Almost everything written about the Instagram algorithm is a paraphrase of a paraphrase. This is an attempt to separate what the platform has actually published from what has accumulated around it, because the second category is quoted with the same confidence as the first.
What is published
There is no single algorithm. Instagram has said since 2021 that each surface — Feed, Stories, Explore, Reels — uses its own ranking, tailored to how people use that surface. Advice that refers to “the algorithm” as one system is already imprecise.
Ranking is described in signal categories. The 2023 ranking explanation lists, for Reels: your activity, your history of interacting with the person who posted, information about the reel, and information about the person who posted. Note what these are — categories, not weights. Instagram has not published the relative importance of the inputs inside them.
Specific demotions are named. The same document states that Instagram aims to make “low-resolution or watermarked reels, reels that are muted or contain borders, reels that are majority text, or reels that have already been posted on Instagram” less visible. This is the most concrete and actionable sentence the platform has published, and it is mostly ignored in favour of tactics. We wrote it up separately as five things Instagram says it demotes.
Reels is built around accounts you do not follow. The creator guidance describes Reels as oriented toward discovery and entertainment, with the majority of content coming from accounts the viewer does not follow, and lists how many and how quickly people like, comment, share and save among the inputs.
What is widely repeated with no primary source
Exact signal weights. The claim that sends carry “three to five times” the weight of a like appears across dozens of marketing blogs. We have not found a primary statement behind it. The underlying ordering — that sends matter more than likes for reaching non-followers — is reported consistently enough to take seriously. The multiplier is not.
Precise thresholds. Numbers like a specific watch-through percentage required for distribution, or an exact posting frequency, circulate widely. These are inferences from small private datasets at best, and invented at worst. Instagram has never published a threshold.
Shadowbanning as a described mechanism. Reduced distribution is real and named — the demotion list above is reduced distribution. What does not exist in any published source is the elaborate machinery the term usually implies.
The gap between “Instagram said this” and “everyone says Instagram said this” is where most short-form advice lives.
The January 2025 reporting
The most-cited recent claim is that Mosseri named watch time, likes per reach and sends per reach as the three heaviest signals. It is reported consistently across many outlets, which is meaningful, but it reaches us through secondary coverage rather than a durable published post. We treat it as well-attested and say so when we use it — see watch time is the signal — rather than presenting it as documentation.
Why we bother with the distinction
Because a claim's sourcing predicts how long it stays true. Published descriptions of how a surface works change slowly and get updated. Numbers invented to fill a paragraph never get corrected, and they propagate by being exactly the kind of concrete detail that makes writing feel authoritative.
If you take one habit from this page: when you read a precise figure about a platform, look for who first said it. Often the trail ends at a blog citing another blog. For a case where that trail has been followed all the way down, see the eight-second attention span.