Reels are how Instagram distributes content to non-followers — most Reel reach comes from people who don't follow you, decided per post by how viewers respond. Understanding what the Reels ranking system rewards (and what disqualifies a Reel before ranking even starts) is the whole game — and it differs enough from TikTok that a TikTok reposted to Reels usually underperforms.
Specs, lengths, and features change fast — re-verify quarterly. The principles are durable.
Discovery-first, not follower-first
The vast majority of Instagram distribution is AI recommendation, not the follower graph — reach is earned per Reel, not borrowed from follower count. A small account can out-reach a big one on a single Reel. That also means every Reel has to earn its reach from scratch, and follower count is a vanity number next to reach, sends, and saves.
The signals that expand reach
Instagram has been unusually explicit about its ranking signals. In rough order of weight:
- Watch time / watch-through — the single most important signal. The first ~3 seconds decide whether a Reel gets pushed wider or throttled. Completion and rewatches are strong.
- Sends per reach (DM shares) — the dominant signal for reaching new people, weighted roughly 3–5× more than likes for discovery. A Reel people forward travels.
- Likes and comments per reach — still count, but matter more for existing followers than for discovery.
The practical combination that travels: a hook that holds (watch time) + an idea worth forwarding
(sends). Everything in hooks-and-retention.md exists to maximize those two.
The eligibility gates (before ranking happens)
A Reel must clear these to be recommended to non-followers at all:
- Original content — reposts get far less distribution; heavy reposting can exclude the whole account from recommendations.
- No other-platform watermarks — a TikTok or CapCut logo is a known, avoidable reach killer. Export clean; never repost the watermarked file.
- Has audio, under the length cap, safe/suitable for a broad audience.
- Clear niche — the system has to be able to categorize you to know who to show you to.
Made-for-Instagram, original, watermark-free is the price of entry.
Sound-off is the default viewing state
Most Reels viewers watch on mute (feed and DM contexts especially). So the script is written
muted-first: the on-screen text must carry the hook and the story on its own; the audio adds
energy and nuance for the minority who unmute. Burned-in captions lift retention — always on, kept
inside the safe zone (see instagram-reels-publishing).
Length tiers
- ~15–35s — the reach sweet spot: easiest to hold watch-through and loop.
- ~35–90s — earns its length for tutorials, storytelling, and transformation arcs; re-hook in the middle.
- Up to ~3 min — still eligible for non-follower recommendations (longer uploads exist via third-party tools but lose recommendation reach — verify current caps).
- Let the idea pick the length. Padding kills watch-through; dead air kills everything.
The Reel lives in three places — plan the frame
A Reel renders full 9:16 in the Reels tab, cropped in the feed, and cropped again on the
profile grid — so keep hook text, faces, and CTAs centered, and plan the cover frame at
script time. The cover is chosen at publish and can't be edited after; exact specs, safe zones,
and the cover brief live in instagram-reels-publishing.
Trial Reels — the free test
Instagram's Trial Reels show a Reel only to non-followers first. If it performs cold, roll it out to followers; if not, you've learned at zero cost to your follower-facing stats. Use trials for unproven hooks, formats, and topics — it's the cheapest experiment on the platform.
Cadence, measurement & disclosure
- Consistency over bursts — regular Reels give the system more tests; treat posting-time advice as general guidance, not analytics (WoopSocial has no analytics surface — judge via native Instagram Insights: watch time, sends-per-reach, saves, profile visits).
- AI-generated media — follow Meta's AI-labeling rules and the brand's policy; disclosure on
Instagram is native/manual (see
veo-prompt-pack.md). Never fabricate testimonials or stats.
