Settle heatmap sample size before you change a page because of a map. No single line makes a conclusion certain, but rough thresholds stop you from acting on a few dozen clicks. What matters most is where you count: per page and per device type, not across the whole site. A hundred visits on phones plus a hundred on desktops are two small samples, not one average.
This guide is about how much data you need. How to interpret colours, overlooked CTAs and false affordances is covered in how to read heatmaps without guessing. Here you will decide when a map is dense enough, what to do when a small site never reaches the threshold, and how to confirm a pattern on individual visits.
The short answer: count per page and per device
A tool may show you a map even when traffic is very low; Microsoft Clarity's documentation says there is no minimum traffic limit. That is convenient, but deciding when a map is ready for conclusions is up to you. A colourful image built from twenty visits looks just as convincing as one built from a thousand.
Microsoft Clarity's documentation says a heatmap can be generated without any minimum traffic, including for a group of pages. The source describes how the tool behaves, not how many visits you need for a reliable conclusion. Microsoft Learn: Heatmaps overview.
The number that matters is the views of one page on one device type in the selected period. A phone has a different layout, section order and first screen than a desktop, so mixing those visits produces a map that describes no real situation. If the page layout changed during collection, count from the day of the change, not from the start of the month. A homepage carousel adds one more caveat: the map background shows a single slide, so you check clicks on the others in the element ranking, as explained in our guide to the homepage carousel.
Rough thresholds and what they are good for
The thresholds below come from a Zoho PageSense guide and are a starting point, not a significance test. The same guide points out that heatmaps are not statistical tools and offer no p-value or significance threshold. Treat them as a minimum density, not proof.
Scroll the table sideways to see all columns.
| Goal of the analysis | Sample per page and device | What it supports |
|---|---|---|
| Spotting a technical bug | Individual visits confirmed by a recording | Fixing it without waiting for a map |
| A first click pattern | 200 to 300 visits | Writing down a hypothesis to check |
| Scroll map | Around 200 visits | Judging whether key content sits too low |
| Changing or removing an element, such as in checkout | Around 500 sessions | Deciding to change an important element |
Zoho PageSense gives 200 to 300 visitors as the minimum at which click patterns begin to form, suggests around 500 sessions before changing or removing an element, and recommends counting devices separately. This is practitioner guidance, not a study result or a statistical significance threshold. Zoho PageSense: How much data do you need.
The first row is the exception to the whole idea of thresholds. A button that does not respond on phones is a bug even if you see it in a single visit. You do not need a sample to fix a fault; you need one to claim that most visitors overlook something. How to read scroll depth around prices and CTAs is covered in the guide to page scroll depth.
Work out how many days the sample will take
Before reading a map, check recent page views and the device split in your analytics. Divide the target sample by daily views on one device to get a rough number of collection days. If the answer is a quarter of a year, you need a different plan, not more patience.
Take a fictional carpentry workshop: its services page gets about 1,200 views a month, three quarters of them on phones. Roughly 900 mobile visits deliver 200 to 300 visits in just over a week. Roughly 300 desktop visits mean desktop needs about a month to reach the same threshold. The mobile map is ready much sooner, and its conclusion does not carry over to desktop.
Page: /services/. Device: phone. Period: from the last layout change. Target sample: 250 views. Question: do visitors reach the pricing before they see the form?
When you will never reach the threshold
Small sites often have pages that will not collect a few hundred visits on one device even in a year. That means using the map differently, not dropping it. Instead of waiting, widen the sample where that is fair and shift the burden of proof to individual visits.
- Extend the date range.Collect for longer, but only over a period without layout changes, big campaigns or seasonal exceptions. Treat each such change as the start of a new sample.
- Group pages that share a template.Product pages, posts or service pages with an identical layout can be analysed as a group when the question concerns a shared element, such as the contact button.
- Narrow the question to one element.Ask about a specific button or section rather than the whole page. With a small sample, comparing two areas is clearer than ranking twenty.
- Watch a series of recordings.Zoho PageSense suggests that 20 to 30 recordings with real interaction can reveal a consistent pattern. Keep watching until new visits stop adding new behaviour.
Grouping has a limit: do not combine the home page with the pricing page just to reach a number. If the layout, goal or traffic source differ, more visits produce a map you cannot tie to any decision. Then use the qualitative sources from the guide to A/B testing with low traffic.
When a pattern is stable enough
A simple check needs no statistics. Split the collection period into two halves and compare the maps for the same page and device. If the same elements lead the click ranking and the point where most visitors stop scrolling lands in a similar place, the pattern is stable enough to write down a hypothesis. If the order jumps around, you need more data or a narrower question.
Nielsen Norman Group recommends 39 participants when the goal is stable heatmaps from an eyetracking study, and 5 people for a qualitative usability test. These figures come from lab eye tracking, not website click maps, but they show that an aggregate view needs more material than finding problems does. NN/g: How Many Test Users.
- The map covers one page or a group of pages with an identical layout.
- Phone and desktop visits are analysed separately.
- Neither the page layout nor the main campaign changed during collection.
- Both halves of the period show a similar order of the most clicked elements.
- The same pattern appears in several recordings you have watched.
A stable pattern is still an observation, not proof of a cause. It tells you what repeats, not why. You can set the order of fixes by the strength of evidence, as described in the guide to prioritising UX fixes.
How to confirm a thin map with recordings
A recording does not replace a sample, but it shows whether a hot spot means what you suspect. Pick visits from the same page and device that include the clicked element, and note what happened around the click. If in several visits a click on an inactive heading is followed by more searching, you have a stronger signal than a patch of colour alone. The difference between an aggregate view and a single story is explained in session recordings or heatmaps.
In Meetlead Analytics you install the Heatmaps app for a selected website. The click map shows a ranking of elements, the click count and the number of retained recordings linked to each point in the selected range, and from there you can open a filtered list of those recordings. Whether recordings are saved, and for how long, depends on your plan. A method for watching them without picking at random is in how to analyse session recordings.
Questions about sample size
Do 200 visits make the result statistically significant? No. A heatmap gives you no significance level, and practitioner thresholds only indicate when a pattern stops being a random arrangement of a few clicks. Proving that a change raised conversions is a different question that needs a different method. The map helps you choose what to check.
Can I combine data from several months? Yes, as long as the page layout, the offer and the main traffic source stayed the same. Record the collection dates and the layout version with every map. If there was a rebuild in the middle of the period, treat the two parts as separate maps, even if each one is too small for firm conclusions.
