The Complete Guide to Creating an Online Survey: From Question Design to Analyzing Results

An online survey is the cheapest way to hear the voice of a customer, an employee or a student. Yet most surveys never lead to a decision, because they asked the wrong question, or asked the wrong people, or read the results wrong. This guide is the path an experienced researcher walks for every survey, from the decision sentence to the one-page executive summary, with worked numbers, tables and a checklist. If you want to start right away, you can create an online survey in Porsino for free in a few minutes; if you're in a hurry, skip straight to the pre-launch checklist.

Step 1: Before any question, write the decision sentence

A good survey doesn't start with "what do we want to know." It starts with "what will we decide after this survey." Complete this sentence:

"If the results show that ……, then we will ……."

If you can't write this sentence, whatever data you collect will only ever be a pretty chart. The decision sentence pins down three things at once: the primary metric (which number?), the threshold (above or below what?) and the action.

Goal	Survey type	Primary metric	Example decision sentence
Satisfaction after a purchase or service	Transactional CSAT	Percent satisfied (4 and 5 out of 5)	If a branch's CSAT is below 70%, retrain that branch
Loyalty and recommendation	Relational NPS (quarterly)	% promoters − % detractors	If NPS stays below 20, launch a call-back program for detractors
Employee satisfaction and engagement	Monthly pulse + annual	Mean item score by department	A department below 3 out of 5 for two consecutive months gets a root-cause meeting
Choosing among several options	Ranking / prioritization	Borda score + "first choice"	The top three options go into next year's budget
Understanding a market or a need	Exploratory questionnaire	Response distribution + open question	If more than 40% mention problem X, X becomes the product priority

One strict rule: any question that doesn't serve the decision sentence gets cut. It's the only way to keep a form short.

Step 2: Whom do we ask, and how many?

Two separate decisions: the population (everyone whose opinion matters) and the sample (the people we actually ask). If the population is small, say 80 employees, ask everyone and aim for a high response rate. If it's large, calculate a sample size.

Cochran's formula in one minute

For a large population, with a 5% margin of error and 95% confidence:

n = (z² × p × q) / e²  =  (1.96² × 0.5 × 0.5) / 0.05²  ≈ 385

In other words, about 385 valid responses are enough, whether the population is one million or ten million. If the population is finite (say 2,000 customers), the finite population correction brings that down to about 322. To run the formula with your own numbers, use the Cochran sample size calculator; to find out whether a two-point difference is real or noise, there's the significance test.

Population size	Sample needed (5% error)	Sample needed (3% error)
100	80	92
500	218	341
2,000	322	696
10,000	370	965
Infinite	385	1,068

And a practical warning we cover in detail in "How many responses are enough": below 100 responses, a handful of people switching sides moves NPS by tens of points, so don't talk about a "trend" until you've reached a sufficient number.

Who answers matters more than how many

A 300-person sample drawn only from this week's happy customers is worth less than a random sample of 80. If you post the link only in your fan channel, the number you get is the fans' opinion, not the market's. The practical route: build a contact list and send personal invitations to a random sample, so that coverage is random and each person answers exactly once.

Step 3: Question design: the rules that save your results

Which question type for which job

Question type	When	Watch out for
Single choice	When the options are mutually exclusive (one pick)	Options must be exhaustive and mutually exclusive; don't forget "Other"
Multiple choice	"Which of these do you use"	Cap the number of selections (max 3), or people tick everything
5- or 7-point Likert scale	Agreement / satisfaction / importance	Label every point, not just the two ends (5 or 7?)
NPS (0 to 10)	Loyalty; once per survey	Don't average it; promoters minus detractors (why)
Star rating	Quick satisfaction on mobile	5 stars; make the meaning of 3 stars clear with a label
Matrix	Several items on one scale	At most 6 rows and 5 columns; on mobile it becomes cards (where it's a disaster)
Ranking	Prioritizing 3 to 7 options	Nobody ranks more than 7 options properly
Open-ended (text)	"Why" and "what should we change"	At most 2 open questions, preferably optional and at the end

Seven rules for writing questions

1. One concept per question. "How were the speed and quality of support?" is two questions; someone who got fast but poor service has no answer.
2. Ask neutrally. "How satisfied are you with the beautiful new design?" contains its own answer. We walk through five types of bias, with examples, in this article.
3. The respondent's language. "User experience" means nothing to a shop's customer; ask "how easy was it to find the product."
4. A specific time frame. "In the past three months, how many times…" rather than "how often do you usually…".
5. An exit option. Offer "Don't know / Not applicable" on questions not everyone has experienced; otherwise they pick the middle of the scale at random.
6. Funnel order. From general to specific, from easy to sensitive; demographic questions (age, gender, department) at the end.
7. Consistent numbers and scales. If 1 means "strongly disagree" in one place, it means that everywhere; flipping the direction mid-form corrupts the data.

"Bad → good" examples

Problem question	Problem	Better version
Were you satisfied with our fast, professional service?	Two concepts + leading	How long did you wait for a reply? / How satisfied were you with the agent's manner?
How often do you visit the site?	No time frame	How many times did you visit the site in the past week? (0 / 1–2 / 3–5 / more)
What do you think of the price?	Vague	The price relative to the value you received: far too high / a bit high / fair / a bit low

Step 4: Length, structure and the respondent's experience

• Target time: 3 to 7 minutes, roughly 10 to 20 questions. State the approximate time in the introduction; someone who knows it's "4 minutes" is less likely to abandon halfway.
• The most important question among the first 3. Drop-off is highest after the first page; if NPS is critical for you, ask it right at the start.
• Conditional logic instead of "if you answered no, go to question 12." Someone who "didn't buy" shouldn't see the purchase-experience questions; with conditional logic the form gets shorter for each person.
• Mobile first. More than half of responses come from phones; wide matrices, big tables and long texts get abandoned right there. The "card" or "conversational" layout (one question at a time) gives the best completion rate on mobile.
• A progress bar and meaningful pagination. One topic per page; a short heading at the top of each page.
• Required questions only when truly necessary. Every extra required field costs a few percentage points of drop-off.

Step 5: Putting it into practice in Porsino, step by step

1. Start from a template. The template gallery has more than 300 ready-made forms; customer satisfaction, employee satisfaction, NPS, course and event evaluation are all there, and you only change the brand name and a few items.
2. Lay out the fields. Stars, scales, NPS, matrix, ranking, capped multiple choice, open text, all with Persian-language validation and the Jalali (Persian) calendar. Half- or third-width fields let you place short questions side by side.
3. Logic and quotas. Conditional display of questions, jumps between pages, capacity limits on options (for example, "at most 30 people for the morning shift").
4. Anonymous or identified? For employee surveys, switch on anonymous mode so that IP addresses and device details are not stored, and respondents see that on the form; for customers, switch on "one response per mobile number / national ID" to keep out duplicate votes.
5. Design. A ready-made theme or your own colors and logo; a two-column layout for desktop, a plain card style for formal forms.
6. Mobile preview and testing. Fill in the form on your own phone and time it; then have someone outside the team fill it in without any explanation and watch where they hesitate.

You can build all of this right now in Porsino's online survey builder; the free plan is enough to get started.

Step 6: Distribution: where and when to post the link

Channel	Typical response rate	Best for	Tip
SMS with a short link	High (often double digits)	Iranian customers, right after a transaction	Send it the same day as the experience, not a week later
Email	Medium	Organizations, B2B customers	Short subject line; one reminder after 3 days
Single-use invitation	High and trackable	When each person must answer exactly once	You see who hasn't answered and remind only them
QR code on a receipt / table / banner	Low but targeted	Restaurants, trade shows, branches	Put a motivating line next to it: "20 seconds, no name needed"
Embedded in a site or app	Depends on placement	In-the-moment feedback	Show it after a key action, not at login
Kiosk (a tablet on site)	High on location	Clinics, bank branches, reception desks	Kiosk mode resets itself after every response

Timing: for customers, the same day as the experience; for employees, mid-week rather than Saturday morning (the start of the Iranian work week); for research, a 7-to-10-day window with one reminder on day four. Once you've reached your sample size, close the form so that late, different-looking responses don't shift the result.

Step 7: Nine techniques for raising the response rate

1. In the introduction, say why you're asking and what you'll do with the results.
2. State the real time ("3 minutes") and stick to it.
3. Send it from a real person's name, not "the Quality Department."
4. Make the first question easy and inviting (a star rating).
5. Test on mobile; wherever there's horizontal scrolling, redesign.
6. One reminder, only to those who haven't answered (possible with personal invitations).
7. Genuinely guarantee anonymity, and say that you have.
8. Announce the results of the previous round ("You said the line was too long; we opened a third counter"). This is the strongest motivator for the next round.
9. A prize draw or small reward only when it won't create bias; for satisfaction surveys it's usually unnecessary.

Step 8: Analysis: what to look at and what not to

Every metric has its own reading

• CSAT = the percentage of people who gave 4 or 5 (out of 5). Not the mean. A mean of 3.8 could be "everyone gave 4" or "half gave 5 and half gave 2," two completely different situations (why the average sometimes lies).
• NPS = % promoters (9 and 10) minus % detractors (0 to 6). With a small sample, write the margin of error next to the number.
• The distribution, not just the mean. Look at the distribution chart for every item; a bimodal shape means you have two different groups that must be analyzed separately.

From "what" to "why": key drivers

When you have one overall metric (say, overall satisfaction) and several specific items (speed, price, manner, quality), a "key drivers" analysis tells you which item moves most in step with the overall metric while scoring low. That's where you should invest. An important, low-scoring item is an "improvement priority"; an important, high-scoring item is a "strength" you should highlight in your marketing. Porsino builds this report automatically, and it's described in the reports guide.

Segmentation, significance and data quality

• Segment by department, branch, city or age group: an "overall" problem is often one branch's problem.
• Significance: the difference between 68% and 72% with 50 responses per group is not significant; before getting excited, run a two-proportion test.
• Suspicious responses: a response completed in 20 seconds, or one that ticks the same column in every row of a matrix, is a sign of carelessness; look at the report once with them and once without.
• Open-ended questions: categorize them (price / speed / manner / …) and count the frequency of each category; one real quote next to each number brings the report to life.

The report for management: one page

A manager won't read charts for 20 questions; they want to know how things stand, where the risks are, and what the next three actions are. Porsino's "executive summary" builds exactly that: a one-line verdict, a health score, strengths and risks, and three actions, downloadable as a PDF with proper Persian fonts, alongside the full per-question report and an Excel export for further analysis.

Step 9: Act, and close the loop

A survey that doesn't lead to action gets a lower response rate next time; people realize they aren't being heard. Three simple things:

1. An owner and a deadline for every "improvement priority."
2. Announce the result and the action to the same audience (even a two-line text message).
3. Repeat periodically with the same questions so the trend becomes visible; don't change the main question, or you lose comparability.

The 20-point checklist before you send

1. The decision sentence is written.
2. Every question connects to one of the decisions.
3. Completion time is under 7 minutes and stated in the introduction.
4. No question contains two concepts.
5. No question is leading.
6. All scales run in the same direction and are labeled.
7. Single-choice options are exhaustive, mutually exclusive, and include "Other."
8. Multiple-choice questions have a cap.
9. Sensitive and demographic questions are at the end.
10. A "Don't know / Not applicable" option is there wherever it's needed.
11. Irrelevant questions are hidden with conditional logic.
12. At most two open-ended questions, both optional.
13. Tested on mobile; no horizontal scrolling.
14. Anonymous vs. identified has been decided and is stated on the form.
15. One response per person is enabled (where needed).
16. A closing date and a response cap are set.
17. The thank-you message says when and how the results will be shared.
18. Someone outside the team has filled in the form with no guidance.
19. The required sample size is known, and you know when you'll close the form.
20. You know who will read the results and in what format.

Seven common mistakes that ruin everything

• Asking whoever happens to be around, then generalizing to everyone.
• Averaging NPS or an ordinal scale and reporting "3.7 out of 5" without the distribution.
• The 40-question form, "because since we're asking anyway…".
• A required open-ended question at the start of the form.
• Comparing results with last year's survey, whose questions were different.
• Leaving the form open indefinitely and mixing three months of responses together.
• A 30-page report without a single sentence saying "so what do we do."

Frequently asked questions

Is a free account enough for a survey, or do I need to pay?

To start, free is enough; what pays for itself is the analytical reports (key drivers, segmentation, executive summary) and the distribution tools (invitations, SMS), not the form itself. In Porsino, building and publishing a survey is free, and these extras open with one of the monthly, quarterly or yearly plans whenever you need them, with no automatic renewal.

How many questions is the right number?

Ten to twenty questions, three to seven minutes. If you need more, split it into two surveys spaced apart, or use conditional logic so each person only sees the part that applies to them.

Where should I put the open-ended question?

One right after the primary metric ("What's the main reason for your score?") and one at the end ("Is there anything we didn't ask?"). Both optional.

How do I prevent duplicate responses?

Single-use invitations are the best way; if the link is public, use "one response per mobile number" with SMS verification, or at the very least one response per device.

Does an employee survey really stay anonymous?

When anonymous mode is on, the IP address, browser and device fingerprint are not stored, and respondents see this on the form. For departments with fewer than 5 people, don't show results by department, so nobody can be identified.

How should I present the results after collection?

One executive page (status, risks, three actions) plus the full appendix. Prefer the distribution chart to the mean, and write the sample size next to every number.

---

Ready? Start with Porsino's online survey builder or add one of the survey templates to your account with one click. If you're building a research or thesis questionnaire, read the guide to designing a standard questionnaire.