Instagram's head keeps pointing creators to the same short list of ranking signals and the publishing habits that fit them: the DM send is the growth lever, watch time gates video, and original first-party work is what the system now promotes.
2026-09-25 · by Moe Ameen
Instagram head Adam Mosseri has continued to spell out, across his ongoing creator guidance, the short list of signals that decide how far a post travels — and the publishing habits that follow from them. The through-line has not changed: watch time, likes per reach, and sends per reach are the signals that matter most across the app, a framing Mosseri first laid out in early 2025 and has reaffirmed through 2026. "Per reach" is the operative phrase — Instagram grades the rate at which the people who saw a post engaged with it, not the raw totals, which is exactly what lets a small account outrank a large one on merit.
Of the three, the DM send is the one Mosseri keeps returning to. Sends per reach is the strongest signal for reaching people who do not already follow you: when someone forwards your post into a friend's messages, they hand Instagram a high-confidence recommendation to a specific non-follower, which is the audience expansion the recommendation system is built to manufacture. Reported estimates put a send at several times the weight of a like for that unconnected reach, and Mosseri's own nuance is that likes count for slightly more on content shown to your existing followers, while sends count for slightly more on content shown to people who do not follow you. Watch time — total seconds watched, replays included, and how much of a video gets completed — remains the workhorse for video, with the first few seconds deciding whether distribution expands at all.
The publishing practices track the signals directly. The guidance points creators to make content specific enough that a viewer forwards it to one other person — the "send this to someone" moment — to post original, first-party work rather than reposts (Instagram's 2026 originality standard cut accounts that primarily repost and aggregate off from recommendations), and to write keyword-legible captions, since Search now reads the words in a caption far more than it counts hashtags. On the testing side, Trial Reels — which show a video to non-followers first, surface engagement metrics after roughly 24 hours, and can auto-share to your followers if the clip performs well within its first 72 hours — became schedulable in 2026, turning hook-testing into a batchable habit rather than a gamble on the follower feed. None of the individual points are new to people who study the platform; the notable part is that Instagram's own head keeps restating them as the current playbook. Treat the exact ranking weights as proprietary and evolving — what is confirmed is the direction.
Mosseri's list is really one instruction wearing several faces: make content a stranger will forward, watch to the end, and that only you could have made. That is a creative-and-volume problem before it is an algorithm problem — you cannot manufacture a send, but you can take far more specific, forward-worthy shots at earning one. [Kompozy](/) is the generation-and-publishing engine that produces those shots at the rate testing actually requires. It writes net-new, first-party content — [Persona Shorts](/glossary/persona-shorts) talking-head Reels in the human, selfie-cam register the originality standard rewards, brand-exact [Carousels](/glossary/hyperframes), photo posts, and [Text Posts](/glossary/output-buckets) — each one built around a single specific idea sharp enough to be sent to one person, not a generic post built to be admired.
The send-first framing changes what you generate, and that is the point of leaning on an engine: a surprising stat, a tip specific to one job, a take that only your point of view produces — generated as a batch of variants so you can test which hook earns the forward instead of betting the feed on a single guess. A [Persona Brief](/glossary/persona-brief) keeps the voice yours and filters the default AI register that gets scrolled past, and Gemini face-lock keeps the presenter identical across every clip so the account stays recognizable. Then [Autopilot](/glossary/autopilot) schedules and publishes the winners across the eight social platforms plus blog and email, behind a per-post review gate. Be clear on the boundary, because honesty converts: Kompozy does not send the DMs or fake the watch time — those come from the content being genuinely good and genuinely yours. What it removes is the production ceiling that stops most accounts from making enough original, send-worthy shots to find the ones that travel. For the ranking mechanics behind this, see [Instagram algorithm strategies for 2026](/guides/instagram-algorithm-strategies-2026).
Instagram head Adam Mosseri has consistently named three signals as the ones that matter most across the app: watch time, likes per reach, and sends per reach. "Per reach" matters because Instagram grades the rate at which viewers engaged, not raw counts. Of the three, sends per reach — forwarding a post via DM — is the strongest signal for reaching people who do not follow you, because it hands the system a high-confidence recommendation to a specific non-follower.
For reaching non-followers, yes. Mosseri has repeatedly called sends per reach one of the biggest signals Instagram uses, and it is widely reported as carrying several times the weight of a like for that unconnected reach. His own nuance is that likes count slightly more for content shown to your existing followers, while sends count slightly more for content shown to people who do not follow you yet — so if growth is the goal, design for the forward.
Trial Reels let you publish a video to non-followers first, without it appearing on your profile grid or in your followers' feed. After roughly 24 hours you can review engagement metrics, and Instagram can auto-share the Reel to your followers if it performs well within the first 72 hours. In 2026 Instagram made Trial Reels schedulable, so you can batch a set of hook tests in one sitting rather than posting them one at a time.
Build each post around one specific idea worth forwarding to one other person: a stat surprising enough to send to a colleague, a tip specific enough to send to a teammate, or a take that only your point of view produces. Vague, generic posts get admired at best; specific, useful, or genuinely funny ones get forwarded — and the forward is what buys reach to non-followers. Test several distinct angles rather than guessing which one lands.