Reading the retention curve

The retention curve is a graph with time on the horizontal axis and percentage of viewers still watching on the vertical axis. A perfectly flat line at 100% would mean every viewer who started the video watched it to the end; that does not happen in practice. What you want is a line that starts high, drops gradually, and stays above a meaningful threshold by the end. What you do not want is a steep cliff early, because heavy early exits are exactly the kind of watch behavior that can work against wider distribution.

The shape of the curve is more useful than any single number. An average completion rate of 60% could mean a slow, steady decline from 100% to 20% across a 60-second video, which is actually a good result. Or it could mean a cliff to 60% at second 3 and then flatness, which means you lost nearly half your audience before the hook even landed, and the remaining viewers are unusually committed. Same number, completely different problem. Read the shape, not just the average.

Each cliff in the curve maps to a specific failure point in the video. A cliff at second 3 is a thumb-stopper or hook failure. A cliff at second 10-15 is a promise delivery failure. A cliff at roughly the midpoint is a pacing or content structure failure. Once you know which cliff you have, you know which part of the video to recut: not the whole thing, just the segment that precedes the drop. The Ad Bench maps your TikTok, Reels, and Shorts analytics against these thresholds automatically; see scoring rubric for how retention feeds into the overall ad score. These breakpoints explain attention loss; they are not fixed platform thresholds.

In most accounts, a large share of viewers, often close to half, leave in the first three seconds. (The exact figure varies by platform, audience, and metric definition; read it off your own analytics rather than an industry average.) That early exodus is not by itself evidence of a bad hook. It is the baseline exit rate before the hook has had a chance to land. These viewers left before they heard a word. The problem is the thumb-stopper: the first frame did not create enough interrupt to override the scroll reflex.

A weak hook line cannot fix a broken first frame. The viewer who exits at second 1.5 never processed the hook. They saw a frame that did not signal anything urgent, unfamiliar, or entertaining, and their thumb kept moving. The fix is upstream of the script: the first frame must contain a face, clear motion, or a text element that poses a question or makes a claim that requires resolution. None of those require a re-write; they require a re-shoot of the opening two seconds with the frame composition in mind. See thumbnail strategy for the mechanics of a thumb-stopper frame.

The hook line must start within the first second of audio. Not the first sentence, the first meaningful word. Many ads open with a beat of silence, a breath, or a title card before the creator starts speaking. Each one of those beats is a window for the viewer to swipe. Cut the lead-in entirely. The hook is the first thing out of your mouth.

The 10-15 second cliff is the “promise delivery” drop. The viewers who made it past the hook gave you their attention because the first three seconds implied a specific payoff. By second 10-15 they expect to see at least the beginning of that payoff. If the video is still in setup mode (still explaining context, still introducing the product, still building to the point), these viewers leave because the contract was broken.

The most common cause is buried lede. The creator knows the payoff is coming and is building tension toward it; the viewer does not know that and reads the delay as filler. Fix: deliver the core premise of the video by second 8 at the latest. Not the full explanation, the premise. “I made $4,000 in a weekend using this one mechanic” is a premise. The explanation of the mechanic is the body of the video. Lead with the premise, then explain it.

Most performance ads are structured as 30 seconds of setup before the point. That is a script problem, not an editing problem. No amount of cutting will fix a video where the value proposition appears at second 25. Rewrite the opening to front-load the conclusion, then use the body to justify it. This is the same “inverted pyramid” structure that governs good journalism: lead with the news, then provide the context.

A cliff in the middle third of the video, roughly 30-50% through, is the “boring middle” problem. The hook worked, the premise was delivered, and the viewer committed to watching. Then the video ran out of new information and the viewer felt the drag. This is a pacing and content structure issue, not a hook issue. The viewers who leave here were already engaged; they just hit a flat section with no new peaks.

The most effective fix is a re-hook at the midpoint. This is a line of dialogue or an on-screen text card that previews the next payoff: “and that's not even the best part,” or “here's where it gets strange,” or a specific numerical tease: “the second method made 3x more than the first.” The re-hook functions exactly like the opening hook: it creates a question that requires resolution, which forces the viewer to stay until the question is answered.

A visual cut to something new (a different location, a product in use rather than on a surface, a reaction shot) can extend engagement without requiring a script change. The eye responds to novelty independently of the narrative. A B-roll cut at the midpoint that introduces a new visual context buys a few extra seconds even when the audio is still in the same sequence. Use both: a re-hook line delivered on a new visual. For pacing norms per platform, see scoring rubric.

There is no universal completion percentage that guarantees wider distribution on every platform. TikTok has described finishing a longer video as a strong indicator of interest,¹ but the recommendation system reads it alongside watch time, rewatches, skips, shares, saves, comments, follows, and other context.

This means the structure of your ad needs to be designed around the retention curve, not just the hook. An ad that gets viewers to second 3 easily but bleeds them through the middle is losing the behavior that helps recommendation. Strong structures tend to follow a pattern: strong first frame, premise by second 8, one visual or content peak every 10-15 seconds, and an ending that delivers on the hook's promise with a specific outcome: a number, a result, a before-and-after. Vague endings are low-completion endings because the viewer senses the video is wrapping up without having paid off and they skip the credits.

Design the length of the ad around the content, not around a target duration. A shorter ad with a tight retention curve can outperform a longer ad that loses viewers, but completion is not the only outcome that matters. Compare the curve with saves, shares, comments, profile visits, clicks, and conversions before deciding what to cut. Cut every second that the curve would drop for, and keep every second that earns its place.

1. TikTok Help Center. "How TikTok recommends content." Accessed August 24, 2026. support.tiktok.com
2. The Ad Bench methodology, current scoring rubric. Reviewed August 24, 2026.