split into segments, you need enough in each segment to survive the split — which is where sample size stops being negotiable.
How long should an online survey be?
Shorter than your draft. Drop-off climbs with each additional screen and the people who abandon skew busy, which biases what's left. Ten questions is a reasonable ceiling for a cold audience; an engaged audience will tolerate more, but you should still ask whether each extra question would change anything.
Should I make questions required?
Only where a blank answer makes the response unusable. Required questions on sensitive or awkward topics don't produce honest answers — they produce abandoned surveys and false ones. Optional-plus-a-reason-why-you're-asking gets better data than mandatory ever does.
Can I run an NPS survey on the free plan?
The dedicated NPS question type is one of three that require a paid plan to publish, alongside e-signature and payment. Everything else — including rating scales, logic, scoring, and the conversational layout — is on the free plan.
Is a conversational survey better than a single-page one?
For mobile audiences and longer surveys, one question at a time generally holds attention better and avoids matrix straightlining. For short surveys sent to people at a desk, a single page is faster and lets respondents see the whole commitment up front, which some audiences prefer. Both layouts are available free here, so it's worth testing rather than assuming.
Keep reading
Better forms.
Better data.
Build your first form in under 60 seconds. Free forever for personal use, no credit card required.