Your analytics can tell you that 68% of visitors leave the pricing page without clicking anything, that mobile visitors abandon the booking form at step two, and that people who watch the demo video convert twice as often. What it cannot tell you is why. Are people confused by the pricing tiers? Worried about a hidden fee? Comparing you with a competitor? Simply not ready yet?
Guessing the answer is how teams waste months testing button colours. Asking is faster. A single well-placed question on the right page, shown to the right visitors at the right moment, will often surface the objection that is costing you sales, in the visitor's own words.
This guide covers the types of on-site surveys and polls and when to use each, how to write questions that get honest answers, how to target and time them, how many responses you need, how to analyse open-ended answers, and how to turn what you learn into tests that move conversion rates.
What analytics records and what it can't explain
Behavioural data and survey data answer different questions. You need both, and the most common mistake is expecting one to do the other's job.
| Question | Analytics and session recordings | On-site surveys |
|---|---|---|
| Where do visitors drop off? | Yes, precisely | No |
| How many people do it? | Yes | Only within the sample who answer |
| Why did they leave? | Only by inference | Yes, in their words |
| What were they trying to do? | Partly (pages viewed, searches) | Yes |
| What almost stopped buyers from buying? | No | Yes (post-purchase survey) |
| Which competitor are they comparing? | No | Yes |
The practical sequence is: use analytics to find the leak, use recordings to see the behaviour, then use a survey to ask about the reason. Semrush's guide to conversion optimisation makes the same point about collecting data from users:
“User data helps you identify why visitors leave without converting.”
— Semrush, Conversion Rate Optimization
If you haven't set up session recordings yet, our guide to Microsoft Clarity covers a free option, and it pairs well with the survey methods below. Make sure your analytics data is clean first, too; our GA4 audit checklist catches the tracking problems that make funnel drop-offs look worse (or better) than they are.
Five on-site survey formats and the question each answers
| Format | Where it appears | Best question it answers | Typical length |
|---|---|---|---|
| On-page poll | Small widget in a corner of a specific page | "Is anything stopping you from signing up today?" | 1 question plus optional follow-up |
| Exit-intent survey | When the cursor moves to leave, or after inactivity on mobile | "What made you decide not to book today?" | 1 to 3 questions |
| Post-purchase or thank-you survey | Confirmation page or confirmation email | "What nearly stopped you from buying?" | 2 to 6 questions |
| Intent survey | Shortly after landing on key pages | "What brings you here today?" | 1 question, multiple choice |
| Feedback tab | Persistent button on every page | Bugs, confusion, broken pages | Visitor-led |
The post-purchase survey is the most underrated. Buyers have just overcome every objection, they're in a good mood, and they know exactly what almost stopped them. Their answers tell you which objections to address earlier in the journey for people who didn't buy. A source question ("Where did you first hear about us?") on the same survey also captures word-of-mouth and podcast influence that click tracking misses; see our guide to capturing source data in your CRM for how to combine the two.
Exit surveys need care on mobile, where there is no cursor to track. Most tools use triggers such as a fast scroll up or a period of inactivity instead. For the design side of interrupting visitors without annoying them, our guide to website popups that convert applies equally to survey widgets.
Writing survey questions visitors answer honestly
Most bad survey data comes from bad questions. Visitors give the answer the question invites, skip questions that take effort, and abandon surveys that feel like homework. A few principles fix most of it:
- Ask about behaviour and specifics, not opinions in general. "What information is missing from this page?" beats "How would you rate this page?"
- Use open questions to discover, closed questions to measure. Start open; once you know the common answers, convert them to multiple choice with an "Other" option.
- Ask one thing at a time. "Was the pricing clear and fair?" is two questions; someone who found it clear but expensive can't answer.
- Remove leading words. "How easy was our simple checkout?" tells people what to say.
- Don't ask people to predict the future. "Would you buy if we added X?" produces optimistic yes answers that don't hold up.
- Keep the language of the page. If your page says "plan", don't ask about "subscription tiers".
| Weak question | Problem | Better version |
|---|---|---|
| How satisfied are you with our website? | Vague; produces a number you can't act on | What, if anything, is stopping you from starting a trial today? |
| Did you find our prices competitive and clear? | Two questions in one; leading | Is anything about the pricing unclear? |
| Would you use a live chat feature? | Prediction; people say yes | Did you have a question you couldn't find an answer to? What was it? |
| Why didn't you buy? | Feels confrontational | What made you decide not to order today? (Any answer helps.) |
| How did you hear about us? (12 options) | Too many options; first ones picked most | Where did you first hear about us? (open text, or 5 to 6 short options plus Other) |
Semrush's CRO guide suggests exit questions in the same spirit, for example "What stopped you from completing your purchase today?". For more on the small words around forms and buttons, see our guide to microcopy.
Targeting and timing: who sees which survey, and when
A question shown to everyone, immediately, on every page produces lots of noise and annoys people. Target narrowly:
| Goal | Page | Who | Trigger |
|---|---|---|---|
| Find pricing objections | Pricing page | Visitors who have been on the page 30+ seconds | Time on page or scroll depth |
| Find checkout friction | Cart and checkout steps | Visitors who start but don't complete | Exit intent on cart; never mid-payment |
| Understand paid traffic mismatch | Paid landing pages | Visitors from a specific campaign (UTM) | After 20 seconds or on exit |
| Understand intent | Home and key category pages | New visitors | After 10 to 15 seconds |
| Find near-miss objections | Thank-you page | New customers | On load |
Add a frequency cap so nobody sees the same survey twice, and exclude logged-in customers from prospect surveys. Exclude your own team too. If you've defined internal traffic for analytics, the same office IP list can usually be reused in the survey tool.
Segmenting by traffic source is especially revealing for paid campaigns. If visitors from one ad group keep saying "I was looking for something else", the problem is the keyword or ad promise, not the page. Our guide to PPC landing pages covers fixing message match once you've found it.
How many responses you need before acting
For open-ended discovery questions, you are looking for themes, not percentages. Read responses as they arrive and keep a running list of themes. When new responses stop adding new themes, you have enough to act on, which for a focused question on one page is often somewhere around 50 to 150 answers. If one objection is mentioned again and again in the first 40 answers, you don't need 400 to know it matters.
For closed questions where you want to compare proportions ("Is price or features the bigger concern?"), you need more responses, and you should be cautious about small differences. A split of 52% versus 48% from 100 answers tells you very little; 70% versus 30% tells you a lot.
Two biases to keep in mind:
- Non-response bias. The people who answer are not a random sample of visitors. Frustrated and very happy people respond more.
- Stated versus real reasons. "Too expensive" sometimes means "I don't see enough value" or "I'm not the decision maker". Follow-up questions and recordings help separate them.
Turning open-ended answers into themes you can count
Open text is where the value is, and also where most teams stall. A simple coding process gets you from 200 messy answers to a ranked list of objections in an afternoon:
- Export all responses with page, device, traffic source and date.
- Read the first 50 without coding, noting recurring ideas.
- Create 6 to 12 codes from those ideas (for example: price unclear, shipping cost, trust, missing info, comparing options, not ready, technical problem).
- Code every response, allowing more than one code per answer. Add new codes only if an idea recurs.
- Count codes and break them down by device and traffic source.
- Pull three verbatim quotes per top code; these are what persuade stakeholders.
- Have a second person code a sample of 30 to check you agree. Large disagreements mean your codes need clearer definitions.
AI tools can speed up the first pass of coding, but check their work against a sample you code yourself; they tend to create too many categories and miss sarcasm.
Example: coded exit survey on a hypothetical online course checkout, 180 responses.
| Code | Responses | Share | Typical verbatim |
|---|---|---|---|
| Unsure if course suits my level | 52 | 29% | "Not sure if this is too basic for me" |
| Price or payment | 41 | 23% | "Can I pay monthly?" |
| Comparing with other courses | 30 | 17% | "Looking at a couple of others first" |
| Not ready yet | 27 | 15% | "Will come back after the holidays" |
| Refund and trust concerns | 18 | 10% | "What if I don't like it?" |
| Technical problem | 12 | 7% | "Card kept getting declined" |
The top objection, uncertainty about level, was not visible anywhere in analytics, and nobody on the team had guessed it.
From visitor objection to A/B test hypothesis
Each top theme becomes a hypothesis in a simple format: "Because [evidence], we believe [change] for [audience] will [effect], measured by [metric]." Keep each test focused. Semrush's A/B testing guide gives the same advice:
“When running your first A/B test, focus on testing one variable to clearly see how changes affect results.”
— Semrush, A/B Testing
From the course example above:
- Level uncertainty (29%): add a "Who this course is for, and who it isn't for" block and a two-minute sample lesson near the price. Primary metric: checkout completion.
- Payment (23%): test showing the instalment option on the course page, not only at checkout.
- Refund concerns (10%): move the guarantee next to the buy button, in plain words.
- Technical problems (7%): not a test; a bug to fix immediately.
Not every finding needs a test. Bugs and missing information should simply be fixed. Save testing for changes where you genuinely don't know the outcome, and make sure you have enough traffic to reach a reliable answer. Our guide to A/B test statistical significance explains how long to run a test, and conversion rate optimization for beginners covers the wider process.
In the example, the "who this is for" block and sample lesson lifted checkout completion from 3.1% to 3.8% over four weeks. That's a hypothetical figure, but the pattern is common: the biggest wins come from answering the question visitors actually had.
Keeping surveys from damaging the experience they measure
- Never interrupt payment. No popups between entering card details and confirming.
- Make dismissal easy and permanent for that visitor.
- Check mobile layouts. A widget that covers the buy button on a small screen does more harm than any insight is worth.
- Be clear about data use. If you ask for an email address to follow up, say why, and treat survey responses as personal data where they can identify someone.
- Don't collect what you won't read. If nobody will code the answers this month, pause the survey.
- Close the loop. Share findings with sales and support; the same objections usually appear in their conversations. Our guide to win-loss analysis covers how to collect them systematically from deals.
For existing customers, use separate satisfaction measures rather than conversion surveys; our guide to NPS and CSAT explains the difference.
Questions about running on-site surveys and polls
Do on-site surveys hurt conversion rates?
A badly timed survey can, especially one that interrupts a checkout or appears the moment someone lands. A small, well-targeted survey that appears after meaningful engagement, or on exit, rarely has a measurable effect. If you're worried, show the survey to a random half of eligible visitors and compare conversion rates.
How many questions should an on-site survey have?
One for an on-page poll, up to three for an exit survey, and no more than five to eight for a post-purchase survey. Each extra question lowers completion. Make the first question the most important one, because many people will answer only that.
Should I offer an incentive for answering?
For one-question polls, no; the effort is too small to need it and incentives attract careless answers. For longer post-purchase or customer surveys, a small incentive such as a prize draw can lift completion. Make sure any prize draw follows the rules where you operate.
What tools can run on-site surveys?
Dedicated survey and feedback tools such as Hotjar, Qualaroo and Survicate handle targeting and triggers well. Many popup and CRO platforms include polls, and form builders can host longer surveys. Pick one that can target by page, device, behaviour and traffic source, and that exports responses with those attributes.
How is this different from NPS or CSAT?
NPS and CSAT measure how existing customers feel, usually after they've bought. On-site conversion surveys ask prospective customers what is stopping them now. Both are useful, but they answer different questions and should be analysed separately.
Launching your first conversion survey this week
Pick the one page where your funnel leaks most. Write one open question in the visitor's language ("What, if anything, is stopping you from... today?"), target it to engaged visitors on that page, cap it at one showing per person and run it until new answers stop adding new themes. Code the answers, fix anything broken, and turn the top objection into your next test.
If you'd like help turning visitor research into conversion gains, our conversion optimization team runs surveys, recordings and testing programmes for sites of every size. Get a quote to find out what's stopping your visitors.
Further reading on on-site surveys
- Capturing UTM and Source Data in Forms and Your CRM
- How to Estimate Competitor Website Traffic (and How Accurate It Is)
- GA4 Audit Checklist: Make Sure Your Data Can Be Trusted
