How to

How to retrain a feed that keeps showing you things you do not want

You did not search for it, you did not follow it, and it keeps arriving anyway. Feeds learn from behaviour, and the behaviour they learn from is not the one you would choose.

Short answer

Feeds weight watch time and lingering far more heavily than the buttons you press. Retraining one means changing what you do, not just what you tell it: scroll past instantly, never open the comments, use the explicit "not interested" controls every single time, and stop giving it anything to work with.

Step by step

How to do it

01Step

Stop giving it watch time, even hostile watch time

Watching something to be annoyed by it is still watching it. Recommendation systems mostly cannot tell the difference between fascination and disgust; they see dwell time. Scroll past within a second, every time.

02Step

Never open the comments

Opening comments is one of the strongest engagement signals you can send. It reliably teaches the feed that this is exactly the sort of thing you stop for.

03Step

Use the explicit controls every time, not once

"Not interested", "hide", and reporting are weak signals individually and meaningful in aggregate. Once will do nothing. Fifty times over a fortnight does something.

04Step

Prune what you follow and what you have liked

Your follows and your like history are durable inputs that keep reasserting themselves. Unfollow the adjacent accounts, not just the obvious ones, and clear old likes where the platform allows it.

05Step

Feed it what you do want, deliberately

Search for and watch the things you would rather see, all the way through. An empty preference gets filled by whatever is generally engaging, which is rarely what you want.

06Step

Give it two weeks

None of this is instant. Most people see a noticeable shift within one to two weeks of consistent behaviour, and it decays back if you drift.

Why the buttons feel useless

Because they are weak inputs competing with strong ones. What you actually watched, how long you paused, what you replayed and what you opened are the signals with weight behind them. A tap on "not interested" is a suggestion.

This is why people conclude the controls are broken. They are not broken, they are outvoted, usually by the user themselves.

The hostile-watching trap

The single most common way people train a feed towards content they hate is by watching it to be annoyed. Outrage holds attention as effectively as interest, and the system is optimising for attention.

If something makes you want to stay and argue, that is precisely the moment to leave. Your feed is watching what you do, not what you feel about it.

If the feed will not shift, or if the content is something you have decided not to see at all, a block ends the argument. Kellet can block the app or the site, on a schedule, with a delay on turning it back on.

Where retraining stops working

Retraining is the right tool when you want to keep using a platform and change what it shows you. It is the wrong tool for content you have decided you do not want to see under any circumstances, because a recommendation system will always eventually test you again.

It also cannot help with the thing you go looking for yourself. That is a different problem with a different answer.

The nuclear options

  • A fresh account.Starts the model from nothing. Effective, and you lose your follows and history.
  • Turn off personalisation.Where offered, this replaces a tuned feed with a generic one. Often duller, which is frequently the point.
  • Block the feed, keep the app.Some platforms separate the algorithmic feed from messaging, so you can cut one and keep the other.
  • Block the app during your worst hours.If the feed only hurts you late at night, that is a scheduling problem rather than an algorithm problem.
Straight answers

Common questions

How long does it take to retrain a feed?
Most people report a noticeable difference within one to two weeks of consistent behaviour. Platforms do not publish how quickly their models adapt, so treat any precise figure with suspicion, including that one.
Does clearing my watch history reset it?
It helps on platforms that let you do it, because history is a direct input. It does not clear everything the model has learned, and follows and likes will keep pulling it back.
Why does it come back after a while?
Recommendation systems keep exploring. They will periodically re-test things you rejected to see whether you have changed. This is deliberate and it is why retraining is maintenance rather than a one-off fix.
One decision left

Calmer water is one decision away.

Kellet blocks the distractions you choose and lets you stay accountable to yourself, or to someone you trust.