Since Adam Mosseri's remarks in early 2025, watch time has been consistently reported as the heaviest ranking input across Instagram's surfaces, alongside likes per reach and sends per reach. That much is widely agreed. The part that gets lost in the retelling is the unit.
Watch time is time. Seconds, accumulated across plays, including repeat plays of the same video by the same viewer. It is not percentage completion, and treating the two as interchangeable leads to the wrong edit.
Where the two disagree
A ten-second video watched to ninety per cent produces nine seconds. A forty-second video watched to half produces twenty. By completion rate the first is dramatically better. By the signal that is reported to matter, the second is more than twice as strong.
Completion rate is a real and useful diagnostic — it tells you whether the video held the people it got. It is just not the thing being counted.
Optimising completion rate has an obvious degenerate solution: make the video shorter. Optimising watch time does not.
What replays do to the arithmetic
If repeat plays accumulate, then a short video that invites a second viewing is competing on the same axis as a long video that holds attention once. This is the mechanism behind the seamless loop — not a trick, just the only way a very short piece can reach the totals that longer pieces reach by duration.
It also explains why loop-friendly content clusters in particular formats. A video whose ending rejoins its beginning without a visible seam gets a second view from viewers who did not consciously decide to rewatch.
What this changes in the edit
Three things, concretely.
- Stop cutting for completion rate. Trimming a video purely so a higher fraction of people reach the end improves a number nobody is counting.
- Protect the middle. Seconds accumulate where viewers stay, and the middle is where they leave — see where each format breaks.
- Treat the ending as a decision. Either it loops, or it resolves. An ending that trails off does neither, and it is the most common ending.
What it does not mean
It does not mean longer is better. That inference is the reason the length question keeps getting answered badly. Length adds available time and adds exit opportunities at the same time; the trade only goes one way if the material can carry it.
It also does not mean the other signals are decorative. Likes per reach and sends per reach are reported in the same breath, and sends in particular are described as the strongest lever for reaching people who do not follow you — consistent with Instagram's own statement that Reels is built around discovering content from accounts you do not follow.
A caveat on the sourcing
The January 2025 framing is reported widely and consistently, but almost entirely through secondary coverage rather than a durable primary post. The claim that sends carry a specific multiple of a like — the “three to five times” figure that circulates — traces back to no primary source we can find, and we do not repeat it. Instagram's own published ranking explanation names signal categories, not weights.