Instagram Growth
How the Instagram Reels Algorithm Works in 2026
The three signals that decide Reels distribution, how the cold start works, why Instagram now reads your video, and a diagnostic for a Reel that died at 400 views.
Published Updated 9 min read
On this page
- The three ranking signals that matter
- How a Reel actually gets distributed
- Instagram now reads the video
- Original content wins, reposts get throttled
- Watch time and skip rate: the part you can actually control
- Sends per reach is the growth metric
- What actually kills Reels reach
- A seven-day diagnostic for a Reel that flopped
- Cadence, format mix and the carousel counterweight
- Frequently asked questions
- How does the Reels algorithm work in 2026?
- Do hashtags help Reels reach?
- Why did my Reel get 300 views?
- Should I delete a Reel that flopped?
- Does posting time still matter?
Every Instagram Reels algorithm explanation eventually collapses into a list of tricks: post at 6pm, use trending audio, add 30 hashtags, delete and repost. None of those are the algorithm. The algorithm is a prediction system that asks one question about every video: will this specific person watch this specific clip to the end, and will they send it to someone?
In 2026 that question is being answered with better inputs than ever, including a model that reads what is actually in your video. Which means the tactics that used to fake signal stopped working, and the ones that produce real signal got more valuable.
Here is how the Reels algorithm is actually structured this year, what each stage rewards, and how to diagnose a Reel that died at 400 views.
The three ranking signals that matter #
Instagram has been unusually explicit. Adam Mosseri has named three signals as carrying the most weight across the app, and they apply directly to Reels:
| Signal | What it measures | Where it matters most |
|---|---|---|
| Watch time | Total seconds watched, weighted by completion and replays | Reaching non-followers |
| Likes per reach | Likes divided by accounts reached, not raw likes | Reaching your existing followers |
| Sends per reach | DM shares divided by accounts reached | Reaching non-followers |
Two things fall out of that table immediately. First, raw likes are a vanity metric: the ratio is what counts, so a Reel seen by 100,000 people with 2,000 likes is weaker than one seen by 10,000 with 900. Second, sends are the growth signal. Likes mostly tell Instagram that your followers still like you; sends tell it that strangers should see this.
Everything else — comments, saves, profile visits, follows — is a secondary signal. They matter, but they matter because they correlate with the three above, not because they are separately weighted.
How a Reel actually gets distributed #
Distribution is staged, and each stage has a different bar.
Stage 1: the trial audience. Your Reel is shown to a small slice of people, a mix of followers and non-followers whose interests match the topic Instagram thinks the video is about. This is the cold start. Performance here is measured over the first few hundred views, and the numbers that decide whether the next stage happens are completion rate and sends.
Stage 2: expansion. If the trial audience watches and shares, the video is pushed to a larger, less similar audience. Content that survives this stage usually has a hook that works without context, because the second audience does not know you.
Stage 3: saturation. Eventually the audiences that would watch it have watched it. Sends per reach falls, distribution slows, and the Reel settles into a long tail that keeps trickling views from search and profile visits for months.
The practical consequence: a Reel’s first hour is diagnostic, not final. Most Reels that “flop” were never given stage 2, because completion or sends were too low in stage 1. Deleting and reposting rarely helps, because the next cold start will measure the same video the same way.
Instagram now reads the video #
The biggest structural change in 2026 is that Instagram classifies Reels by their content, not just by their engagement. The model looks at visuals, on-screen text, audio and caption together to decide what the video is about, then matches it to interest graphs.
This has two effects that creators feel but rarely name:
- Topic consistency is rewarded. An account that posts consistently about one thing teaches the model what audience to test it with. An account that alternates between cooking, crypto and comedy forces a fresh cold start every time, and the trial audiences get worse.
- On-screen text and spoken words are indexed. Saying “sourdough starter troubleshooting” out loud, and putting the phrase on screen, gives the classifier an unambiguous label. Aesthetic-only Reels with no verbal anchor are harder to place.
If your reach dropped without your content changing, look at whether your topics drifted. A single off-topic viral hit can reset the model’s expectations for the next ten posts.
Original content wins, reposts get throttled #
Instagram gives more distribution to the original creator than to accounts that repost. Accounts whose feeds are mostly other people’s content stop being recommended to non-followers entirely. This is the aggregator penalty, and it applies to brands reposting creator content without adding anything, and to creators reposting TikToks with the watermark intact.
The line is not “never repost”. It is “add something the original did not have”: a reaction, a correction, context, a stitch, a second angle. A repost with commentary is original content. A repost with a new caption is not.
We wrote about how the penalty works in practice in Instagram original content and the aggregator penalty.
Watch time and skip rate: the part you can actually control #
Watch time is not the same as length. A 12-second Reel watched twice has more watch time than a 60-second Reel abandoned at eight seconds, and it also has a much better completion rate. Short, dense and rewatchable beats long and thorough, which is why the best-performing Reels in most niches are 7–15 seconds.
Skip rate is the inverse measure: how fast people swipe past. It is decided almost entirely in the first one to three seconds, which is why the hook is the highest-leverage part of the video. A weak hook on strong content produces exactly the pattern most creators misread as “the algorithm hates me”: decent watch time from the people who stayed, terrible reach because most people left.
Three fixes that move watch time more than editing tricks:
- Start with the payoff, not the setup. Show the finished dish, the fixed bug, the final design. Then explain how you got there.
- Cut the first two seconds entirely. Most Reels have an intro that exists because the creator needed a run-up. The viewer does not.
- Design for a second watch. A detail people missed the first time, a list they want to screenshot, a punchline that lands differently once you know the end.
Our breakdown of Reels watch time and skip rate has the retention benchmarks by length, and Reels hooks covers the first three seconds in detail.
Sends per reach is the growth metric #
If you only track one number on Reels, track sends per reach. It is the cleanest proxy for “would someone send this to a friend”, which is the behaviour that pushes a video beyond your followers.
Calculating it is simple: DM shares divided by accounts reached, as a percentage. A Reel with 5% sends per reach is exceptional; 1–2% is healthy; under 0.5% means the video was watched but not worth passing on.
How to engineer sends without begging for them:
- Make it about a person the viewer knows. “Send this to the friend who always orders the same thing” converts because it names a real relationship.
- Give people social currency. A fact, a correction or a shortcut they can use to look useful in a group chat.
- Be specific enough to be quotable. “Post at 6pm” is forgettable. “The 6pm advice is wrong for accounts under 5K” is a sentence someone repeats.
- Create mild disagreement. Content that takes a defensible side gets forwarded to argue with, and arguments are sends.
What actually kills Reels reach #
Watermarks. A TikTok watermark signals reposted content and depresses distribution. Download clean or re-export.
Topic drift. Three unrelated topics in a week means three cold starts with weaker audiences.
Slow openings. A logo animation, a “hey guys”, a two-second pan of the room. All of it is skip rate.
Text that repeats the voiceover. On-screen text should add information, not transcribe it. Duplicated words waste the classifier’s signal and the viewer’s attention.
Posting into a dead hour. Timing matters less than it used to, because Reels are distributed over days rather than minutes, but posting when your audience is asleep still wastes the first-hour window.
Deleting and reposting. It resets the cold start without changing the inputs. Fix and repost in six to eight weeks instead, when a new audience exists.
Chasing a trend you cannot improve on. If the trend’s joke is already complete, your version is a repost with extra steps.
A seven-day diagnostic for a Reel that flopped #
Wait 72 hours before judging. Then work through this in order.
| Check | What it tells you | Fix |
|---|---|---|
| Views under 1,000 with low completion | Hook failed; the trial audience left early | Re-cut the first two seconds, repost in 6–8 weeks |
| Good completion, low sends | Entertaining but not useful or shareable | Add a takeaway or a reason to forward it |
| High likes per reach, low reach | Your followers liked it; strangers were never shown it | Make the topic legible in the first frame |
| Strong saves, weak views | Useful to a narrow group | Good problem: repurpose into a carousel and a pinned post |
| Views fine, follows near zero | No reason to stick around | Add a clear through-line for the account, not just the video |
The same diagnostic works in reverse: when a Reel outperforms, identify which row it hit and repeat that input deliberately rather than copying the surface detail (the audio, the caption style) that probably did not cause it.
Cadence, format mix and the carousel counterweight #
Reels are the reach format, carousels are the save format. Buffer’s 2026 analysis of four million posts found Reels get 36% more reach than carousels, while carousels get 12% more engagement than Reels. Running only Reels maximises reach and starves the saves that build a content library people return to.
A workable operational mix for most accounts: three Reels and two carousels a week, with Stories carrying the daily relationship work. If you cannot sustain three Reels, two well-hooked Reels beat five rushed ones, because the algorithm is measuring completion, not volume.
Our guide to posting frequency and format mix has the per-format cadence numbers, and sends per reach goes deeper on the metric that decides whether a Reel leaves your follower base.
Frequently asked questions #
How does the Reels algorithm work in 2026? #
It predicts whether a specific viewer will watch a specific video and send it to someone. The heaviest inputs are watch time, likes per reach and sends per reach, applied in staged distribution: a small trial audience first, then expansion if completion and sends clear the bar.
Do hashtags help Reels reach? #
Barely. They function as topic labels; captions, on-screen text and spoken words carry more classification weight. Five relevant tags is the ceiling and the practical maximum.
Why did my Reel get 300 views? #
Almost always a failed hook: the trial audience left before the content started. Low completion suppresses stage-two distribution regardless of how good the rest of the video is.
Should I delete a Reel that flopped? #
No. Deleting removes the long-tail search traffic and resets nothing useful. Diagnose it, fix the hook in a re-cut, and publish the improved version in six to eight weeks.
Does posting time still matter? #
Less than it did. Reels are distributed over days, so timing mostly affects the first-hour sample. Our 2026 posting-time data covers the windows that still matter by audience type.
The algorithm is not a puzzle to be gamed. It is a prediction market on attention, and the inputs it can read in 2026 are close to the ones a human viewer uses: does this look worth three seconds, does it hold, and would I send it to someone. Make those three true and the distribution takes care of itself.