How to retrain the YouTube algorithm

Why YouTube is the easy case

Recommendations here are unusually legible: they are largely a function of what you watched, and that record is editable. Very few platforms let you reach into the input like that.

It also means the fix is durable. Remove the cause from history and the effect genuinely fades, rather than being outvoted the next time you linger.

Shorts is a different product

Everything above applies to the main recommendation system. Shorts runs on the short-form logic instead, where watch time dominates and history matters less.

If Shorts is what keeps serving you things you do not want, retraining it means the same discipline as TikTok: leave within a second, every time.

Remove a video from history and it stops counting

This is the control that makes YouTube genuinely easier to retrain than any other platform, and it is the one people rarely use, because it is buried and looks like housekeeping rather than tuning.

Individual entries can be deleted from watch history, and removing one withdraws the signal it was providing. Nothing else on this site offers that. Elsewhere you can clear everything or nothing; here you can remove the six videos from one bad evening and keep a decade of accurate signal about what you actually want.

The practical routine is to delete the entries you regret rather than clearing the lot. A full wipe gets you a generic feed with nothing useful in it, and most people find that worse and turn history back on within a week, which restores the old recommendations along with it.

The channels doing the damage are usually ones you subscribed to

Retraining advice focuses on recommendations, and on this platform a large share of the problem arrives from channels you deliberately chose, sometimes years ago and often for a reason that no longer applies.

Subscriptions are sticky in a way follows elsewhere are not, partly because unsubscribing feels like a judgement on someone whose work you once liked. The result is a home page shaped by an older version of your interests, topped up by a recommendation system doing its honest best with that input.

An unsubscribe pass is usually more effective than any amount of algorithm tuning. Go through the list and remove anything you would not deliberately search for today. The feed changes noticeably within days, and unlike retraining it does not require you to behave perfectly for a fortnight.