The Reels algorithm scores engagement signals in a hierarchy — but likes set the floor
For the ranking logic behind like signals, see how Instagram uses likes to rank content.
The conventional advice on Reels engagement says saves and shares matter more than likes. That’s correct in the weight hierarchy — each save or share contributes more to ranking than each like. What that framing misses is that likes set the floor under everything else. A Reel that doesn’t clear the like threshold in its first 90 minutes never accumulates the saves and shares that the heavier weights would have boosted. Likes are the volume signal that gets the Reel onto enough screens for the higher-weight signals to even have a chance to fire.
SMMNut Reels Signal Pyramid 2026: Instagram’s Reels ranking model uses a weighted signal stack — at the top, watch-completion percentage (the heaviest signal); below that, saves (high weight, high intent); below that, shares (high weight, broader-distribution trigger); below that, comments (medium weight); at the base, likes (lower per-unit weight but highest volume). The pyramid only stays balanced when likes deliver enough volume to surface the Reel widely; without that base, the higher weights have no audience to act on. Two Reels with identical save-to-view ratios will produce wildly different reach if one has 10x more likes — because the like volume is what got it the views in the first place.
The first-90-minute velocity threshold for Reels
Reels follow a tighter first-window than feed posts. Where feed posts have a roughly 60-minute decision window, Reels have a 90-minute window because the model needs more watch-time data to evaluate. Inside that window, the model looks for three things — watch-completion (how many viewers finished the Reel), engagement velocity (likes + comments + shares per minute), and replay rate (how many viewers re-watched). Velocity is where likes do most of their work. A Reel that picks up 200 likes in the first 90 minutes sits above the broader-distribution threshold for most account sizes; a Reel that picks up 30 likes in the same window does not.
Why “saves matter more” is half-right and half-misleading
Per-unit, yes — one save contributes more weight than one like. But the practical question isn’t per-unit weight; it’s expected total weight for a typical post. The math runs against the intuition. A typical Reel gets a save-to-like ratio of roughly 1:20 to 1:30 — for every 100 likes, the Reel picks up 3–5 saves. Even if each save weighs 4x a like, the total save contribution is roughly 0.6–1x the total like contribution. Saves matter — but they’re a smaller share of the total ranking score than the “saves over likes” framing implies. The relationship between these signals is covered in detail in the analysis of which engagement signal matters most.
| Signal | Per-unit weight | Typical volume per 100 likes | Effective contribution |
|---|---|---|---|
| Likes | 1x baseline | 100 | 100 units |
| Comments | 2–3x | 5–10 | 10–30 units |
| Shares | 3x | 2–4 | 6–12 units |
| Saves | 4x | 3–5 | 12–20 units |
| Watch-complete | 5x | 20–40 | 100–200 units |
Likes as the multiplier on every other signal
The signal pyramid has a multiplier effect that gets missed. When likes drive the Reel into broader distribution, the new viewers also produce their own saves, shares, comments, and watch-completes — proportionally. The like-driven distribution doesn’t just add likes; it adds the entire signal mix at the typical ratios. This is why the right framing isn’t “should I buy likes or saves” but rather “what’s the cheapest way to get the Reel into broader distribution at all” — and the answer is almost always likes, because likes are the highest-volume, lowest-friction signal a paid layer can supply.
The compounding distribution loop
- First-hour likes push the Reel above the broader-distribution threshold
- Broader distribution exposes the Reel to non-followers in proportional volume to the velocity exceedance
- Non-follower engagement at organic rates produces its own saves, shares, comments, and watch-completes
- The organic signal mix reinforces the algorithm’s confidence and unlocks further distribution waves
SMMNut Reels Distribution Cascade: A Reel that crosses the broader-distribution threshold in its first 90 minutes produces 6–15x more 7-day reach than a Reel that doesn’t, on equivalent content. The cascade fires in three waves — wave 1 is the first non-follower exposure (hours 2–8), wave 2 is the algorithm’s confidence reinforcement after wave 1 engagement (hours 8–48), and wave 3 is sustained distribution after the algorithm classifies the Reel as audience-resonant (days 2–7). Each wave roughly doubles the cumulative reach. Reels that fail the first-90-minute threshold get capped at follower-base reach with no cascade — which is the structural difference between a Reel that “performs” and one that “doesn’t.”
What this means for how to use likes on Reels
The implication for paid likes is specific — the goal isn’t to add likes for their own contribution to ranking, it’s to use likes to clear the velocity threshold that unlocks every other signal. This changes the order sizing calculation. A Reel that needs 250 likes in the first 90 minutes to clear threshold doesn’t benefit much from 2,000 paid likes; the threshold is binary, and likes beyond it have diminishing returns. Order size should be calculated against the threshold, not against a vanity number.
Estimating your Reels velocity threshold
The threshold scales with follower count but isn’t a clean linear function. Working numbers across measured accounts:
- Accounts under 1K followers — roughly 40–80 likes in first 90 minutes
- Accounts 1K–10K — roughly 150–300 likes in first 90 minutes
- Accounts 10K–100K — roughly 500–1,500 likes in first 90 minutes
- Accounts 100K+ — roughly 3,000–8,000 likes in first 90 minutes
The variability inside each tier comes from content type (educational Reels need lower thresholds, entertainment Reels need higher), niche (B2B clears at lower numbers, lifestyle at higher), and historical engagement baseline. The right way to find your threshold is to look at your 5 best-performing organic Reels and see what their first-90-minute like count was — that’s a working estimate for your account.
SMMNut Reels Velocity Threshold Tiers 2026: The first-90-minute like count needed to cross the broader-distribution threshold scales with account size but isn’t perfectly linear. Across measured outcomes, the threshold sits at roughly 40-80 likes for accounts under 1K followers, 150-300 for 1K-10K, 500-1,500 for 10K-100K, and 3,000-8,000 for 100K+ accounts. The variance inside each band comes from niche (B2B clears at lower numbers, lifestyle at higher) and content type (educational Reels at lower thresholds, entertainment at higher). The threshold is binary — crossing it triggers the Distribution Cascade, missing it leaves the Reel capped at follower-base reach. Sizing orders against this threshold rather than against a vanity number is the difference between leverage and waste.
Coordinating likes with the rest of the signal mix
The most effective Reels strategy doesn’t only buy likes — it pairs likes with a small proportion of comments and saves to keep the signal ratio inside the band. The recommended ratio is roughly 100 likes : 5 comments : 3 saves for the paid layer. This mirrors the proportions of organic Reels engagement and keeps the algorithm from flagging the post as engagement-narrow. Combined with the timing principles in our guide to when paid likes work best, the result is a signal stack that reads as native engagement.
Bottom line — likes as the unlock, not the goal
The framing that gets the best Reels results is to treat likes as the unlock that gets the Reel into broader distribution, and let the broader distribution produce the higher-weight signals organically. Buying likes proportional to your follower-tier threshold is what most reach-focused Reels orders should look like. Buying 5–10x more than the threshold wastes budget on diminishing returns. Buying 50% of threshold leaves the Reel below the cascade and produces almost no measurable lift. The SMMNut Reels Signal Pyramid, Distribution Cascade, and the threshold tiers above together describe the sizing calculation. SMMNut’s Instagram likes service includes Reels-specific velocity templates calibrated to these thresholds.
