A streaming home screen looks like a shelf of categories, but almost nothing about its arrangement is fixed. Both the rows and their contents are assembled per household, per session.
Two rankings run at once
The first ranking chooses which rows appear and in what order. Each candidate row is scored on how likely it is to produce a play from this particular profile right now.
The second ranking orders titles inside each row. A row that survives to the top of the page is then filled with the individual titles judged most likely to be selected from it.
Because both layers are personalised, two households can see rows with identical names holding almost entirely different titles, which makes the shelf metaphor misleading.
Position carries most of the weight
Attention on a home screen falls away sharply from the top left. The first row and the first few tiles within it collect a disproportionate share of every play the page generates.
That concentration means placement is a stronger lever than the catalogue itself. A title buried four rows down on a television interface is, in practical terms, absent from the service.
It also explains the small number of visible rows. Adding more shelves does not spread attention evenly; it pushes titles into positions where they are effectively never seen.
Why the artwork changes too
The tile image is not a fixed poster. Services generate several images for the same title, each emphasising a different character, tone or moment, and select among them per viewer.
The selection follows the same logic as the ranking, drawing on which imagery has previously preceded a play for similar viewing histories. A comedy can be presented as a romance or as an ensemble piece.
This is why a title can feel unfamiliar when someone else describes it. The two of you may have been shown genuinely different faces of the same programme.
Browsing and searching ask different questions
Search answers a question the viewer already has, so it is judged on precision. The ranking has one clear target and either finds it or fails visibly.
A home screen answers a question nobody has asked. It is judged on whether a session that began without intent ends in a play, which rewards breadth and recognisable hooks over accuracy.
The two systems therefore optimise against each other. A row built to start an undecided session will look like a poor answer to anyone who arrived knowing exactly what they wanted.
The catalogue behind the page
Rows are also constrained by rights. A title available in one country and not another cannot appear, so the same personalisation logic produces structurally different pages by territory.
Recency plays a part as well. Newly added titles are surfaced aggressively for a short window, which is why a service can feel busy on Friday and static a fortnight later.