How to Run a Hotjar Survey That Produces Fixable Answers

October 31, 2026

A Hotjar survey earns its keep when each response can become a ticket. That requires three decisions before you launch: who sees it, on which page, and what single question they answer.

Why do most on-site surveys return nothing useful?

Because they ask for an opinion instead of a report. “How was your experience today?” produces thumbs-up icons and silence. “What were you trying to do when you got stuck?” produces the sentence you need.

The second shape works because it assumes a failure. Nobody is being asked to grade you; they are being asked to describe a moment.

Hotjar‘s polling and survey widgets support open text, rating scales and multiple choice. Open text is the one that generates findings, and it is the one people avoid because the responses need reading.

How do you target a survey so responses are comparable?

Three targeting rules, applied together:

  1. Page-specific — trigger on the page where the task happens, not sitewide.
  2. Behaviour-based — show after a specific event, such as three failed searches or a 45-second dwell with no click.
  3. Excluded segments — hide from employees, from anyone who answered in the last 90 days, and from accounts created today (their answers are about the signup flow, not the product).

Sitewide exit-intent surveys with no targeting produce a mix of every page and every intent, and the data becomes a mood, not a finding.

What question should you ask first?

One question per survey, and this is the best first one: “What were you trying to do just now?” An optional follow-up: “What stopped you?”

Rating scales are cheaper to analyse and much less useful — they tell you a number moved, not why. Use ratings only as a filter: show the open-text question to people who rated 1 or 2.

Avoid asking “how can we improve” as a first question. It invites feature requests from people who were never going to be customers.

How many responses do you need before acting?

Fewer than you think for qualitative signals. Fifteen to twenty responses to an open-text question surface recurring themes reliably; after that you mostly hear variations.

That means a sample rate of 10–20% is usually enough, and a higher rate just annoys visitors. On the free Basic plan at 35 sessions a day, a 20% sample gives you a handful of responses a week — enough for one page, not for a whole site.

Goal Sample Time to signal
Find a broken flow 10–15 responses 1–2 weeks
Rank known complaints 50+ responses 3–4 weeks
Track sentiment movement 200+ responses Quarterly

How do you turn responses into work?

Read them in one sitting, weekly, and tag each one into a cause bucket. Three buckets are usually enough: confusing label, missing capability, technical failure.

Then check the bucket against recordings. A response saying “the button didn’t work” needs the recording to establish whether it was a misclick or a genuine error — Hotjar puts recordings and surveys in the same product, so this is one click, not a project.

Route technical failures to engineering, confusing labels to design, missing capability to the roadmap. If a bucket gets more than five responses in a month, it becomes a backlog item with evidence attached.

What does Hotjar not tell you?

Whether the people who answered are representative. Survey respondents skew toward either very happy or very annoyed users. Pair the qualitative layer with funnel numbers from Amplitude or PostHog so you know what share of users hit the problem at all.

It also cannot see behind a login problem. If the survey never renders for a user whose session broke, you get silence from exactly the people you needed.

Are paid plans worth it for surveys?

At around $32/month for the entry paid tier, yes if surveys are the reason you picked Hotjar over free Clarity. If you are only doing heatmaps, the free 35-session cap covers a single page and paid is hard to justify.

Key Takeaways

  • Ask “what were you trying to do?”, not “how did we do?” — assumptions of failure produce specifics.
  • Target by page, trigger by behaviour, and exclude employees and repeat respondents.
  • 15–20 open-text responses is enough to surface recurring themes.
  • Tag responses weekly into three buckets and pair each with a recording.
  • Respondents skew to extremes; use funnels to size the problem.
Tags: