The basics
What is the Attention Map?
The Attention Map shows where visitors actually pay attention on a page — not just how far they scrolled or where they clicked. It measures, for every 5% band of the page and for the key elements on it, how long the content was genuinely in view, how long the mouse lingered there, how long people hesitated before clicking, and how much revenue came from the sessions that engaged.
How is this different from a scroll map or a click map?
A scroll map tells you how far down the page someone got. A click map tells you where they acted. Neither tells you what held them. Most of what happens on a page — reading, comparing, hesitating, losing interest — never becomes a click. The Attention Map measures that middle layer.
It also separates two things scroll depth blends together: how many people reached a section, and how many of them engaged with it. A footer only 4% of visitors reach can still be the most engaging block on the page for the people who get there.
What's the difference between the Measured and AI Prediction tabs?
Measured is real behaviour from your real visitors.
AI Prediction is a model's forecast of where attention will land, based on the visual design of the page alone. No traffic required.
Compare puts them side by side with a shared depth highlighter.
Measured tells you what happened. Prediction tells you what the design is inviting people to look at. The gap between them is usually the interesting part.
What do I use it for?
Three common jobs:
Finding content that earns attention but is buried too far down the page.
Finding elements that people look at for a long time before acting — usually a copy or clarity problem, not a design one.
Deciding what to test next, with evidence rather than opinion.
Do I need to install anything new?
No. The Attention Map runs on your existing Heatmap installation. There's no second snippet and no tagging work required to get started.
What it measures
What is a depth zone?
The page is divided into horizontal bands, 5% of the page height each — 20 bands from the top of the page to the bottom. Every metric in the depth view is reported per band. Wider bands (10%, 20%, 25%) are available for very long pages.
What do Reach % and Attention % mean?
Reach % — how many visitors got this far down the page, as a share of the most-reached zone on the page.
Attention % — of the visitors who reached that zone, the share who had it genuinely in view for at least 2 seconds.
Reach is about position. Attention is about whether the content did anything once it was seen. A zone with high reach and low attention is content people scroll straight past.
What is dwell?
Seconds that a zone or element was visible in the viewport, on a visible browser tab, with the visitor active. Time in a background tab doesn't count. Neither does time after the visitor goes idle — accrual pauses after 5 minutes of no scrolling, clicking, or typing, and resumes when they come back.
What is float?
Seconds the mouse cursor hovered over that zone or element. On desktop, cursor position is a reasonable proxy for reading position — people tend to park the mouse near what they're looking at.
What is hesitancy? What's a bad number?
Hesitancy is how long someone hovered on an element before clicking it, measured only among people who did click. It's reported as a median and a maximum.
There's no universal threshold, but the shape of the signal is consistent: low hesitancy means the element was obvious, high hesitancy means people had to think. High hesitancy on a CTA almost always points at the copy or the offer — the button isn't communicating clearly enough what happens next.
The companion metric is fast-click %: the share of clickers who clicked within 3 seconds. High fast-click plus low hesitancy is a healthy CTA.
Which elements get tracked? Can I choose them?
Up to 10 elements per page (12 maximum). They're detected automatically and classified as calls-to-action, media, or content blocks.
You can also mark specific elements yourself by adding a data-hm-watch attribute to them in your theme, with a name of your choosing. Tagged elements keep the name you give them, which makes them easier to find in reports.
How does an element count as "seen"?
An element has to be at least 40% visible before it starts accruing time. A zone has to overlap the viewport by at least 5%. This is deliberate — it stops content that merely clipped the edge of the screen from being counted as viewed.
What does the revenue column on an element mean?
Revenue from sessions that engaged with that element and purchased in the same session, plus purchase count and revenue per session. It's same-session attribution, so treat it as "elements that appear on the path to purchase" rather than proof of causation. A split test is still what proves causation.
What is active vs. open time?
Two numbers side by side: the median time the page was actually visible and being used, versus the median total time the tab was open. The gap is how much of your reported "time on page" is real. Pages where the two numbers diverge sharply usually have visitors parking the tab and coming back later — high time on page that isn't engagement.
Reading the data
What are insight cards?
Generated summaries of what stands out on the page. Six types:
Card | What it flags |
Hidden gem | A zone with high attention but low reach — good content, buried too deep |
Move-up candidate | Content that would likely earn more if it sat higher on the page |
Hesitancy hotspot | An element people hover a long time before clicking |
Active vs. open time | A large gap between real attention and reported time on page |
Money zone | The depth band most associated with purchasing sessions |
Revenue driver | The element most associated with purchasing sessions |
They're a starting point for a hypothesis, not a verdict. Confirm anything surprising against the underlying numbers before you act on it.
Why doesn't an element I care about appear in the report?
Most likely one of three reasons:
Not enough viewers. Elements need at least 5 distinct viewers before they're reported. Below that the numbers are noise.
Not detected. It wasn't picked up as a call-to-action, media, or content block. Add a
data-hm-watchattribute to track it explicitly.Element cap. Only the top elements on the page are tracked, up to 12.
Can I filter the data?
Yes — date range, device, new vs. returning visitors, buyers vs. non-buyers, traffic source, and country.
Can I see attention for buyers only?
Yes, and it's one of the most useful things in the report. Buyer reach and buyer attention are reported per zone, so you can see which sections of the page the people who actually purchased spent time on. Comparing buyers against everyone tells you which content is doing persuasive work and which is just occupying space.
AI Prediction
How does the prediction work?
We capture a full-page screenshot of the page and run it through a predictive attention model. The model returns a predicted attention intensity for every point on the page — displayed as an Attention score from 0 to 100 — plus predicted attention by depth, and heatmap, focus, and contrast views of the page. [[VENDOR-ATTRIBUTION]]
Is this eye-tracking?
No. It's a model trained to predict where human attention goes based on the visual properties of a design — contrast, size, position, faces, text density. No visitors are involved. That's the point: you can run it on a page that has no traffic yet, including a page that isn't live.
When measured and predicted disagree, which one is right?
Both, about different things. The prediction describes what the design pulls the eye toward. The measured data describes what your actual visitors, with their actual intent, did.
The two useful patterns:
Predicted high, measured low — the design is shouting about something visitors don't care about. Usually a relevance or messaging problem.
Predicted low, measured high — visitors are hunting for something the design is hiding. Usually the fastest win on the page: make the thing they're already looking for easier to find.
Why does the prediction look different from my live page?
The screenshot is taken with cookie and consent banners suppressed and at a single viewport, so the captured page may differ from what you'd see yourself — particularly on pages with personalization, geo-specific content, or an active A/B test.
Can I run the prediction on a page that isn't live yet?
Not in this release — prediction runs against a URL we can reach. Pre-launch design review is on the roadmap.
Data accuracy and privacy
How soon does data appear, and how often does it update?
Attention data begins appearing within a couple of hours of a site being enabled, and updates hourly after that. It isn't a live view.
Why don't the depth percentages match my scroll depth report?
Because the Attention Map only aggregates visitors who saw a comparable version of the page. Pages vary in height between visitors — A/B test variants, geo-specific blocks, lazy-loaded sections — and averaging a visitor who saw a 4,000px page with one who saw a 12,000px page makes the depth percentages meaningless. So the depth view keeps pageviews whose page height falls within ±25% of the median for that page and excludes the rest.
The result is a depth map you can trust for positional decisions, at the cost of not matching a raw scroll number exactly. Both are correct; they're answering different questions.
Why are hover and hesitancy near zero on mobile?
Float and hesitancy are cursor metrics, and touch devices have no cursor. On mobile, use dwell, reach, attention %, and clicks. Read hover-based metrics as desktop signals only.
What happens on pages that change height as they load?
When a page grows substantially past its initial height — lazy-loaded sections, infinite scroll — the depth grid is locked to a stable reference height so that zone boundaries mean the same thing for every visitor. Without this, the same pixel would land in different zones for different people and the depth map would be unreadable.
How are rage clicks and bots handled?
Pageviews with an abnormal burst of clicks in a short window are flagged so they can be excluded from analysis rather than distorting engagement numbers.
What data does this collect? Is it privacy-safe?
The Attention Map records timing and interaction data — how long regions and elements were in view, cursor position over the page, and click timing. It doesn't capture form contents, keystrokes, or personal information. Everything is reported in aggregate.
Visitors who should not be tracked at all can be excluded using the standard Heatmap do-not-track flag, which suppresses attention collection entirely for that visitor.
Does this slow my site down?
No measurable impact. The tracker samples once a second and sends compact batched updates rather than a stream of events. Tracker performance overhead is tested against live production sites as part of every release.
Does this depend on session recording?
No. Attention collection runs on every pageview independently of whether that visit is being recorded, so your attention data is complete even when session recording samples a visit out.
Access and troubleshooting
How do I get access?
The Attention Map is rolling out progressively across accounts. [[ACCESS-PATH]]
I've been enabled but I see no data.
Check in this order:
Has it been at least two hours since enablement?
Is the page you're looking at one that actually gets traffic in the selected date range?
Is the do-not-track flag set on your site?
Are your filters excluding everything — a narrow date range plus a device plus a source will empty most reports.
If all four are clear, contact Support@heatmap.com with the site and page URL.
A number looks impossibly high or doubled.
Report it to Support@heatmap.com with the page URL, date range, and filters applied. Include a screenshot if you can — it makes diagnosis much faster.
How far back does the data go?
Attention data starts accruing when your site is enabled. It isn't backfilled from before that point.
Need help?
If you have questions or run into any issues while using Heatmap, our support team is here to help. Contact us anytime at Support@heatmap.com.