← Lab notebook

Instagram/Facebook Reels Retention Benchmarks

There is no single official Instagram or Facebook Reels retention benchmark in the Meta sources reviewed for this guide. If you build software and publish Reels, mixing seconds, percentages, denominators, windows, or paid scope can make an invalid comparison look decisive. Meta defines watch-time metrics and supplies retention tools, but it does not publish one platform-wide percentage or number of seconds that every Reel should beat.

Use external averages as context, not targets. Compare each Reel with a like-for-like account baseline: the same platform, reporting window, paid or organic scope, duration band, and metric denominator. In the supplied three-post case, paid distribution dominates Instagram, Facebook’s largest displayed drops happen within two seconds, and mismatched Instagram watch-time fields prevent a clean performance verdict.

This guide separates published facts, screenshot observations, arithmetic, and interpretation so you can decide what to test next without inventing a benchmark.

No universal official Reels retention benchmark appears in the reviewed Meta sources

The official sources explain how to read metrics, not what score counts as universally good. In April 2023, Meta defined Instagram total watch time and average watch time. Total watch time includes replay time. Average watch time equals total watch time divided by total plays. In November 2023, Meta said Reels Plays include Initial Plays and Replays and announced a moment-by-moment retention chart.

The Facebook announcement is similar. Meta described Facebook retention graphs that show how long viewers watched, alongside reach, interactions, follower attribution, and distribution health. None of those reviewed announcements provides one Instagram- or Facebook-wide average.

That absence matters because a universal target would collapse different Reel lengths, audiences, distribution modes, account sizes, and denominators. A 6-second loop and a 60-second explanation can have the same average watch time while producing very different viewing behavior.

Three-step benchmark ladder separates official Meta metric definitions, Fanpage Karma’s historical 8.9-second Instagram and 10.9-second Facebook sample means, and a comparable account baseline matched by platform, window, distribution scope, and duration.
The third-party seconds provide context, but the decision benchmark comes from a like-for-like account set after the scope gate. On a narrow screen, scroll horizontally only if needed.

The transparent third-party average is context, not a target

Fanpage Karma offers a useful historical reference because it publishes its sample and period. Its 2024 study covered 419,393 Instagram Reels from 1,088 accounts and 246,366 Facebook Reels from 998 accounts, posted from August 2023 through July 2024. It reported average viewing times of 8.9 seconds on Instagram and 10.9 seconds on Facebook.

Those values still are not universal retention benchmarks. Fanpage Karma’s study does not give a Reel-duration distribution or a length-normalized percentage in the selected methodology. It also does not supply enough account-selection, paid-scope, or uncertainty detail to turn either mean into an AI Maker Lab target.

A second benchmark source demonstrates the scope problem from another angle. Dash Social’s 2026 Instagram Reels benchmark uses 3.3K+ global brands from July through December 2025. It covers handles with at least 1,000 followers and organic, boosted, and promoted content while excluding ads. It reports industry engagement rates from 0.1% to 0.5%, calculated as engagement divided by views. That is an engagement benchmark, not retention. Its no-ads scope does not match the ad-dominated case below.

Reels retention metrics answer different questions

Read each metric as an answer to one question. Do not substitute a percentage for seconds, a curve for completion, or a view count for an action.

Five distinct Reels measurement cards map average watch time to seconds, average watched to a duration-normalized percentage, the retention curve to drop and replay points, completion to end-reaching starts, and actions to saves, shares, interactions, and follows.
Seconds watched, percentage watched, curve shape, completion, and actions are related measures—not substitutes for one another. On a narrow screen, scroll horizontally only if needed.

On a narrow screen, scroll the metric table horizontally.

MetricUnit or formulaQuestion it answersCommon mistake
Total watch timeSeconds or minutes, including replay time under Meta’s 2023 Instagram definitionHow much playback time accumulated?Comparing totals across different play counts
Average watch time (AWT)total watch time ÷ total plays under Meta’s published Instagram definitionHow many seconds did a play consume on average?Treating seconds as a completion percentage
Average watchedAWT ÷ Reel duration × 100 as a derived normalizationWhat share of the Reel duration did the mean play cover?Calling the result a completion rate
Retention curveViewer share plotted against elapsed timeWhere did viewing fall, hold, or replay?Reducing the curve to one average
Completion ratecompleted eligible plays ÷ eligible starts × 100, using the export’s stated denominatorWhat share of defined starts reached the end?Inventing completions from average watched
Three-second View RatePercentage continuing after three seconds in the reported Instagram featureDid the opening retain the metric’s defined audience past three seconds?Treating three-second views as a rate without starts
ViewsPlays or views under the export’s stated definitionHow much playback occurred?Treating a replay-inclusive view as a unique person
Interactions, saves, shares, followsCounts or rates with an explicit denominatorWhat action occurred after exposure?Treating engagement as buyer intent or demand

The average-watched formula is useful for comparing different durations, but it is an operational normalization—not an official universal target. Because Meta’s published total watch time includes replay time, a replay-heavy Reel can also complicate a simple duration-normalized reading.

Completion stays separate. If a platform export does not provide completed plays and its denominator, do not manufacture completion from AWT or average watched. Likewise, a count of three-second views is not a three-second rate until you also know the eligible starts.

TechCrunch reported in January 2025 that Instagram’s View Rate shows the percentage of followers and non-followers continuing after three seconds. It also described Views Over Time, which compares accumulated views with an account’s usual views over the same period. That report is dated and secondary, so verify the current in-product label before building an automated scorecard around it.

The three-post case shows paid scope and an opening question

The supplied screenshots cover three posts: Game grammar, Platformer, and Translation hooks. Their strongest signal is not a universal creative winner. It is a measurement warning: Instagram distribution is mostly classified as ads, Facebook shows early drop points, and several fields lack compatible denominators.

Three-post case map shows about 94 percent paid classified Instagram reach, about 84 percent paid classified views, 9 interactions, 2 saves, no shares or follows, and Facebook rows whose largest displayed drop occurs at one or two seconds; a warning says the Instagram watch-time arithmetic does not reproduce reported average watch time.
The screenshots point to paid distribution and an opening-retention question, while their mismatched scopes prevent a platform winner or bug claim. On a narrow screen, scroll horizontally only if needed.

Instagram is primarily a paid-distribution case

On a narrow screen, scroll the Instagram comparison table horizontally.

PostTop-line viewsReach: organic + adsSource-classified views: organic + adsInteractionsTotal watch timeReported AWTFollows
Game grammar75750 + 1,50562 + 69249m21s8s0
Platformer425121 + 884138 + 28737m24s3s0
Translation hooks12712 + 37414 + 11322m6s8s0

Across the classified Instagram reach values, ads account for 2,763 ÷ 2,946, or about 94%. Across source-classified views, ads account for 1,092 ÷ 1,306, or about 84%. These are shares of the supplied classified values, not claims about the whole account or Instagram generally.

The source-classified views total 1,306, while top-line views total 1,309. Preserving that three-view difference is safer than silently forcing the fields to match. The screenshots also show 9 interactions, 2 saves, 0 shares, and 0 follows across the three posts.

Those action counts do not establish audience quality, buyer intent, product demand, or conversion. They also do not prove the posts failed. Without aligned reporting windows, impression or reach denominators, objectives, and a comparable baseline, they are bounded observations.

The Instagram AWT fields need a scope warning

The displayed total-watch-time and AWT fields do not reconcile under Meta’s published formula when top-line views are used as plays.

On a narrow screen, scroll the watch-time reconciliation table horizontally.

PostDisplayed total watch time in secondsTop-line viewswatch seconds ÷ viewsDisplayed AWT
Game grammar5617570.74s8s
Platformer4444251.04s3s
Translation hooks1261270.99s8s

The calculation exposes non-comparability; it does not prove a Meta bug. Views may not equal the plays used by the AWT formula. The screenshots may also represent different windows, audience slices, attribution scopes, or product surfaces. The correct next step is to export the native fields with their date ranges and definitions, then reconcile the scopes before judging performance.

Facebook points to an early-drop test, not a winner

On a narrow screen, scroll the Facebook comparison table horizontally.

PostAWTAverage watchedThree-second viewsBiggest drop
Game grammar3s8%20:02
Platformer6s13%20:02
Translation hooks3s5%40:01

Platformer has the highest displayed Facebook AWT and average watched inside this three-post set: 6 seconds and 13%. That makes it the leading descriptive row for those two fields, not a proven winning creative. Durations, starts, paid scope, reporting windows, and randomized assignment are missing.

Translation hooks has the most displayed three-second views, 4, but the lowest average watched, 5%, and the earliest biggest drop, 0:01. Without a start denominator, four three-second views cannot be interpreted as a better three-second rate.

Every displayed biggest drop occurs within the first two seconds. That makes the opening a reasonable next test area. The screenshots do not show the drop magnitude or curve denominator, so they cannot tell us how many viewers left or which exact edit caused it.

Build a comparable account scorecard

A practical benchmark starts with a frozen comparison contract. Export every Reel in a fixed analysis window, then keep only rows that match the comparison you intend to make. This guide’s analysis rule is to report the sample size and never treat a small set’s median as stable.

Use one row per Reel with these fields:

Scorecard fieldRecord exactly
IdentityPost URL or id, final asset hash/version, caption version
ScopePlatform, account, organic/boosted/promoted/ads, follower/non-follower slice
WindowPublish time, observation cutoff, attribution/reporting date range
CreativeTopic, duration, hook family, frame-zero subject, CTA, audio treatment
DistributionViews/plays, reach, impressions when available, paid spend and targeting when applicable
RetentionTotal watch time, AWT, average watched, three-second starts/rate, completion when supplied, first major curve drop
ActionsInteractions, likes, comments, saves, shares, follows, profile visits, link actions when intentionally measured
LimitsMissing denominators, field mismatch, UI definition, anomalous delivery, unresolved data scope

Then calculate account baselines only inside comparable groups. For this guide, record the median as a consistent summary convention and label the sample size; do not claim robustness or stability for a small set. Keep AWT and average watched together: seconds preserve absolute attention, while the percentage controls for duration. Inspect the retention curve beside both because one mean cannot show whether the loss occurred at frame zero, mid-explanation, or after the payoff.

Six-step Reels measurement loop exports native fields, freezes platform and distribution scope, groups comparable posts, calculates a scorecard, inspects the retention curve, and tests one creative variable before repeating.
A useful account baseline is a repeatable comparison protocol, not a borrowed universal threshold. On a narrow screen, scroll horizontally only if needed.

Test one creative variable at a time

The next experiment should be narrow enough to interpret. For these screenshots, the observed Facebook drop points make an opening test more defensible than a total rewrite.

Write the experiment before publishing either version:

  1. Question: Does a concrete frame-zero problem state hold more viewers through the opening than the current opening?
  2. Primary metric: Choose one available opening measure, such as a reported three-second View Rate or retention at a fixed timestamp.
  3. Guardrails: Keep AWT, average watched, and downstream actions visible so a stronger opening does not hide a weaker payoff.
  4. Changed variable: Change only the opening claim, first frame, or opening evidence order.
  5. Held conditions: Keep platform, duration band, audience/distribution plan, observation window, and CTA as comparable as practical.
  6. Identity: Freeze the exact video, caption, cover, and measurement cutoff used for each row.
  7. Interpretation: Call a sequential platform-distributed comparison quasi-experimental unless the platform randomizes assignment for the tested surface.

A single comparison should update the next hypothesis, not become a universal rule. Repeat the same scorecard across comparable Reels before promoting an account observation into an operating target.

What this analysis establishes

The research establishes four bounded conclusions:

  • The reviewed Meta sources do not publish one universal Reels retention average.
  • The Fanpage Karma seconds are historical sample means rather than targets.
  • The supplied Instagram case is paid-dominated and arithmetically scope-mismatched.
  • The Facebook screenshots justify testing the opening without naming a causal winner.

The analysis does not establish buyer intent, demand, conversion, an algorithm threshold, or a platform defect. It also cannot compare Instagram and Facebook directly until duration, reporting window, distribution scope, starts, and denominators align.

For the next controlled test, adapt one evidence-bounded opening with our Instagram Reel hooks guide and record it with the scorecard above.

Sources

Keep reading