Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Conversion Rate Optimisation › Conversion Research

Reviewing Session Recordings and Heatmaps

How to review session recordings and heatmaps for conversion research: one question, one step and segment, a tally of behaviours, maps and privacy checks.
Tutorialintermediate10 min readUpdated

What recordings and maps can and cannot show

The conversion research methods guide places recordings and heatmaps in the behavioural group: they show what people did, not what they thought. A recording shows a visitor's scrolling, tapping, typing and hesitation on a page. A click map shows where many visitors clicked or tapped. A scroll map shows how far down the page each share of visitors reached, the same measure the scroll depth entry defines for analytics.

They are good at three things. They show where a page did not behave as a visitor expected, such as an element tapped repeatedly or a form field cleared and re-typed. They show whether the part of a page that answers the visitor's question is ever reached. And they show the sequence of actions before a visitor left, which a funnel report reduces to a single exit. They cannot say why a visitor did something, how common the behaviour is across all visitors, or whether the visitor would have converted without the problem. They also carry the reviewer's bias: as the methods guide puts it, people find what they look for, and a handful of dramatic sessions can feel like a pattern when it is not one.

The behaviours worth coding are the ones the methods guide lists, plus long pauses before leaving. Each has more than one possible cause, which is why a behaviour is counted and compared before it is interpreted.

BehaviourWhat it may signalWhat else could explain it
Repeated taps or clicks on the same elementIt did not respond as expected, or looks clickable and is notA slow response while the page was busy; a broken control on one device
Back and forth between two pagesA comparison the page did not supportOrdinary shopping behaviour on a considered purchase
Form fields cleared, re-typed or abandonedA validation rule or label the visitor could not satisfyA visitor checking a detail elsewhere before continuing
Clicks on headings, images or text that are not linksThe visitor expected more detail thereAccidental taps while scrolling on a phone
A long pause, then leavingDoubt, missing information or a price surpriseAn interruption unrelated to the page
Scrolling past the call to action and backThe visitor was looking for something before committingReading the page in full before acting, then acting

Before you start

  • A located step. A funnel analysis that names the step, device and segment losing the most people; the funnel analysis tutorial produces it, including a segment of the visitors who left the step, and a conversion research sprint runs this review straight after it.
  • Access to the tool and its settings. Not just to the recordings, but to the masking, filtering and loading settings, or to the person who controls them.
  • The site's consent decision for the tool. How the recording tag is classified in the consent banner and when it is allowed to load, from the privacy and consent guide and the site's own measurement plan.
  • A coding sheet. A table with one row per session and one column per behaviour, plus a column for notes and a link to the session.
  • Two hours of uninterrupted attention. Recordings reviewed in gaps between meetings are remembered, not counted.

The steps

  1. Write the viewing question

    Write one sentence that says what you are looking for, where and among whom: "What do phone visitors who reach the delivery step and leave do on that step before leaving?" The question comes from the funnel, not from a hunch, and it names a step, a device and a segment. Write beside it what you would expect to see if the leading explanation were true and what would contradict it. A question with no expected answer is still useful; a review with no question is the most common way recordings waste a day.

  2. Check consent, masking and page speed before collecting

    The recording tool is a tag like any other. Whether it may load before the visitor makes a choice follows the consent decision the site has made, as the privacy and consent guide explains for tags in general, and the tool belongs in the consent management platform under that decision. Recordings can capture what people type, and the GDPR, the EU regulation listed in this page's sources, defines personal data broadly, including names, identification numbers, location data and online identifiers, so mask every input by default, as the personal data guide asks of analytics in general, and check that order and account pages show no names. Finally, check what the script costs: the speed and mobile guide names session recording among the third-party tags that compete for the main thread and loads everything optional after the page is interactive, and the slow interaction diagnosis loads such scripts on idle or after the visitor interacts. This page is educational, not legal advice.

  3. Filter to one step, one device and one segment

    Apply the filters before watching anything: sessions that reached the located step, on the located device, from the segment the funnel pointed to, that left without converting. Add the period, so the batch reflects the page as it is now rather than before the last release. If the tool cannot filter to the step, use the page address and an exit from it. Write the filters at the top of the coding sheet, because a reviewer who changes filters halfway through is reviewing two different batches.

  4. Build a coding sheet of behaviours

    Turn the behaviours into columns: repeated taps on the same element, back and forth between pages, fields cleared or re-typed, clicks on things that are not links, long pauses before leaving, and any behaviour the viewing question predicts. Add a column for the element involved, a free-text note and the session link. Keep the definitions short and write them down, so that two reviewers would tick the same box for the same session. A behaviour that needs a paragraph to define is two behaviours.

  5. Set the evidence rule before watching

    Decide, with whoever owns the backlog, how many sessions in the batch must show a behaviour before it counts as a finding, and write the rule on the sheet. The number comes from the team's own judgement about the batch and the step, never from a published figure, and it is set before watching because a rule chosen afterwards always fits what was seen. Decide too when the review stops: when the batch is finished, or earlier if the tally is already clear in either direction. The academy research sprint checkpoint tests when a finding from one method is still a lead rather than evidence.

  6. Watch and tally without stopping to theorise

    Watch each session from arrival at the step to exit, tick every behaviour you see and note the element involved. Do not stop to explain a session or to search for the cause; an explanation formed early shapes what you notice in every session after it. Skip sessions that are plainly bots, tests or staff, and note how many you skipped. Where a session shows something new and important that is not on the sheet, add a column and keep watching, but record from which session the new column started so the count stays honest.

  7. Read click and scroll maps for the same segment

    Open the maps for the same step, device and segment, never for all visitors, because an aggregate blends phones and desktops, new and returning visitors, paid and organic traffic. On a click map, look for clicks on things that are not links, which show where visitors expected something to respond, and for controls that draw few clicks although the page depends on them. On a scroll map, find where the share of visitors still on the page falls away and what sits just below that point. As general practice, check what the map was drawn from: sticky headers, pop-ups and content that loads late can place clicks and scroll positions in the wrong spot, and the scroll depth entry notes that the same page is much longer on a phone than on a wide screen.

  8. Compare what lost and converted visitors did

    Repeat a shorter version of the tally for visitors who reached the same step on the same device and continued. A behaviour that appears as often among visitors who converted is part of how people use the page, not a cause of leaving. A behaviour that is common among those who left and rare among those who continued is the candidate finding. Compare the maps the same way: a section that converting visitors reach and leaving visitors do not is worth more attention than a section nobody reaches.

  9. Write the finding with its count and examples

    Write the finding in the what, so what, now what shape. What: the behaviour, the element, the count against the evidence rule and the comparison with converted visitors. So what: the step's reach from the funnel, and what the behaviour plausibly costs. Now what: the next method to confirm it and the change it suggests. Attach links to a few sessions that show it clearly, chosen after the count, not before. Leave the cause as a candidate; recordings show what happened, and a poll or a usability session is how you hear why.

  10. Hand it to the backlog and the hypothesis

    Add the finding to the research backlog in the conversion research hub with its location, evidence, reach and decision. The research backlog gives every finding the same fields, whichever method produced it. If it is a bug, such as a control that does not respond on one browser, fix it and close the row. If it needs a second source, plan the poll or the sessions at the same step. When the evidence holds, the hypothesis tutorial turns the row into a problem statement, a change and a metric. Where the finding is about a section visitors never reach, the anatomy of a landing page guide shows which job each section does and in what order.

A pottery shop's product-page review (illustrative)

Illustrative example — not a real result

Kilnquiver Pottery reviews its product pages on phones

Kilnquiver Pottery is an invented online shop selling handmade mugs, bowls and vases. Its funnel analysis shows that phone visitors who open a product page add to basket far less often than desktop visitors, on product pages that receive around 9,000 phone visits a month. The marketer writes the viewing question "What do phone visitors who open a product page and leave without adding to basket do on the page?" and checks the recording tool first: it loads only after consent under the site's banner, every input is masked, and the script is set to load after the page becomes interactive. She filters recordings to phone visits from the last fortnight that viewed a product page and left, agrees an evidence rule with the owner before watching, and codes each session. The tally clears the rule for one behaviour only: repeated taps on the glaze swatches, which look like buttons that switch the photo but are plain images. A shorter tally of visitors who added to basket shows the same taps far less often. The scroll map for phone visitors shows that only a small share reach the care and delivery details, which sit below a long maker's story, while the click map shows taps on the words "Dishwasher safe?" in a heading that does not expand. She records one counted finding, the swatch taps with the comparison and links to a few clear sessions, and a second row, the unreached care and delivery details, marked as needing a poll before it becomes a hypothesis. The swatch hypothesis proposes making each swatch switch the main photo, with add-to-basket rate on phones as the metric.

Common mistakes

  • Watching without a question. Unfiltered viewing produces stories, not counts.
  • Changing filters halfway. Two batches tallied as one give a count that means nothing.
  • Setting the evidence rule afterwards. A threshold chosen after watching always fits what was seen.
  • Reading maps for all visitors. Aggregates hide the segment where the problem lives.
  • Skipping the comparison. A behaviour as common among buyers as among leavers is not a cause.
  • Unmasked inputs. Recordings that capture what people type create personal data nobody needed.
  • Ignoring the cost of the script. A recording tag that loads first can slow the page it is studying.
  • Explaining instead of counting. The cause comes from a poll or a session, not from a reviewer's guess.
  • Quoting a borrowed number. How many sessions to watch is a team decision, not a published figure.

Frequently asked questions

How many session recordings should we watch?

No published figure settles it, and none should be borrowed. The methods guide's answer is enough to see a behaviour repeat under a specific question, which is usually a modest batch filtered to one step and one segment rather than hours of unfiltered viewing. Before watching, the team decides how many sessions must show a behaviour before it counts as a finding, keeps a tally while watching, and stops when the tally is clear either way. A few dramatic sessions are an anecdote; a counted pattern across a filtered batch is evidence.

Do session recordings need consent?

Treat the recording tool as a tag like any other. Whether it may load before the visitor makes a choice follows the consent decision the site has made for each tag, which the privacy and consent guide describes, and the rules differ by jurisdiction. Separately, recordings can capture personal data typed into forms, and the GDPR defines personal data broadly, so inputs are masked by default whatever the consent position. This page is educational, not legal advice; agree the specifics with whoever handles data protection in the business.

What does a rage click actually tell you?

That a visitor pressed the same thing repeatedly in quick succession, usually because it did not respond the way they expected. The cause can be an element that looks clickable and is not, such as an image, a heading or a swatch; a control that responds slowly because the page is busy; or a control that is broken on that device. A single rage click proves nothing. Count how often it happens at the step, check whether the element is meant to respond, and compare with visitors who converted before writing it down as a finding.

Terms used on this page

Frameworks

Where this page is used

Also related

Sources

  1. Regulation (EU) 2016/679 (GDPR) — EUR-Lex

Related