How does the Instagram algorithm actually work?

Is there one algorithm?

No, and this is the single most useful thing to understand. Adam Mosseri has said repeatedly that Instagram uses different ranking systems for different surfaces. Feed has one, Reels has another, Stories another, Explore another. Each system is tuned for what people do there.

Stories ranking leans heavily on relationship, because people mostly watch stories from accounts they care about. Reels ranking leans on entertainment signals like watch time and sends, because most Reels you see come from accounts you do not follow. Explore sits somewhere in between.

Practically, this means a post that flops in feed can still travel on Reels, and a format that works in Stories tells you almost nothing about what will work in Explore. Judge each surface on its own numbers.

What signals actually matter?

Fewer than the guru threads suggest, and Meta has walked through the main ones on the record. For Reels, the big ones are watch time (did people finish it, did they rewatch it), likes per reach, and sends per reach, meaning how often people share it through DMs. Mosseri has specifically called out sends as a signal creators should watch.

For feed, relationship matters more: do you DM this person, do you comment on their posts, do you look them up by name. Recency counts everywhere. A three-day-old post rarely gets pushed hard, which is part of why consistent posting beats occasional masterpieces.

Comments matter too, though my read from running accounts is that a DM share is worth more than a generic 'love this' comment. Instagram has never published exact weightings, so anyone quoting precise percentages is making them up.

• Watch time and completion rate, especially on Reels
• Likes per reach, not raw likes
• Sends per reach (DM shares), which Mosseri has highlighted directly
• Relationship signals: DMs, profile visits, comments over time
• Recency of the post

What can you actually control?

You cannot control the ranking system, but you control the inputs it reads. Hooks decide watch time, so the first second of a Reel matters more than the last thirty. Making content people want to send to a friend is the most reliable growth lever I know of right now: relatable, useful, or funny enough to forward.

You also control consistency. The system needs a steady stream of your content to learn who responds to it. Two posts a month gives it almost nothing to work with.

What you cannot control: who else posted that day, what topics Instagram is currently pushing, and the general fact that reach is noisy. Two near-identical posts can perform very differently for no visible reason. That is normal, not a penalty.

Is the shadowban real?

Instagram says it does not secretly suppress accounts that follow the rules, and it now shows account status tools so you can check whether your content is eligible for recommendations. If a post breaks recommendation guidelines, it can still reach followers but will not be pushed to non-followers, which feels like a shadowban even though Instagram would call it something else.

Most 'I got shadowbanned' cases I have looked at were actually a run of weak hooks, a topic change the audience did not follow, or normal reach variance. Check Settings, then Account Status, before assuming a conspiracy. If everything shows as fine there, the honest answer is that the content underperformed.

Does the algorithm hate links or certain words?

The 'algorithm punishes link stickers' claim has never been confirmed by Meta, and Mosseri has directly denied that Instagram downranks posts for mentioning competitors like TikTok. What is true: content that sends people off-platform tends to get fewer of the engagement signals that drive reach, because people who tap a link stop watching. The effect is real, the punishment framing is wrong.

Same with engagement-bait folklore like 'never edit a caption after posting.' Instagram has never documented any such penalty. When a claim is unverifiable, file it under superstition and get back to the inputs you can actually measure.