The YouTube algorithm stopped recommending and started herding

I opened YouTube this morning to check if a channel I follow had posted, and the homepage gave me: a Shorts shelf I didn't scroll to, two "reaction" channels clipping a podcast I've never watched, and a video I already watched three days ago, recommended to me again like we'd never met. I had to tap into Subscriptions manually to find out the channel I actually came for had, in fact, posted twelve hours ago. It just hadn't bothered to tell me.

This isn't a new complaint. People have been saying some version of this for a few years now, that YouTube's recommendations used to feel like they came from someone who'd actually watched what you watch, and now they feel like a slot machine pulling whichever lever pays the platform best. I've mostly agreed and moved on with my day. But I've hit the point where I want to actually work out why it feels this way, instead of just grumbling at my phone.

There used to be a version of this that worked

Go back far enough and YouTube's recommendations were driven pretty crudely by clicks and view counts, which produced its own disaster: a race to the most clickable thumbnail regardless of whether the video delivered anything. The fix, a shift toward optimizing for watch time instead of clicks, actually made things better for a while. It's the era people are nostalgic for, the one where the sidebar would hand you something you didn't know you wanted, a niche essay, an old talk, a video from a channel with four hundred subscribers that happened to be exactly what you needed that day. It felt less like being sold to and more like being understood.

I don't think that era was some golden age of pure benevolent design. It optimized for watch time because watch time meant more ads shown, and more ads shown meant more revenue. The interests of "keep this person engaged" and "show this person something good" happened to overlap a lot more than they do now. That's the part that changed, not the intent, the overlap.

What broke the overlap

A few things stacked on top of each other, and I think it's fair to say no single one of them is the villain on its own.

Short-form exploded, TikTok proved you could hold attention in fifteen seconds instead of fifteen minutes, and every platform with a recommendation feed responded by building its own version and then pushing it into every surface it could reach. Shorts didn't stay in their own tab. They colonized the homepage shelf, the search results, even the subscriptions feed on some days, whether or not you've ever tapped on one. I haven't watched a Short in months and mine still shows up first.

Watch-time optimization, once it matured, stopped rewarding "genuinely interesting" and started rewarding "shaped like something the retention curve likes." Which is a real, measurable thing, and also a thing you can game without making a better video. Ten-minute videos ballooned into eighteen minutes because ad breaks slot in better and the algorithm doesn't penalize padding if the retention graph still looks acceptable. Thumbnails converged on the same three or four visual tricks, the red circle, the shocked face, the arrow pointing at something, because YouTube's own thumbnail-testing tools reward whatever gets the click, and click-through is still a huge input even in a "watch time" world. Creators I used to think of as restrained started doing it too, and I don't blame them, the ones who didn't got buried.

And recommendations narrowed instead of widening. The thing I miss most isn't any specific feature, it's the discovery, the feeling that the algorithm might surface something outside my existing loop. Now mine is a closed circuit: five or six channels, endless permutations of the same three topics, recommended back to me in slightly different packaging every day. I don't know if that's the model getting more confident or more conservative, but the net effect is the same: it stopped introducing me to anything.

The word people reach for

The term that gets thrown around a lot for this pattern, not just on YouTube but across most ad-funded platforms as they age, is "enshittification," coined by Cory Doctorow to describe a fairly consistent lifecycle: a platform is good to users to build a userbase, good to the businesses and creators built on top of it once the userbase is locked in, and then extracts value from everyone once neither group can easily leave. I don't think YouTube has hit the ugliest stage of that curve. Creators still make real careers there and a lot of genuinely great work still gets made and found. But the recommendation quality is exactly the kind of thing that erodes first and quietly, because a slightly worse homepage doesn't make you delete the app, it just makes you a little more annoyed every day, which is apparently a cost the platform can absorb indefinitely.

What I actually want

Not much, honestly. A Subscriptions tab that reliably shows me videos from channels I subscribed to, in something resembling the order they were posted, without the algorithm deciding some of them aren't worth surfacing. A homepage that's allowed to not have a Shorts shelf if I've never once engaged with one. And the willingness to occasionally hand me something a little outside my lane again, even at the cost of a slightly lower retention number on that one recommendation slot.

I don't expect any of that. The incentives that produced the current homepage aren't going anywhere, and I'm one person complaining into a blog post, not a shareholder. But I did want to at least write down why it feels different, instead of just feeling it and closing the tab.