How to retrain the Instagram algorithm

Explore and Reels are trained separately

People often fix one and are confused that the other is unchanged. They draw on overlapping but distinct signals, so both need the same corrective treatment.

Reels is the harder of the two, because vertical video rewards watch time more sharply than a grid of images does.

Sensitive content control is weaker than it sounds

Instagram offers a sensitive content setting with a "Less" option, and it reduces some suggestions rather than removing a category. It is a dial, not a filter.

It also only affects recommended content, so anything from accounts you follow is unaffected.

The signals that count, in order

Retraining works when you understand which behaviours the system weighs heavily and which it barely notices. Most people spend their effort on the second group.

• How long you stayed.Dwell time on a single post is the loudest signal you emit, and you emit it without doing anything. Stopping to look at something you disapprove of teaches exactly the same lesson as stopping to look at something you like.
• Whether you watched it twice.A loop counts. On a short video, hesitating reads as a rewatch, which reads as strong interest.
• Opening the profile or the comments.A deliberate act, weighted accordingly. Reading the comments to see other people being annoyed is indistinguishable from enthusiasm.
• Explicit thumbs-down.Real but weak, and it applies to one post rather than a theme. It is the control people rely on most and the one that moves the feed least.

Why hate-watching is the fastest way to make it worse

The single most common way people train a feed toward the thing they are trying to escape is by disapproving of it slowly. There is no signal for disapproval, only for attention, and the system cannot tell the difference between fascination and irritation.

This is why a feed can drift toward material you actively dislike and why it feels almost adversarial when it happens. It is not adversarial. It is a system optimising for the thing you demonstrably do, which is stop and look, rather than the thing you would say if asked.

The discipline that works is unglamorous: scroll past within about a second, without stopping to react, every time. It is much harder than tapping "not interested", and it is the part that actually moves the feed.

The honest caveat is that this only holds while you can do it consistently. If a feed is something you reach for at midnight when your judgement is worst, retraining is a project you will keep restarting. That is the point at which removing access does more than tuning the recommendations does.