Sends per reach: the Reels signal to design for
Mosseri named sends per reach as a Reels ranking signal ahead of likes. Six structures that trigger sends, and how to track the ratio without a benchmark.
Adam Mosseri has publicly named the three signals that matter most for Reels: watch time, sends per reach, and likes per reach. Two of those are outputs you can only influence indirectly. The third is the only one you can build a video's structure around, because a send is a decision with a specific shape — someone has to think of a particular person and then act.
That distinction is the useful part. You cannot write a video that makes people like it; you can write a video that makes someone think of their flatmate. Sends are a design problem with a solvable brief, and the ordering Mosseri described — sends carrying more weight than likes for reaching people who do not already follow you — makes it a structural constraint rather than a bonus.
What was actually said, and what wasn't
The trio is his, stated publicly, and the ordering for non-follower reach is his. What is not his — and what circulates constantly as though it were — is a multiplier. You will see "sends are worth 3–5x a like" repeated in strategy decks. I cannot source that to Instagram or to Mosseri, and I would not design a content calendar around a number nobody can attribute.
The practical difference matters. If you believe the multiplier, you start trading watch time for sends, cutting clips shorter to raise share rate, adding "send this to someone who needs it" to every caption. If you only believe the ordering, you do something narrower and safer: you stop treating sends as a vanity byproduct and start treating them as one of three things a Reel is supposed to produce, alongside — not instead of — the completion the first signal implies.
Why a send is a structural constraint
Every other engagement action is about the viewer's relationship to the content. A send is about the viewer's relationship to a third person. That means the video has to contain something transferable: a claim, a joke, an instruction, a judgement — something that survives being pulled out of the feed and dropped into a DM thread with no context.
This is why "great video, no sends" is such a common pattern. A well-made piece that is genuinely enjoyable can be completely untransferable, because enjoying something and having a reason to forward it are different states. The test to apply while scripting is blunt: what is the sentence the sender types along with it? If you cannot write that sentence, the Reel will not be sent, however good it is.
Notice that this also constrains length. A send has to be worth someone else's time, and the sender knows it. Long clips get sent less per view than short ones, not because sharing is effortful but because forwarding a 90-second video is a bigger ask of the recipient. That tension between watch time and sends is real, and pretending it does not exist is how accounts end up optimising one into the ground.
Six structures that produce sends
These are structures, not caption tricks. Each one puts the recipient inside the video.
| Structure | The mechanic | The sentence the sender types |
|---|---|---|
| Named-type callout | Describes a recognisable person type precisely enough to be identified | "this is you" |
| Settle-the-argument | Takes a clear side on something people actually disagree about | "told you" |
| Do-this-with-me | Requires a second person to execute | "want to try this?" |
| Compressed reference | A checklist, a spec, a set of numbers nobody will memorise | "saving this for you" |
| In-group joke | Only lands if you share a specific context | no sentence needed |
| Before-you-buy | A verdict on a decision someone is currently making | "read this before you order" |
Two notes from watching these run. The named-type callout is the highest-yield and the easiest to do badly — precision is everything, because "people who procrastinate" identifies nobody while "people who clean the kitchen at 11pm instead of sleeping" identifies someone specific. And the compressed reference is the one that most often wins on sends while losing on watch time, since viewers screenshot and leave. If watch time is your bottleneck, do not lead with that structure.
The one to avoid is the explicit ask. "Send this to a friend who needs it" appended to a video with no transferable content does not produce sends; it produces the appearance of trying. The ask works only when the video already gave someone a reason, at which point it is a nudge rather than a substitute.
Isolating the metric
Instagram's reporting vocabulary and Mosseri's vocabulary do not line up perfectly — what he describes as a send is the share-to-DM action, and what your Insights surfaces may be labelled differently. Do not spend time reconciling the two. Pick whichever share or send count your account exposes for a Reel, use the same field every week, and divide it by that post's reach.
Three rules for making the number readable:
- Always divide by reach, never report raw sends. Raw sends track distribution, so a post that reached five times as many people will show more sends while being less shareable. The ratio is the only version that compares across posts.
- Compare against your own median, not a benchmark. There is no credible published floor for sends per reach, and the ones you will find are vendor figures without inspectable samples. Take the median of your last 30 Reels and treat that as the line.
- Read it against watch time, not instead of it. A post that spikes on sends and collapses on completion is usually a screenshot-bait format. Worth knowing, occasionally worth doing on purpose, but not a template.
Logging is the part that fails. If you are pulling numbers by hand you will stop within three weeks, so make it a scheduled task rather than an intention — the agent can check social performance across connected accounts, which is enough to keep a weekly row going without opening the app.
Designing for sends without breaking the other two signals
The trap is treating the three signals as a portfolio to rebalance. They are not independent — a Reel with no watch time does not get the reach for sends to happen at all, so watch time is a gate rather than a competitor.
A workable order of operations:
Decide the transferable thing first. Before the script, write the sentence the sender types. If it does not exist, either add a structure from the table or accept that this post is a watch-time post and stop trying to make it both.
Put it somewhere retrievable. The transferable element needs to be findable without scrubbing. In practice that means it lands as a stated claim or an on-screen line, not buried in the middle of a voiceover. Timed text overlays are the mechanism; the discipline is that the overlay carries the claim rather than transcribing the audio.
Keep it legible muted. A Reel forwarded into a DM is very likely watched with sound off, possibly twice. Captions are not optional here — the caption track is what makes the transferable line survive the handoff. Adding captions handles the transcription, and keeping the style consistent means a forwarded clip still reads as yours.
Do not stretch it. If the transferable thing lands at 14 seconds, the Reel is 14 seconds. Padding to hit a length target costs completion and gains nothing on sends. Where the format genuinely needs more room, the cut rate Reels rewards sits a little slower than TikTok's, which buys you some density without reading as frantic.
The broader account-level version of this — what to publish, in what mix, against which pillar — is covered in the Reels growth playbook. This post is the narrower question of what a single Reel has to contain.
FAQ
Does raising sends actually raise reach, or is that assumed?
The ranking role is stated; the causal chain from your one post to a specific reach outcome is not something anyone outside Instagram can verify. What you can do is check the correlation on your own account: plot sends per reach against non-follower reach share across your last 30 Reels. If your account shows no relationship, believe your data over the general principle.
Should I ask for sends in the caption?
Only when the video earned it. A direct ask attached to genuinely transferable content converts some fence-sitters. Attached to content with nothing to transfer, it reads as a request for a favour, and repeated across every post it becomes background noise your regulars stop seeing.
Do carousels and static posts follow the same logic?
The transferability principle carries over, and reference-style carousels are historically strong on sends because they compress something worth keeping. The ranking signals Mosseri named were about Reels specifically, so do not assume the weighting transfers to other surfaces. Measure the ratio separately by format.
What if my sends are fine but my reach is not growing?
Then sends are probably not your constraint. Work backwards through the gate: if watch time is below your own median, the video is not earning the distribution that lets sends happen. Fix the opening and the completion first, and re-check the sends ratio once the post is actually being served.