How to use Attention Analysis
Attention analysis infers where visitors focus from mouse movement, dwell, and scroll behavior — the closest practical proxy for eye-tracking without specialised hardware.
Attention Analysis
Inferring what users focus on from mouse movement, dwell time, and scroll behavior — eye-tracking's practical proxy.
Attention analysis is the practice of reconstructing where a visitor's attention falls on a page using passive behavioral signals — cursor position and motion, time spent in a region, and scroll depth and velocity. It sits inside the broader discipline of behavioral analytics and answers a question raw click data can't: what did the visitor actually consider before they clicked, scrolled past, or left?
Unlike webcam-based eye-tracking, attention analysis requires no hardware, no opt-in panel, and no lab. It runs on every session at scale, which makes it the default attention layer for e-commerce CRO. The trade-off: signals are inferred, not measured, so reading them well matters as much as collecting them.
On a typical Shopify product page, fewer than one in three visitors clicks anything before they bounce. Click data tells you what happened; attention analysis tells you what almost happened — the hero image they paused on, the size selector they hovered but didn't pick, the review section they scrolled past in under a second.
That gap between intent and action is where most CRO wins live. If you only test the elements people clicked, you optimise the funnel that's already working. Attention analysis surfaces the elements that competed for the click and lost.
The three signals attention analysis runs on
Mouse movement is the loudest signal. Roughly 70% of desktop users move their cursor along the line they're reading or toward the element they're considering, so cursor heatmaps approximate gaze on text-heavy and form-heavy pages. The correlation drops on image-led pages where visitors read with their eyes and leave the cursor parked.
Dwell time on a region — how long a visible element stays in the viewport before the user scrolls or interacts — is the second signal. Long dwell on a price block or a returns policy almost always indicates friction. Long dwell on a review carousel usually indicates engagement. The label depends on what comes next.
Scroll behavior is the third. Slow, steady scrolling means the page is being read. Fast scrolling with brief stops means the visitor is scanning for one specific thing. A sudden scroll-back is a tell that something earlier in the page caught attention that the visitor wants to re-examine — usually price, sizing, or a guarantee.
Mobile changes the rules
On mobile, there's no cursor — attention has to be inferred from tap targets, scroll velocity, and pause-on-element. Treat mobile and desktop attention models as separate analyses; collapsing them hides device-specific friction (especially in checkout).
Reading the signals correctly
Not every signal carries equal weight. Cursor-to-gaze correlation is strongest on forms (checkout, account signup) and weakest on visually-driven landing pages. Dwell time is most reliable when the element is above the fold and fully visible — partially-visible elements skew the data.
The chart below shows the rough reliability of each signal across page types. Use it to decide which signal to trust when they disagree — and they will disagree, frequently.
Signal reliability by page type
Cursor tracking
Dwell time
Scroll behavior
The pattern: on a checkout form, trust the cursor. On a category grid, trust scroll behavior. On a product page, dwell is your strongest read. When two signals agree, you have something close to a confirmed observation. When they disagree, the conservative move is to design a test rather than draw a conclusion.
What attention looks like on real pages
The table below shows ballpark attention shares for a typical apparel product page — the proportion of a visitor's on-page time spent dwelling on or hovering near each element. Use it as a baseline; your store will vary by category, price point, and audience.
Notice how much attention the hero image and price-area receive relative to the actual buy button. The buy button is the destination — it doesn't need attention, it needs clarity. Trust signals, sizing, and reviews are where the visitor decides whether to use it.
Attention share by element — apparel product page (desktop)
| Page element | Share of attention time | Average dwell per visit | Bounce-correlated? |
|---|---|---|---|
| Hero / product image gallery | 28-34% | 11-14s | Yes — short dwell predicts bounce |
| Price + discount block | 14-18% | 5-7s | Yes — long dwell predicts bounce |
| Size / variant selector | 10-14% | 4-6s | Yes — hover-without-select predicts exit |
| Reviews / social proof | 12-16% | 8-12s | No — long dwell is engagement |
| Shipping & returns info | 6-9% | 3-5s | Yes — long dwell predicts bounce |
| Description / product details | 8-12% | 6-10s | No |
| Add-to-cart button | 3-5% | 1-2s | Inverse — short dwell is healthy |
The actionable read: any element where long dwell correlates with bounce is a friction candidate. Price block dwell over seven seconds usually means the visitor is doing mental math the page should be doing for them — financing options, bundled discount, or a clearer per-unit price often resolves it.
Turning attention data into tests
Attention analysis is diagnostic, not prescriptive. It tells you where to look; it doesn't tell you what to change. The reliable workflow is to use attention data to generate hypotheses, then validate them with an A/B test. Skipping the test is how teams ship attention-driven changes that quietly hurt conversion.
A useful template: identify the element with anomalous attention (much higher or lower than the benchmark), state what you think is happening in one sentence, name the metric the change should move, then design the smallest variant that tests it. If you can't write that sentence, you're not ready to test.
Don't redesign on heatmaps alone
Cursor heatmaps are a generation tool, not a verdict. Two visitors with identical heatmaps can have opposite intents — one engaged, one confused. Always pair attention analysis with session replay for context and an A/B test for proof.
Attention analysis FAQ
On form-heavy and text-heavy pages, cursor position correlates with gaze around 65-75% of the time. On image-led pages it drops below 50%. Cursor data is good enough to find friction and generate hypotheses, but it's not a substitute for eye-tracking when you need pixel-level gaze data.
Partially. You lose the cursor signal entirely, so mobile attention analysis relies on scroll velocity, dwell time per viewport-visible element, and tap-target proximity. It's still useful — especially for spotting checkout friction — but treat it as a separate model from desktop.
Behavioral analytics is the parent discipline, covering every kind of user-action signal: clicks, events, funnels, retention. Attention analysis is the sub-discipline focused specifically on what visitors consider before they act — the pre-click layer that explains why your funnel converts the way it does.
For desktop, around 1,000 sessions per page gives stable cursor and dwell patterns. For mobile, 2,000-3,000 sessions because the signals are noisier. Below that, you'll see patterns that disappear when more data arrives — common cause of bad redesigns.
Lightweight modern implementations add 5-15ms to page load and sample cursor data at 10-20Hz, which is imperceptible. Older heatmap tools that record at 60Hz and ship full DOM snapshots can add 100ms+ and hurt Core Web Vitals. Audit the actual performance impact before deploying.
Cursor position and dwell time are generally not personal data on their own, but combined with session replay they often are. If you record actual sessions you need consent and a clear privacy notice. Aggregated heatmaps without session replay typically fall under legitimate interest, but check with your DPO.
Before, to generate hypotheses, and after, to understand why a test won or lost. A test result tells you which variant performed better; attention data on each variant tells you why — which is what you need to apply the lesson to the next test.
Click heatmaps show where visitors clicked — a record of action. Attention heatmaps show where the cursor lingered or dwelt — a record of consideration. An element with high attention and low clicks is usually a near-miss: it caught interest but didn't earn the action.
No. Attention analysis tells you what happened, not why. A visitor who dwells nine seconds on shipping info might be skeptical, confused, or just a careful reader. Interviews and surveys give you the why; attention data tells you where to point them.
Pick the single element with the most anomalous attention signal — usually a long-dwell field in checkout or an ignored CTA on a landing page. Write a one-sentence hypothesis, ship a two-variant A/B test, and measure against a primary funnel metric. One focused test per week beats a quarterly redesign.
Get an AI expert review of your site
Paste your URL — Metricuno's AI runs the same heuristic checks a senior CRO consultant would, scoring your page and prioritising the fixes that'll move conversion fastest.