Scroll depth is the most quoted number on a heatmap and the most often misread. It is worth knowing exactly what it counts, because two of the three obvious readings of it are wrong, and both of them flatter the page.
It measures the bottom of the window, not the top
The measurement is the position of the bottom edge of the visitor's window against the full height of the document. Open a page and do nothing at all, and you have already recorded a depth greater than zero — because the first screenful is content you were shown.
This is the right definition and it surprises people. The alternative, measuring the top of the window, says a visitor who read the first screen and left saw none of the page, which is plainly untrue. But it does mean the number is not "how far down they went"; it is "how much of the page was put in front of them".
It is ten buckets, not a percentage
What is stored is a whole number from 0 to 10 — which tenth of the page the window's bottom edge reached. Not 63%, not 62.8%. There are three reasons, and only the first is about privacy.
- A precise percentage is close to an identifier. An exact scroll position, repeated across a few pages, is a surprisingly distinctive trail. A decile is not.
- The precision is not real. Document height changes as images load, fonts swap and embedded content settles. A number quoted to a decimal place implies a stability that web pages do not have.
- Nobody acts on the difference. No decision changes between 62% and 63%. The decisions change at "most people never see the second half", which a tenth resolves perfectly well.
The two readings worth acting on
A cliff. A big drop between two adjacent buckets is the useful finding — something at that point in the page stops people. Usually it is a false ending: a horizontal rule, a large block of white space, or a section that looks like a footer. Visitors are not bored, they think the page has finished.
A call to action below the cliff. Cross-reference the depth where you lose half the audience against the position of the thing you want them to click. If the button sits in the eighth tenth and half the readers stop at the fourth, the button's problem is not its wording.
What is not worth acting on is a gentle taper. Attention declining smoothly down a long page is the normal shape of reading, not a fault, and flattening it is not a goal.
What it cannot tell you
It cannot tell you that anything was read, as above. It cannot tell you the order things were seen in, because the stored value is the deepest point reached rather than a path. And it cannot tell you about one person — there is no identifier on it, so a scroll map is only ever a shape made of many visits.
That last limitation is deliberate and it is the trade this product makes everywhere: the aggregate shape is the thing worth knowing, and the per-person trail is the thing that needs a consent banner. Scroll depth is a good example of a measurement that gives up almost nothing useful by being anonymous.
Scroll maps arrive with click and attention maps from the script already on your site, and if you want the rest of the picture first, how to read a heatmap covers what each of the three maps is good at.