What is survey design?

What makes a good on-site survey question?

Ask about behavior and history, not prediction. "What were you looking for today" is answerable from memory. "Would you buy this" asks the respondent to forecast their own behavior, and a forecast is not evidence.

One question at a time. Double-barreled questions such as "was the pricing clear and fair" cannot be answered when the two halves differ.

Open-ended where you need language. The purpose of an on-site survey in CRO is frequently to capture voice of customer vocabulary, and multiple choice returns your own words back to you.

Neutral wording. "How helpful was this page" presumes helpfulness. "What, if anything, was missing from this page" does not.

Short. On-site surveys compete with the visitor's actual task, and every additional question costs completion.

The highest-return single question in B2B CRO is "what nearly stopped you from getting in touch today", asked on the confirmation page. Respondents have already converted, so goodwill is high and the answer names an objection that nearly cost the sale.

When should a survey fire?

Placement decides the sample, and the sample decides what the answers can support. A survey answered only by converters describes only converters.

On the confirmation page. The highest response rate available, because the respondent is finished with their task. Its limit is built in: everyone who answers is someone the page already worked on.

On exit intent from a key page. Reaches non-converters, which is the population holding the information you most need. Lower response rates and higher annoyance cost. See exit intent popup for the tradeoffs.

After a threshold of engagement, such as 60 seconds or 50% scroll depth. Avoids interrupting people who arrived by accident.

Never immediately on load, which interrupts before the visitor has formed an opinion and depresses response quality along with conversion.

How do you run an on-site survey on a Webflow site?

Webflow has no native survey widget, so the survey arrives as a third-party script and the platform decides where it can fire.

Script placement. Site-wide code lives in Site Settings before the closing body tag and loads on every page. Keep the script site-wide and do the page targeting inside the survey tool, which is easier to maintain than pasting code into individual pages. Tools such as Hotjar cover this.

There is no confirmation page by default. A Webflow form swaps in its success state without changing the URL, so a survey configured to fire on a thank-you page never fires. Set an explicit redirect on the form, or trigger the survey on the success state instead.

A trigger set on a CMS template fires on every item in the collection, which is either the reach you wanted or a survey appearing on hundreds of pages you did not intend.

Which biases distort results?

Non-response bias. People who respond differ systematically from those who do not. Usually the most engaged and the most annoyed, with the indifferent middle missing entirely.

Acquiescence bias, the tendency to agree with statements as presented, which is why "do you agree that" phrasings are unreliable.

Social desirability, where respondents give the answer that reflects well on them. Weaker on anonymous on-site surveys than in interviews, and not absent.

Recall error on anything beyond the immediate session.

Small-sample overreading. Twelve responses is a source of hypotheses, not a measurement. Treating a percentage from a small survey as a finding is a routine error, and quoting it to a stakeholder as "34% of visitors say" is worse.

Two mitigations do most of the work: report raw counts rather than percentages until the sample is large enough that a percentage means something, and hand the output to a test rather than to a decision.

Related terms

Voice of customer · Usability testing · Exit intent popup · Segmentation analysis · Form abandonment

Service: Conversion Rate Optimization.

FAQ

How many questions should an on-site survey have?

One, in most cases. Each additional question reduces completion, and a single well-chosen open question returns more usable material than five closed ones.

What is a good on-site survey response rate?

Low by nature, frequently in the low single digits. The number matters less than whether respondents represent the group you care about, which is determined by when and where the survey fires.

Should on-site surveys be multiple choice or open ended?

Open ended when you need the customer's own language, which is the usual reason to run one in CRO. Multiple choice when you need to quantify something you already understand well enough to enumerate the options.