Scroll depth is not attention — what it really measures
It is the most quoted and least understood number in a small site's analytics. Here is what scroll depth can prove, the four claims it cannot support, and how to read it next to the signals that explain it.
Short answer: scroll depth measures how far down a page a visit reached — the deepest point the viewport touched. It tells you how much of your content was available to be seen, not whether anyone read it, understood it or cared. Treat it as a placement metric and read it next to clicks, goals and the sessions behind it.
What scroll depth measures (and how your tool defines it)
Scroll depth is a movement metric. It records how far the page travelled in the browser window during a visit. Nothing more: it does not know whether the visitor looked at what passed, whether they were reading, or whether they left because of what they found.
That would be harmless if everyone measured it the same way. They don't. Three families of definitions are in use, and they produce three different numbers for the same page, the same visitors and the same week.
| What the report shows | What it really counts | Where it misleads you |
|---|---|---|
| Average scroll depth | The mean of each visit's deepest point, often as a share of page height | A few long scrolls lift the average while most visits stopped early — the number describes neither group |
| Threshold shares | The share of visits that crossed a line: 25%, 50%, 75%, 100% | Crossing a line is not seeing everything above it; a fast scroll past a block still counts |
| Deepest point of a session | The furthest a visitor scrolled anywhere in the visit, all pages included | A second page visited later can be credited with the scroll of a page read first |
Two tools can therefore disagree about the same page without either being wrong. If you migrate your analytics, expect the scroll numbers to move, and do not read that movement as a change in visitor behaviour.
Four things scroll depth does not tell you
These are the four readings I see most often in small-site reports. Each one asks the number to prove something it was never built to prove.
1. It does not tell you that anyone read the page
A visitor looking for a delivery time, a price or a phone number will scroll fast and far. Deep scroll is a common signature of searching, not of reading: reaching the bottom in four seconds is easy. If your question is "did this content land", scroll depth cannot answer it. The question needs a different instrument — time in one section, a click on a link inside the text, or a replay where you can see the pause.
2. It does not tell you the page is engaging
Make a page twice as long and the share of visits reaching the bottom falls, even if not a single visitor changed behaviour. Shorten it and the same number rises. Scroll depth is partly a description of your page length, which is why it can never be compared across page types: a pricing page, a guide and a contact page will produce three different curves by construction, not by merit. Compare a page with itself, over time.
3. It does not tell you where the fold is
The fold is not one place. On a phone it moves with the browser chrome and the font size the visitor chose; on a desktop it depends on the window height. A reassuring average can therefore hide a mobile reality where your call to action sits three swipes down. Split the metric by device before you conclude that everyone saw it.
4. It does not tell you why visitors stopped where they stopped
This is the limit worth remembering. A cliff at 60% tells you where attention ended; it says nothing about why — a price, a technical error, a form asking for a phone number, a paragraph that reads like a contract. Localising the problem is genuinely useful, and it is also only half the job. The other half is watching what happened in the visits that stopped: this is where finding out why visitors leave stops being a guessing game.
Where scroll depth earns its place
Used honestly, scroll depth answers one family of questions very well: is the content that has to be seen actually within reach? Three jobs it does better than any other metric.
- A placement check. Mark where your price, your call to action and your proof sit in the page, then compare that with the share of visits reaching that point. If the offer lives below the zone most visits ever reach, you have found a layout problem without touching the design.
- A regression check. Same page, same traffic mix, before and after a change: a curve that drops is a strong signal that something moved out of the way or that the page grew. It is one of the few ways a low-traffic site can judge a layout change without a proper test.
- A length check. A long page where almost nobody passes halfway is usually a structural verdict: the useful part is buried behind paragraphs nobody asked for. The fix is moving the answer up, not writing less.
In my experience the mistake is not reading these numbers, it is reading them as a score. A page is not better because more people scrolled it. It is better when the visitors who scrolled it did the thing you wanted — and that verdict comes from goals, not from the curve.
How to read it without fooling yourself
Five habits separate a scroll report you can act on from one that just looks precise.
- Read the curve, not the average. The share of visits reaching 25, 50, 75 and 100 percent tells a shape. One mean collapses it, and two very different pages can share the same mean.
- Split by device first. Mobile and desktop are two different layouts on two different screens. An average over both describes neither.
- Compare like with like. Same template, same page type, same traffic mix, same period length. Scroll depth is highly sensitive to page length and to the source that sent the visit.
- Anchor the thresholds to your content. Add your own markers: the position of the call to action, the end of the first argument, the start of the form. Thresholds of 50% on an article and 50% on a checkout page do not mean the same thing.
- Cross-check before you change anything. A number that cannot be confirmed by a second signal (a click, a goal, a replay) is a hypothesis, not a diagnosis. Set a goal on the action you actually care about first — goals that answer a real question rather than tally clicks — so every later reading has something to be judged against.
| What you observe | The likeliest reading | What to check next |
|---|---|---|
| Deep scroll, few clicks on the call to action | People scanned the page for something you did not put where they looked | Placement of the offer, wording of the button, the click map of the same page |
| Shallow scroll on mobile, deep on desktop | Something at the top of the mobile page absorbs the first screen | The mobile layout in a replay, and whether the first screen answers the query |
| A good curve, no conversions | Scrolling is not your leak; the offer or the next step is | The step after the page, and where visitors abandon there |
| A cliff at the same point on several pages | A shared element — banner, block, form — interrupts the flow | The component common to those templates, on both devices |
None of these readings replaces a second number: a high exit rate on a page that should lead somewhere is a different and often more urgent signal than a shallow scroll, and it answers a different question about the same departure.
What to measure alongside it
Scroll depth becomes useful the moment it stops travelling alone. Three companions cover what it cannot.
- A goal. The outcome you wanted from this page. Without it, no scroll number can be called good or bad.
- A click map. Where attention went, which is not where the scroll went. Click and scroll maps are read together — how to read a heatmap covers the five traps that produce wrong conclusions from either.
- Session replays. The only instrument that shows the pause, the repeated tap, the half-filled form and the return to the same sentence. Watch a dozen visits that stopped at your cliff, filtered by device, and the reason usually appears quickly.
That combination is exactly what SessionInsight keeps in one place: analytics, scroll and click heatmaps, session replays and goals on the same screen, so the curve and the visits behind it are never two separate investigations. You can see how it works or start on the free plan — one website, 50,000 pageviews a month, 30 days of history.
FAQ
What is scroll depth?
Scroll depth is how far down a page a visit reached, expressed either as a percentage of the page height or as the share of visits that crossed fixed thresholds such as 25%, 50%, 75% and 100%. It is a reach metric: it describes how much of the page was available to be seen, not what the visitor did with it.
Is high scroll depth good?
Not by itself. A high number can mean the page holds attention, or that the page is long and people are scrolling to find something. I treat it as good only when the sessions behind it end in the action I wanted, or when it shows me where to move the answer up.
Does scroll depth measure reading?
No. Scrolling is not reading: a visitor can reach the bottom of a page in seconds while looking for a price, a phone number or a delivery time. Reading leaves other traces, such as time spent in one section, a click, or a return visit, which is why scroll depth is read next to clicks and session replays.
What is a good average scroll depth?
There is no universal figure, and a number quoted for the whole web tells you nothing about your site: it depends on page length, template and device. Compare a page with its own history, and split mobile from desktop before you draw a conclusion.
How is scroll depth measured?
Most tools record either the share of visits that cross fixed thresholds, or the deepest point each visit reached. The two definitions produce different numbers for the same page, so compare figures from one tool and never across two.
Should scroll depth be a conversion goal?
No. Scroll depth is a diagnostic, not an outcome. Keep goals on things a visitor deliberately did, such as an order, a form or an enquiry, and use scroll depth to decide where to look inside the page.
