SuckButtBecause some things just suck

The Internet Sucks

Why Your Feed Shows Things You Never Followed

Accounts you chose produce a finite amount of material, and recommendation removes that ceiling, which matters because advertising inventory grows with the pool it sits inside.

A neat workspace featuring a laptop displaying Google search, a smartphone, and a notebook on a wooden desk.
Photograph by Caio via Pexels
Editorial note. Independent reporting and analysis. Nothing here is sponsored or paid for. How we work.

You followed a few dozen accounts and the feed is now mostly strangers. The change was not a drift in taste. It was a change in where the supply comes from.

Following produces a finite supply

A feed built strictly from accounts you chose can only be as large as those accounts post. For most people that runs out within a few minutes.

Once it runs out, the session ends. There is no more material, and a platform whose revenue depends on time in the application has just hit a hard ceiling.

Following also decays. People post less, accounts go quiet, and the pool shrinks over time without the user doing anything at all.

Recommendation removes the ceiling

Drawing candidates from everything posted anywhere, and ranking them by predicted interest, turns a small supply into an effectively unlimited one.

The model does not need to know you well. It needs to be better than random across a very large pool, which is a much easier problem.

Sessions stop ending naturally, because there is always another item, and the feed no longer has a bottom to reach.

Advertising needs a large pool to sit in

Adverts are inserted between organic items at a set ratio, so total advertising capacity is a direct function of how many items a user scrolls past.

Expanding the organic supply therefore expands the advertising supply proportionally, without changing the ratio that determines how cluttered the feed feels.

This is the cleanest explanation for why every platform made the same change at roughly the same time, regardless of what each one was originally for.

Engagement per item beats loyalty

Ranking systems are trained on whether an individual item gets a reaction, which favours whatever performs well right now over whatever you chose to follow last year.

A stranger's popular video reliably beats a friend's ordinary update on that measure, so the friend gradually disappears without ever being unfollowed.

Users then follow more accounts to compensate, which does not help, because the ranking applies to the enlarged list exactly as it did to the small one.

The follow graph became a hint

What used to be an instruction is now one input among many, weighted alongside watch time, dwell, shares and behaviour of people the system considers similar to you.

That makes the feed harder to steer deliberately, since the visible controls affect only the input that carries the least weight.

Chronological options survive on several platforms, generally one layer down in a menu and rarely as the default, which is a fair indication of how much they are used.

Questions readers ask

How do I tell if a page is filler?

Check whether any sentence makes a claim that could be wrong. Restated questions, unsourced numbers and answers buried at the end are the common markers.

Are search operators still worth learning?

Yes. Site restriction, date filtering and exact-phrase matching all still work and bypass a great deal of the ranked page.

The Internet Suckssearchcontentpublishingquality
More in The Internet Sucks
Jhilik Mahapatra
Contributing writer, SuckButt

Jhilik writes about everyday irritations and believes packaging is a conspiracy.

Also by Jhilik Mahapatra