The Instagram algorithm: how it works and which signals actually matter

Talking about "the Instagram algorithm" is a simplification that gets in the way. Every surface in the app runs its own system, because each one is used with a different expectation: in Feed people expect to see who they follow, in Reels they expect to discover something new, in Stories they expect to keep up with people they are close to.

What they share is the method. The system takes the available pool of content, runs a prediction of how likely you are to interact with each piece, and orders from that. There is no absolute quality score: there is a prediction made for you specifically, which explains why two people following exactly the same accounts see different feeds.

That changes the right question. It is not "how do I please the algorithm", it is "who is this content so obviously for that they stop scrolling, and can the system predict that".

Instagram publishes which signals matter on each surface. According to the platform itself, in Instagram Ranking Explained, the rough order of importance is this:

Notice the pattern: on relationship surfaces (Feed and Stories), the history between the two accounts dominates. On discovery surfaces (Reels and Explore), the performance of the piece itself climbs the list. That is why a Reel can take off on a small account and a feed post almost never does.

The public conversation is all about what raises reach, but the fastest route in the other direction is negative signals. Skipping a video in the first moments, tapping "not interested", hiding a post, muting or reporting are actions that cut delivery far more efficiently than a like raises it.

There is an asymmetry worth internalising: content delivered to the wrong audience is not neutral, it is expensive. It accumulates negative signal, and that signal informs the next delivery. This is the mechanism behind a reach drop after an off-topic post, and also after a viral that pulled in people with no real interest in the profile.

Dwell time is the other side of it. How long someone stays on your content before scrolling says more than a like, because it does not depend on them deciding to act. That is what makes carousels and video structurally advantageous: both ask for time.

Delivery to people who do not follow you passes an extra filter: the recommendation guidelines. A piece can sit perfectly fine on your profile and still be barred from discovery surfaces, with no warning and no notification. That mechanism is what most people call a shadowban.

Two causes repeat more than all the others. The first is reposting without transformation: third party material published as your own gets reduced distribution by policy, and crediting the source in the caption does not fix it. The second is video carrying another app's watermark, which the system detects automatically.

The practical reading is boring and true: the recommendation path rewards people who produce, not people who aggregate. If your operation depends on reposting, it depends on a lane the platform chose to narrow.

• Pick an interest cluster and stay in it. The system needs to know who to offer your content to. Switching subject resets part of that learning.
• Optimise for shares and saves. They are the interactions that push outside the bubble, and the ones that cost the viewer the most.
• Treat the first seconds as the product. In video, the hook decides whether distribution happens at all.
• Post on a predictable cadence. Regularity matters more than volume.
• Do not chase the "5-3-1 rule" or similar formulas. They are not platform rules, they are market heuristics. They can work as routine discipline, but there is no mechanism behind them.
• Check account status before theorising. There is a screen that reports any active restriction.

• I know which subject my account is classified under and I am posting inside it.
• My content is mine, with no reposting and no other app's watermark.
• I am measuring saves and shares, not just likes.
• I tested the hook separately from the rest of the video.
• I checked the account status screen before concluding there was a penalty.
• I compared performance against the median of the last thirty posts, not the previous one.