The short answer (and the one most creators get wrong)
If you open TikTok, Instagram Reels, or YouTube Shorts today, the platforms themselves will tell you that longer videos are favoured. TikTok pushes creators toward 60-second uploads. Instagram weights Reels that hold attention for the full duration. YouTube Shorts is testing 3-minute uploads in 2026.
So most creators take the hint and upload longer.
Then they watch their retention graph collapse around the 8-second mark and wonder what went wrong.
The platforms are not lying. Longer videos can get pushed harder, but only when retention holds. A 60-second video that loses half its viewers at 12 seconds performs worse than a 20-second video that holds 80% to the end. The algorithm rewards completion and rewatches, not duration.
Here is what 1,400+ scripts generated on ScrollScript, combined with public platform retention benchmarks, actually tell us about optimal length per platform.
TikTok: 21โ34 seconds is the sweet spot
TikTok's own creator-fund data from 2024โ2025 showed videos between 21 and 34 seconds had the highest average completion rate. Not the longest. Not the shortest.
Under 15 seconds, the algorithm does not have enough signal to evaluate whether viewers actually liked the video or just failed to swipe in time. Completion rate stops being meaningful.
Over 45 seconds, you need a genuinely strong structure to hold attention. Most creators do not have the script writing background to sustain a 60-second video.
What our data shows: 68% of scripts generated for TikTok in the 30-second duration bucket. Creators who generate 60-second TikTok scripts return 22% less often than creators who generate 30-second ones. They either cannot execute the longer script or the video under-performs and they quietly stop using the tool.
Practical rule for TikTok length:
- 15 seconds: viral hooks, single punchlines, reaction-style content
- 30 seconds: the default for educational, storytelling, and most creator content
- 60 seconds: only if you have three or more distinct beats in the script and have tested that your delivery holds the viewer past the 15-second dropoff
Instagram Reels: 7โ15 seconds for reach, 30 seconds for saves
Reels is the most bimodal of the three platforms. There are two kinds of Reels that perform, and they have almost opposite length targets.
The reach Reel is 7 to 15 seconds. Quick visual payoff, one punchline, designed to be rewatched. Instagram's algorithm counts rewatches as strong signal, so a 9-second Reel watched twice outperforms an 18-second Reel watched once. This is why so many viral Reels feel suspiciously short.
The saves Reel is 25 to 45 seconds. Educational or list-based content where the viewer needs time to absorb information worth saving. Saves are Instagram's highest-weighted signal in 2026. A Reel with a 12% save rate will out-reach a Reel with a 40% like rate almost every time.
What our data shows: Reels scripts cluster at 30 seconds in our data, but the creators who publish both short and medium length Reels (mixing 15s reach content with 30s save content) see higher 30-day retention on the tool than creators who only publish one length.
Practical rule for Reels length:
- 15 seconds: the "scroll-stopper" format. Hook, one beat, payoff. Designed to be shared.
- 30 seconds: educational or list content designed to be saved
- 60โ90 seconds: rarely worth it unless the content is genuinely narrative (storytelling, confessional, behind-the-scenes)
YouTube Shorts: 30โ45 seconds is the new ceiling
YouTube Shorts behaves differently from the other two platforms because it is connected to your main YouTube channel. Your Shorts retention affects how your long-form videos get recommended. This changes the length calculus.
Shorts under 20 seconds rarely convert Short viewers into channel subscribers, because there is not enough time to demonstrate why someone should follow you.
Shorts over 60 seconds, while allowed, compete directly with the massive library of established YouTube long-form content. Viewers accustomed to 10-minute videos often drop off a 90-second Short because the value density feels low.
The ideal YouTube Shorts length in 2026 is 30 to 45 seconds. This is long enough to demonstrate expertise and pitch your channel, short enough to benefit from the Shorts feed's scroll behaviour, and hits the retention sweet spot where YouTube actively pushes the video beyond your subscriber base.
What our data shows: YouTube Shorts scripts in our 1,400+ dataset average 38 seconds of spoken content when accounting for the reading speed adjustment (native speakers read ~150 words per minute; non-native creators often deliver slower).
Practical rule for Shorts length:
- 15โ20 seconds: only for dance, meme, or single-joke content
- 30โ45 seconds: the default. Educational, commentary, tutorial, storytelling.
- 60 seconds: genuine teaching content with three clear subsections
Why "longer = better" is a trap
Here is the mechanic almost every creator under-weights: the algorithm does not measure how long your video is. It measures what percentage of your video people watch, and how many people finish it or rewatch it.
A 15-second video watched to completion by 90% of viewers is a better algorithmic signal than a 60-second video watched to completion by 35% of viewers. The longer video shows more total watch time, but lower completion rate signals lower quality to the ranking system.
The trap is that creators see a successful 90-second video from a large account and assume length is the reason it performed. It is not. The reason is that the creator knew how to keep a 90-second video interesting. Most creators cannot. They upload a 90-second version of a script that should have been 30 seconds, and the retention curve tells the story.
How duration changes script structure
This is the part almost no creator writes about, so it deserves space.
A 15-second script has time for one hook and one payoff. Structure:
- 0โ3s: Hook (promise + tension)
- 3โ12s: Body (deliver the promise in one beat)
- 12โ15s: Payoff or punchline
You cannot fit a meaningful CTA in 15 seconds. The CTA is the video itself functioning as a signature.
A 30-second script has room for a three-part body:
- 0โ3s: Hook
- 3โ8s: Context or reframe
- 8โ22s: Three distinct beats or points
- 22โ27s: Payoff or resolution
- 27โ30s: CTA (one short line)
A 60-second script needs a mid-video "retention reset" around the 25โ30 second mark, because that is where the second wave of dropoff happens:
- 0โ3s: Hook
- 3โ10s: Context
- 10โ25s: First major beat
- 25โ30s: Retention reset (a question, a new angle, a pattern break)
- 30โ50s: Second major beat + payoff
- 50โ60s: CTA
If you cannot identify where the retention reset goes in your 60-second script, the script is too long. Cut it to 30 seconds.
The length question no one asks
"How long should my video be?" is the wrong question. The right question is: how many distinct beats does my idea actually have?
A one-beat idea should be a 15-second video. A three-beat idea should be 30 seconds. A five-beat idea with a natural mid-point tension shift should be 60 seconds. Forcing a one-beat idea into 45 seconds is the fastest way to tank retention.
When you use ScrollScript to generate a script, the duration you select changes the structure, not just the word count. A 15-second script returns a different architecture than a 30-second one because the beats are different, not because we added filler.
Platform-by-platform cheat sheet
| Platform | Reach length | Save length | Maximum worth trying |
|---|---|---|---|
| TikTok | 21โ34s | 30s | 60s (with 3+ beats) |
| Reels | 7โ15s | 25โ45s | 60s (narrative only) |
| YouTube Shorts | N/A | 30โ45s | 60s |
The best creators test two lengths of the same idea, post both across platforms, and let retention data tell them which version of the idea actually had enough meat for the longer version. Most of the time, the shorter version wins.
Ready to generate scripts at the right length for your platform? ScrollScript generates 3 ready-to-film variants at 15, 30, 60, or 90 second durations. Each is structured for the platform you pick. Try it free โ