Instagram Reel view benchmarks by follower count

What the sample contains

Every Instagram post carrying a view count from the tracked accounts: 9,654 typed as Reels, plus 116 video posts from 15 accounts whose type field was not normalised on sync. Excluding them moves no band median by more than 0.03x, so they are kept and disclosed rather than dropped.

Multiples are computed per post and then taken as a median, not as a ratio of two medians. Accounts are public creator accounts tracked in Reelyze; posts with no view count are excluded, as are accounts with no follower count.

Who is in the sample

The obvious objection to any first-party benchmark is that it measures whichever accounts one product's users happen to follow. That is not what this is. 82% of these accounts are the creators' own, connected by people who signed up for a Reels analytics tool. So the population is creators actively measuring their own performance: a describable group, and not a random sample of Instagram. The remaining 18% are accounts those creators track. The 222 accounts belong to 188 distinct Reelyze accounts, so this is not one watchlist, and no single account dominates: the largest contributes 1.9% of posts and the 50 largest together contribute 52.8%.

Caption keyword analysis identifies 12 niches. The largest is travel at 19.8% of accounts, and the three largest together account for 41%. That is a heuristic rather than a hand-audit, so treat it as indicative, but the sample is not concentrated in one or two verticals.

The posting range overstates the spread. Posts run from 2019-11-23 to 2026-08-23, but 89% of them were published in 2025 or later. Read this as a current benchmark rather than a multi-year average, which is the more useful thing for it to be, but not what a seven-year date range implies on its own.

The obvious way this could be wrong, and what happened when we tested it

Follower counts are recorded as of the last sync, not as of each post. Accounts grow, so an older post gets divided by a larger number than it actually had, and an account is placed in a band by its size today. Both effects push the same way, and between them they could manufacture the whole falling-multiple pattern without any real effect underneath. It is the first thing worth checking and it is not a small objection.

If that were the cause, restricting to recent posts, where the recorded follower count is close to the true one, would flatten the pattern. It steepens. Across the full sample the median multiple runs 2.26x down to 0.33x. Across the last 90 days alone it runs 3.44x down to 0.22x. Removing the suspected distortion makes the effect larger, which is the one outcome the distortion cannot produce.

The bias is real, and it runs against the finding rather than for it. Posts from before 2025 in the under-1,000 band show 0.80x, against 3.44x for recent ones, because those accounts have since grown. Those posts are 10.7% of the sample and they pull the published figures down. The numbers above are the conservative version.

The headline is stable across every window: 38.7% of posts reach 1x across the full sample, 39.5% for 2025 onward, and 39.5% for the last 90 days. Only the pre-2025 tail differs, at 32%.

The two populations are not evenly spread, and it matters

Creators' own accounts and the accounts they track are not mixed evenly across the follower bands. The under-1,000 band is 100% own accounts; the 100,000+ band is only 38.1%, because people track large accounts and rarely track small ones. So the two ends of the pooled table above are not quite comparing the same population, which is the kind of composition shift that can create a trend out of nothing.

Tested by restricting to one population with current follower counts: own accounts, last 90 days. The pattern holds and is steeper than the pooled version, running 3.44x, 0.67x, 0.37x, 0.28x across the four bands. Over full history the own-accounts figure rises again in the top band, on 610 posts, which is the follower-timing effect above plus a thin sample rather than a contradiction.

The headline table stays pooled, because that is the version reproduced figure by figure and cited elsewhere on this site. If you are quoting the size effect rather than the benchmark itself, the single-population numbers in this paragraph are the ones to use.