The Instagram Reels algorithm is one of several ranking systems Instagram runs, one for each part of the app, and Instagram's ranking explainer says it orders reels by predicting how likely each viewer is to reshare a reel, watch it all the way through, like it and open its audio page. Those predictions lean on the viewer's recent activity, their history with you, information about the reel and information about you. Meta's Reels system card adds that a lightweight model first narrows the candidates to approximately 100 reels, removes anything non-recommendable, and then scores and orders what is left.
Everything below comes from Instagram's and Meta's published material: the ranking explainer written by the head of Instagram, the Instagram for Creators blog and Meta's AI system cards. Where other guides quote weightings or multipliers, we leave them out, because none of these sources publish weights.
One app, several algorithms
Instagram says it does not have a singular algorithm. Feed, Stories, Explore, Reels and Search each use their own, tuned to what people go there for: closest friends in Stories, discovery in Explore and entertainment in Reels. The explainer also notes that much of what people watch in Reels comes from accounts they do not follow, which is why a reel can travel far beyond your follower list while a Story mostly cannot.
That split matters when you read your numbers. A reel competing for strangers' attention is ranked on different evidence from a Story shown to friends, so a tactic that lifts one surface can do nothing for another. Each surface needs its own review, which is what the signal table further down is for.
How a new reel reaches people who don't follow you
Instagram describes the route a reel takes in two creator blog posts. In April 2024 it announced that every eligible reel would first be shown to a small audience likely to enjoy it, regardless of whether they follow you, and that the best performers would then be shown to progressively wider groups. Content counted as eligible when it was original, carried no visible watermark and met both the Community Guidelines and the Recommendation Guidelines.
In February 2025 Instagram said those changes were fully launched. The same post lists the signals it uses to predict which reels people might like, naming watch time, retention, shares, likes and comments, and it describes audience matching, where what people with similar interests engage with informs what you are shown. Its stated aim is to distribute reels on engagement rather than on the creator's follower count alone.
Length has its own rule. Instagram's January 2025 update said reels up to three minutes long would become eligible for recommendation to non-followers, starting with creators in the United States, where previously only reels up to 90 seconds qualified.
What the Reels ranking predicts
The ranking explainer lists the predictions that matter most: whether you will reshare a reel, watch it to the end, like it and go to its audio page, which Instagram treats as a hint that a reel might inspire you to make your own. Instagram also surveys people about whether a reel was worth their time or fun, and learns from the answers. The signals behind those predictions, roughly in order of importance, are your activity, your history with the person who posted, information about the reel such as its audio, visuals and popularity, and information about the person who posted.
Meta's Reels Chaining card, last updated in November 2025, goes further, listing predictions for whether a viewer will watch less than three seconds of a reel, comment, reshare, share it off Instagram, follow the author, use the audio and tap “Interested”. One prediction on that card is worth reading twice: how likely a viewer is to watch more of a reel than 95% of users who watched reels of the same length. Retention is judged against reels of similar length, so a long reel is not compared with a ten-second clip.
The early-skip prediction lines up with Instagram's own creative advice. The February 2025 post tells creators to make reels that stand on their own for strangers, to open with a hook in the first three seconds and to check view rate in Insights to see what share of people keep watching beyond those seconds. Meta adds a caution of its own: the models and their input signals change frequently, so treat any list as a set of examples rather than a scoring sheet.
Signal-by-surface table
This table maps what Instagram says each surface predicts to the Insights metric that comes closest to it. The metric is a proxy you can see, not the model's actual input.
| Surface | What Instagram says it predicts or weighs | Closest Insights metric | Source |
|---|---|---|---|
| Reels tab | Reshares, complete watches, likes and audio-page visits; skips under three seconds; watching more than most viewers of same-length reels; follows | Views, watch time, average watch time against the reel's length, shares, follows | Reels Chaining card and ranking explainer |
| Explore | Likes, saves and shares, with popularity signals mattering more than in Feed or Stories; follows; watching most of a video; “Not Interested” taps | Saves, shares, follows and views from non-followers | Explore card and ranking explainer |
| Feed, from accounts you follow | Time spent, comments, likes, shares and profile-photo taps; carousel swipe-throughs; sending a post by direct message; how recently it was posted | Interactions, profile activity and views on the post | Feed card and ranking explainer |
| Suggested posts in Feed | How informative a post is, assessed by a language model; profile clicks; comments; turning audio on; opening reels full screen | Profile activity, interactions and non-follower views | Feed Recommendations card |
| Stories | Viewing history, engagement history and closeness; tapping in, replying, sending a like, swiping away or exiting, answering polls and sliders | Viewers, story interactions such as replies and sticker taps, profile activity | Stories card and ranking explainer |
The definitions for every metric in the third column are collected in our Instagram Insights glossary. If you want a single number to track per reel, keep it to one you can define precisely; our engagement rate worksheet shows three ways to do that.
What makes a reel less visible
Instagram names the conditions that shrink a reel's distribution, and they are mostly production choices you control:
- Low resolution, visible watermarks, muted audio, borders, or frames that are mostly text, all named in the ranking explainer.
- Reels that have already been posted on Instagram. Since 2024, when two identical pieces of content exist, Instagram says it recommends only the original.
- Unoriginal content. Instagram's original content guidelines say accounts that mainly repost others' work without material edits may not be recommended, and an April 2026 update extended that to photos and carousels.
- Anything in the Recommendations Guidelines categories, from sexually suggestive material to engagement bait. For creators near those lines, our map of Instagram's rules for adult creators sets out which content is removed and which is simply not recommended.
- An account that is not eligible for recommendations at all, which you can check in Account Status.
The explainer also answers a common suspicion directly: Instagram says it does not suppress content to push people into buying ads, and that reach should go to whoever makes the content people engage with most.
Weekly signal review sheet
Copy one row per reel into a spreadsheet each week, checking every reel at the same age so the comparison is fair. Fill in only what Insights shows you; leave a cell blank rather than estimating it.
| Column | What to record | Which prediction it approximates |
|---|---|---|
| Reel, date, length | A short name, the publish date and the running time | Context for the same-length retention comparison |
| Views and viewers | Both figures from the reel's insights | How far the staged distribution carried it |
| Average watch time | The figure, then divided by the reel's length | Watching to the end and length-adjusted retention |
| View rate | The share still watching after the opening seconds, where your Insights display it | The early-skip prediction |
| Shares | Shares from the interactions breakdown | Reshares and sends to friends |
| Likes, comments, saves | Each count on its own, not summed | The like, comment and save predictions |
| Follows | Accounts that followed after viewing the reel | The follow-the-author prediction |
| Account Status | Any recommendation flag on the reel or the profile that week | Whether the reel was eligible for distribution at all |
Then answer five questions before planning the next week:
- Which reels held attention best relative to their own length?
- Which reels earned shares and follows, not only likes?
- Did any reel stall at the start, with a weak view rate, and what did its opening look like?
- Did Account Status flag anything that would explain a reel that never travelled?
- What single change will you test next week, and how will you keep everything else the same?
For the last question, the Reels test planner covers controlled test design, and Instagram trial reels let you try an idea on non-followers before your followers see it.
Claims Instagram's own material does not support
- Fixed multipliers. Instagram lists its signals roughly in order of importance and publishes no weights, so any claim that one share equals a set number of likes is not from Instagram's written material.
- One master algorithm. Instagram says the opposite, and Meta publishes a separate system card per surface.
- Trials dragging down regular reels. The trial reels help page says trials are evaluated independently and don't affect the ranking of your standard reels.
- Hashtags as the main Reels lever. Instagram's reach tips mention keywords and hashtags as a way to be found through Search, not as a top Reels ranking signal.
Turning these signals into an account plan is a different job, covered in how to grow on Instagram as a creator.
Limitations of this explanation
Instagram publishes predictions and example signals, never weights, thresholds or the size of each testing audience, so no outside guide can tell you exactly why one reel travelled and another did not. Rankings are personalised, which means the same reel is scored differently for every viewer. The ranking explainer dates from May 2023, while the system cards are revised more often and describe themselves as dynamic, so details here may be out of date by the time you read them.
The Insights metrics in our tables are proxies you can observe, not the inputs the models use, and some of them are labelled by Meta as estimated. Use the sheet to spot patterns across many reels over several weeks, not to diagnose a single post, and expect results to vary with your niche, your audience and the country rollout of each feature.