Glossary
Retention Rate: What It Measures
Retention rate is how much of the video the average viewer watches — what the curve shows, what a drop-off means, and what to read next.
At a glance
What is retention rate, and what should you measure it against?
Retention rate is the share of viewers still watching past a specific timestamp — and the timestamp is the whole question, which is why a bare percentage means nothing. Two matter most in short-form: three seconds, the gating threshold that decides whether the platform tests distribution further, and a later mark that tells you whether the body holds. The term gets misused when teams collapse it with completion rate — the share who reached the very end — or with average view duration, because those answer different questions about the same clip. All three are surfaced natively in TikTok Studio, Instagram Insights and YouTube Studio for Shorts, so the discipline is to read them side by side rather than substituting one for another.
- A retention figure is meaningless without its timestamp — three seconds and the body mark answer different questions.
- Three seconds is the gating threshold that decides whether distribution is tested further.
- Do not collapse retention with completion rate: one is a share past a mark, the other is reaching the end.
- Nor with average view duration — an average hides the shape that retention shows.
- All three are native in TikTok Studio, Instagram Insights and YouTube Studio, so read them together.
By Bell Chen, founder of Superdirector — how the analysis works.
A Curve, Not A Score
Retention rate is the percentage of viewers still watching past a given timestamp. Two checkpoints matter most: three seconds, the gate for wider distribution, and the completion point. Completion and average view duration are points on the curve; only the curve shows which beat broke.
What retention rate actually means
The strict definition: retention rate is the percentage of viewers who continue watching past a specific timestamp in the video. The two timestamps that matter most in short-form are three seconds (the gating threshold that decides whether the platform even tests distribution further) and the completion point (whether the viewer made it to the end). Both metrics are surfaced natively in TikTok Studio, Instagram Insights, and YouTube Studio for Shorts. Where the term gets misused is when teams collapse it with completion rate (the percentage of viewers who reached the very end) or average view duration (total seconds watched divided by viewers). Retention rate is the curve. Completion rate and average view duration are two points on that curve.
Jenny Hoyos, who has shipped more than a dozen YouTube Shorts past 100 million views per video and was profiled at length in Marketing Examined's short-form playbook (marketingexamined.com), gave the operational logic of why the curve matters more than the endpoints. Hoyos said, "It's not about the first three seconds. It's about retention all the way through. If your retention is steady, you'll go viral. If it dips, you won't." The curve tells you which beat broke. The endpoint only tells you that something broke.
The numbers that matter
The most credible cross-platform benchmark set as of mid-2026 comes from three sources. Socialinsider's social media video statistics study (socialinsider.io), built on 111K videos published between January and June 2026, puts average Reels engagement at 0.80 percent for 1K-to-5K-follower accounts down to 0.40 percent at 100K-to-1M, computed against followers — Socialinsider's own read is that small accounts look better on engagement only because their reach is still small, while average views jump more than sixfold once an account crosses into six-figure followers. Its Instagram Reels statistics study (socialinsider.io) tracks skip rates of 60.5 to 65.5 percent within the first three seconds. No benchmark study publishes completion-rate medians or a population watch-time median for Reels, so the size-based completion pattern quoted in older secondhand write-ups has no verified source. The third anchor is platform-side: Buffer's TikTok algorithm guide (buffer.com) lists the engagement signals — likes, shares, saves, comments and, 'perhaps most importantly,' video watch time — without publishing a negative-signal ranking.
House retention floors by platform (house rules-of-thumb, not published study figures): TikTok, clearing the three-second mark is the working floor for the algorithm to widen distribution. Total watch time matters more than completion rate. Instagram Reels, the house floor is 50 to 60 percent retention at the three-second mark (a house floor derived from Socialinsider's published skip-rate band, which puts 60.5 to 65.5 percent of viewers gone within the first three seconds, https://www.socialinsider.io/blog/instagram-reels-statistics/). Sends per reach is one of the top Reels metrics Mosseri recommends reviewing, with sends slightly more important for unconnected audiences. YouTube Shorts, average view duration is the headline metric. The house working floor for Shorts is 75 percent of video length on average view duration (a house rule-of-thumb, not a published study figure); Hoyos's public breakdowns are what put average view duration on the dashboard, not the source of the 75 percent number.
Those are floors, not targets. The benchmark that matters is the creator's own median for the same format, audience, and topic. A 65 percent three-second retention on a tutorial may be worse than a 55 percent three-second retention on a debate clip, because the tutorial format the same creator runs typically holds 70 percent.
How real creators apply it
Paddy Galloway is widely cited as the most analytically rigorous YouTube growth consultant in the field, with public clients including MrBeast, Logan Paul, and Mike Tyson. In his interview with Colin and Samir (youtube.com), Galloway said his workflow opens with the retention graph before anything else: "I'm not looking at views. I'm looking at the retention curve and finding the dips, because those tell me where the script broke," per Galloway. Galloway treats the retention dip as a script problem, not a thumbnail problem. The thumbnail decides the click. The retention curve decides whether the script delivered what the thumbnail promised.
MrBeast has stated multiple times, including on the Lex Fridman podcast Episode 351 in January 2023 (lexfridman.com), that average view duration is the metric his team optimizes around. MrBeast said in the same conversation that he and his team review the retention graph of every published video within 24 hours of release and use the dips to identify which scene types are losing audience, per MrBeast. The discipline is iterative: every video's retention curve is an input to the editorial decisions on the next video.
Jenny Hoyos has been more specific on the cognitive technique behind retention. In her vidIQ profile of her 10M-views-per-Short workflow (vidiq.com), Hoyos described what she calls fast retention: editing the script so that no beat lasts longer than four seconds before a new visual, claim, or sound enters the frame. Hoyos said in the same writeup, "Every word has to earn its place. If a sentence isn't moving the story forward or adding value, it's gone." The retention graph on her Shorts is conspicuously flat, where most creators see a sloping decline.
How to diagnose it on your own content
The four-step audit I run on accounts I advise: First, pull the retention curve for the last ten posts in the same format on the same platform. Native analytics is enough; no third-party tooling required for this step. Second, identify the median three-second retention and median completion across those ten. That is the working baseline. Third, sort the ten clips by three-second retention and look at the top two and bottom two. What does the hook on the top two share that the bottom two miss? Write the answer in one sentence per cluster.
Fourth, find the most common dip point across the ten retention curves. In my experience auditing roughly thirty short-form accounts since late 2025, the dip rarely falls where the creator first guesses. If the dip is at second three, the hook is the fix. If the dip is at second eight, the bridge from hook to payoff is the fix. If the dip is at second twenty-two on a 30-second clip, the payoff is the fix. Each layer is a different shot list for the next ten posts.
The mistake to avoid in this audit is treating the curve as a verdict. The curve is a diagnostic. A weak hook can be fixed in fifteen minutes on the next shoot. A weak payoff can take three iterations to fix because the script structure has to change.
Common mistakes
The most common retention misreading is using a universal benchmark instead of the account's own median. No benchmark study publishes a completion-rate floor for Reels, so any population completion number quoted in a meeting is invented; the honest population anchor is Socialinsider's published skip-rate band (60.5 to 65.5 percent of viewers gone within three seconds), and even that is not a target. A creator who shipped a Reel at 55 percent completion may be above the account's own median, but if the same creator's last ten Reels averaged 68 percent completion, the new post is a 13-point regression and the team should be auditing what broke.
The second mistake is treating retention as a single number rather than a curve. A 60 percent average completion rate hides whether the drop is at second three, at second nine, or at second twenty-two. Each of those drops points to a different fix. Pull the curve, not the endpoint.
The third mistake is optimizing for retention at the cost of payoff. As an editorial test, pair the retention curve with sends per reach to see whether viewers both stayed and passed the clip on. Mosseri's January 2025 caption names watch time, likes and sends as top signals; it does not establish a rule that high retention with low sends is engagement bait or specify the weight of that combination.
Where Superdirector fits
For competitive-set diagnosis, the brand-profile analysis I built in Superdirector pulls the retention pattern across an account's last 30 clips and an adjacent creator's last 30; useful as one diagnostic input among several. The script decisions that change the curve sit upstream of any dashboard.
Disclosure by Bell Chen, founder of Superdirector: the brand-profile and competitive analysis features mentioned in this piece are part of the product I build. Methodology and benchmarks here are sourced from the linked platform documentation, industry reports, and named-creator interviews; treat the tooling note as one input among several.
More tips and examples
Go deeper on the parts that matter to your next project.
One Endpoint, Two Fixes
A curve that collapses in three seconds and one that declines evenly can both finish at 50 percent completion, and they need opposite fixes: a new hook versus more mid-video beats. Hoyos is direct about it: "If your retention is steady, you'll go viral. If it dips, you won't."
Your Median Beats The Panel
Industry panels pool accounts from every size band into one number, so a panel median is population context, not a target. The comparison that changes your next Reel is internal: a 55 percent finish is a regression for a creator whose last ten Reels averaged 68 percent, whatever the panel reports.
Let The Dip Assign The Fix
Tighten the first three seconds for the gating drop, add a visual or claim every three to four seconds for mid-curve decline, and build a payoff or loop for the ending. Paddy Galloway opens at the dips because, per his Colin and Samir interview, they mark where the script broke.