How does Instagram decide which Reel you see next?
Short answer: Meta documents the steps for one Reels system: collect reels you might like, have a lightweight model pick about 100, predict how you'll react, then give them a relevance score and order them. Instagram lists your recent activity first among its Reels signals, so one like is only one input among many.
The mechanism at a glance
The mechanism, step by step
- Candidates are collected. The system collects reels you might like. In our model, your recent activity is one of the inputs. OfficialOur modelIllustrative
- Unsuitable reels are removed. Before ranking, reels that are non-recommendable or may break the Community Guidelines are taken out. Official
- About 100 are shortlisted. A lightweight model picks about 100 of the most relevant reels. OfficialIndustry pattern
- Your reaction is predicted. For each reel: how likely you are to watch more of it than 95% of viewers of same-length reels, reshare it, comment, or watch less than three seconds. Official
- Scored and ordered. Those reels get a relevance score and are put in order by it. OfficialIllustrative
- The top reel plays. In this walkthrough, the reel at the top of the order is the one that plays next. OfficialIllustrative
- Your reaction feeds back. In our model, how you react to it becomes part of your recent activity, which feeds the next pick. Our model
How it works#
Instagram has said it doesn’t run one single algorithm, and that Feed, Stories, Explore and Reels each have their own ranking process. Evidence: Reported For one Reels system, called Reels Chaining, Meta publishes a “system card” that lists the steps. Evidence: Official
Before ranking, the system removes content that is “non-recommendable” or may break the Community Guidelines. Evidence: Official The card’s steps are: collect all the reels you might like, run a lightweight model to pick about 100 of the most relevant, predict how you’ll react to each one, give those reels a relevance score, and put them in order by that score. Evidence: Official
The reactions it predicts include how likely you are to watch less than three seconds, watch more of a reel than 95% of people who watched reels of the same length, comment, reshare, share off Instagram, follow the author, tap “Interested” or use the reel’s audio. Evidence: Official
What goes in? Instagram lists its Reels signals “roughly in order of importance”. First comes your activity: reels you liked, saved, reshared, commented on or engaged with recently. Then your history of interacting with the person who posted, then information about the reel, such as its audio, visuals and popularity, and last, information about the person who posted, such as their follower count and engagement level. Evidence: Official The system card’s example inputs also include whether you’ve been viewing reels for more than 15 seconds, how many times a reel was dismissed, and the authors of reels you watched for less than three seconds. Evidence: Official
So a like is one of several kinds of recent activity Instagram lists, alongside saves, reshares and comments. Evidence: Official
Architecture#
In our model, it works like a funnel with a feedback loop. Evidence: Our model Systems like this commonly use a cheap first-stage model to narrow a huge pool before a more expensive step ranks the shortlist, and the card’s lightweight model picking about 100 reels Evidence: Official fits that pattern. Evidence: Industry pattern
The loop is our reading too: your reaction to the reel that plays becomes part of your recent activity Evidence: Our model, which is the first signal group Instagram lists. Evidence: Official
Who's involved
- Your activityReels you liked, saved, reshared, commented on or skipped recently.
- Integrity filtersRemove reels that are non-recommendable or may break the Community Guidelines.
- Lightweight shortlist modelCheaply narrows the candidates to about 100.
- Reaction predictionsEstimate how likely you are to watch, reshare, comment, follow or swipe away.
- Relevance scoreCombines the predictions and puts the shortlist in order.
- Your controlsNot interested, newest-first, Your Algorithm topics and reset change the inputs.
The walkthrough at the top shows how these parts connect, step by step.
Tradeoff#
Our inference: because Instagram lists recent activity first Evidence: Official, a handful of recent interactions can noticeably shift what you’re shown. Evidence: Our model
Instagram describes some controls. “Not interested” makes the system try to filter out similar content, and “newest reels first” shows reels from accounts you follow in reverse chronological order. Evidence: Official “Your Algorithm” shows a summary of the topics shaping your recommendations and lets you ask for more or less of a topic. Evidence: Official In November 2024 Instagram began testing a reset option that clears your recommended content; afterwards, recommendations personalize again over time. Evidence: Official
Verified vs illustrative#
The steps and the shortlist of about 100 come from Meta’s Reels Chaining system card Evidence: Official, as do the example input signals Evidence: Official and the predicted reactions. Evidence: Official The order of the Reels signal groups comes from Instagram’s ranking explainer, which describes it only as “roughly in order of importance”. Evidence: Official
Some things on this page are our model or illustration rather than Meta’s documentation, and each is labelled where it appears: the reason for the cheap first pass Evidence: Industry pattern, the feedback loop and your activity feeding the next pick Evidence: Our model, the idea that a few recent interactions can shift what you see Evidence: Our model, the top reel playing next in the walkthrough Evidence: Illustrative, and the walkthrough’s dots and their numbers. Evidence: Illustrative
Documented8 claims
Official documentation, or confirmed by at least two independent reputable sources.
- Instagram lists the Reels signals 'roughly in order of importance': (1) your activity, meaning reels you liked, saved, reshared, commented on or engaged with recently; (2) your history of interacting with the person who posted; (3) information about the reel, such as audio, visuals and popularity; (4) information about the person who posted, such as follower count and engagement level. Sources: [1]
- Meta's Reels Chaining system card describes these steps: collect all reels you might like, run a lightweight model to pick about 100 of the most relevant, predict how you will react, give those ~100 a relevance score, and put them in order by score. Sources: [2]
- Example input signals in the system card: whether you've been viewing reels for more than 15 seconds, how many times the reel was dismissed, the authors of reels you watched for less than three seconds, the length of the reel, and which authors you follow. Sources: [2]
- Other predictions in the system card include: how likely you are to watch less than three seconds, watch more of a reel than 95% of people who watched reels of the same length, comment, reshare, share off Instagram, follow the author, tap 'Interested', or use the reel's audio. Sources: [2]
- Before ranking, the system removes content that is 'non-recommendable' or may break Community Guidelines. Sources: [2]
- You can influence the Reels you're shown: 'Not interested' makes the system try to filter out similar content, and 'newest reels first' shows reels from accounts you follow in reverse chronological order. Sources: [2]
- 'Your Algorithm' (launched Dec 10, 2025, starting in Reels in the US) shows an AI-generated summary of the topics shaping your recommendations and lets you ask for more or less of a topic. It reached Explore on Apr 15, 2026 and all of Instagram (Feed, Reels, Explore) on Jun 10, 2026. Sources: [3]
- In Nov 2024 Instagram began testing a 'reset recommendations' option that clears your recommended content across Explore, Reels and Feed. Afterwards the recommendations personalize again over time. Sources: [4]
Reported1 claim
A single report, or accounts that disagree. Treat with care.
- Instagram says it does not run one single algorithm. Each part of the app (Feed, Stories, Explore, Reels) has its own ranking process. Sources: [1]
Industry pattern1 claim
How systems like this are commonly built; not confirmed for this company.
- Large recommendation systems commonly use a cheap first-stage model to narrow a huge pool of candidates before a more expensive model ranks the shortlist.
Our model2 claims
Our reasonable reading of the sources, not a confirmed internal detail.
- Your reaction to the reel that plays becomes part of your recent activity, so it feeds the next pick.
- Because recent activity weighs most, a handful of recent interactions can noticeably shift what you're shown.
Illustrative2 claims
Simplified examples to explain the idea, not factual claims.
- The walkthrough shows the reel at the top of the order playing next. That is a simplification of how the ordered reels are shown.
- In the walkthrough, dots stand for candidate reels. How many there are, which are removed and their order are made up; they are not real counts, scores or weights.
Sources#
- Instagram Ranking Explained, Instagram (Adam Mosseri), published . Accessed . (primary)
- Instagram Reels Chaining, Meta Transparency Center. Accessed . (official)
- Control Your Instagram Reels Algorithm, Instagram, published . Accessed . (primary)
- Reshape Your Instagram With a Recommendations Reset, Meta Newsroom, published . Accessed . (primary)
Updates and corrections
No corrections since publication. Spotted an error? See how corrections work.
Use it
- Instagram Reels ranking: the official sources: The four official pages that explain how Reels are ranked and what you can control, in the order worth reading them.