There’s a moment most of us have experienced. You pause on a video, maybe something unsettling from a conflict zone, just long enough for curiosity to take hold. A few minutes later you’re four videos deep into war footage you never went looking for. The algorithm noticed that pause. It remembered. And now it thinks this is what you want.
That’s the thing about social media algorithms. They are watching everything, not just what you like or follow, but where you slow down, what you rewatch, and what you almost scroll past. Understanding that is the first step to actually doing something about it.
Algorithms learn from behaviour, not settings
Most people’s first instinct when their feed goes wrong is to dig into settings. And yes, platforms give you toggles and controls. But your behaviour is a far stronger signal than anything in a settings menu.
I’ve seen this play out across different platforms. My Instagram has essentially become a couples account, full of funny content I share with my partner because she’s on Instagram but not TikTok. The result is that my Instagram and TikTok feeds are completely different. Same person, same general interests, but two entirely distinct algorithms trained by two entirely different patterns of use. Neither feed is wrong, they just reflect who I am on each platform.
My other half had the unfortunate mistake of dwelling on a video of a toenail injury one day. Weeks of feet she had to endure before realising scrolling faster past these videos meant they slowly disappeared.
Bluesky is an interesting contrast. It runs with very little algorithmic curation by default, which sounds liberating until you realise how much quiet work a good algorithm does on your behalf. I use it regularly and I enjoy the rawness of it, but there are moments where you notice what’s missing. The serendipity that a well-trained feed provides is real, even if the way it gets there feels uncomfortable.
This brings to mind the time I switched from Google to DuckDuckGo years ago. The privacy case was compelling, and I stuck with it for a while. But the gap in search quality was hard to ignore. Google’s results were sharper because Google knew me. Creepy, absolutely. But also genuinely useful in a way that was difficult to replicate elsewhere.
The two levers you actually have
When it comes to reshaping your algorithm, everything broadly falls into one of two approaches: starving it of bad signals, or feeding it better ones. Both matter, and the most effective approach combines them.
Starving the algorithm
The algorithm learns from engagement, and engagement includes more than likes and comments. Hovering on a post, watching a video past the halfway point, even opening something and closing it quickly all send data. Negative engagement, such as leaving an angry comment, can still reinforce a topic rather than suppress it. The recent fuel protests showed many of us this.
To actually starve an unwanted signal, you need to move past content quickly and without interaction. Most platforms also give you an explicit “not interested” option, usually accessible through a three-dot menu on a post or video. Use it and use it aggressively. On platforms like X, muting keywords and topics can effectively remove entire categories of content from your feed without having to unfollow anyone. If you don’t see it, you can’t feed the algorithm engagement on it.
The logic here is simple. If the algorithm has nothing to learn from, it cannot reinforce what it already knows about you.
Feeding it something better
The flip side is actively building new signal. Follow accounts that reflect where you want your feed to go. Watch videos in that space to the end. Save posts. Share things. These are all strong positive signals that tell the algorithm what you actually want more of.
The people who build these apps and algorithms think the world is perfect and that everyone is sitting at home making curated save lists of their favourite content. I know that us Irish are not very likely to make the most of these kinds of features.
But this means you can take algorithm signals to a deliberate extreme, spending a week or two engaging only with content from a specific niche to force a pivot. If you want to move from one type of content to another, the combination of muting and unfollowing the old while actively engaging with the new is far more effective than either approach alone.
The nuclear options
Sometimes you want a clean break rather than a gradual adjustment. A few approaches can get you there faster, though each comes with trade-offs.
Deleting and recreating an account gives you a completely blank slate. You lose your history, your follows, and any personalisation that was working in your favour, but if your feed has become genuinely unpleasant, it can be worth it. Some platforms also have a built-in reset option for recommendations, which is worth checking before going the full deletion route.
Just know that a clean start can be as risky as a long re-point. Platforms will feed you expansive content to see what you’re interested in early doors, and you could be back at the start in a few days.
Switching platforms entirely is an option too, though it is more of an escape than a reset. It does not solve the underlying issue of how to manage an algorithm; it just gives you a fresh start on different terms.
Turning off recommendations entirely, something YouTube allows, is a middle ground. You keep your account and subscriptions but remove the algorithmic feed. What you lose is discovery. What you gain is control.
Secondary accounts and compartmentalisation
One underused approach is running separate accounts for different contexts. I ended up with a couples algorithm on Instagram almost by accident, but there is a logic to doing it deliberately. If part of your usage is for work research and part is for personal interests, mixing those signals trains the algorithm in contradictory directions.
A secondary or anonymous account lets you consume content in a specific niche without it bleeding into your main feed. It is not a perfect solution, and maintaining multiple accounts adds friction, but for people whose professional and personal interests are genuinely distinct, it is worth considering.
Patience is part of the process
None of this happens overnight. Algorithms have memory, and that memory does not clear instantly. Consistent behaviour over days and weeks is what actually shifts recommendations, not a single afternoon of furious unfollowing.
The platforms are also not passive in this. Creators and the platforms themselves actively optimise for engagement, which means the system is always nudging you back towards content that keeps you scrolling. Taking control of your algorithm is an ongoing practice rather than a one-time fix.
The good news is that once you understand what these systems are actually measuring, you stop feeling like a passive recipient of whatever they decide to show you. You are always training the algorithm, whether you intend to or not. You may as well do it deliberately.
