Adding a voiceover does not earn reach. Across 998 Reels of similar length, the more words per second, the less reach
More narration per second went with less reach across 998 Reels of matched length. The trend is clean and small. Measured 24 September 2026.
What the numbers say
A voiceover is a delivery choice, not a growth tactic, and the data is mildly against overdoing it. Looking only at Reels between 15 and 60 seconds, so that length cannot explain the result, 998 Reels sorted by how many characters of speech they carry per second fall into a clean order. The quietest quarter reached a median of 0.65 times their follower count. The next band reached 0.56, then 0.49, and the wall-to-wall talking band, which is most of them at 592 Reels, reached 0.42. Every step down in reach is a step up in words per second, with no reversals. The obvious objection is that quiet Reels are simply shorter, and that is exactly why this is measured inside a single length window.
If you are adding a voiceover because you think talking over the footage will lift reach, this corpus says it will not. Narrate when the words carry something the picture cannot, and stop when they do not.
Sample of 998, measured on 2026-09-24. Scrptly production database, queried 2026-09-24. 998 Reels deduplicated by shortcode, restricted to durations between 15 and 60 seconds, banded by transcript characters divided by duration.
The measurement
Only Reels between 15 and 60 seconds long, so this is not secretly a comparison of short against long. Speech density is transcript characters divided by duration, which is a rough proxy for how continuously someone is talking. Views per follower is the Reel's views divided by its own account's follower count. The four rows hold all 998: 116 plus 99 plus 191 plus 592. Read the median column, which falls in a straight line. The top tenth column does not, and is noisier because each band's top tenth rests on far fewer Reels.
| Characters of speech per second | Reels measured | Median views per follower | Top tenth reaches | Median length seconds |
|---|---|---|---|---|
| Under 3, near silent | 116 | 0.65 | 24.57 | 28.3 |
| 3 to 8, occasional lines | 99 | 0.56 | 6.47 | 23.0 |
| 8 to 14, steady narration | 191 | 0.49 | 10.61 | 31.5 |
| 14 or more, wall to wall | 592 | 0.42 | 6.52 | 42.4 |
One Reel, beat by beat
This is what a near silent Reel does instead of talking. One real Reel from the corpus, not a composite: 51,160 followers, seen 22,969,929 times, which is 448.98 times the following. It runs 58 seconds and its entire transcript is 14 characters long. Nothing is explained out loud. The structure does the work, and every beat below is a fresh small payoff rather than a fresh sentence.
| At | What happens |
|---|---|
| 0:00 to 0:03 | A person sets something up on the ground. No explanation. The setup itself is the question |
| 0:04 to 0:10 | The first small animal arrives and interacts. Quick payoff, early, before anyone can scroll |
| 0:10 to 0:30 | More animals cycle through. The variety keeps resetting attention rather than asking for one long hold |
| 0:30 to 0:50 | A bigger or more appealing animal arrives. This is the payoff the rest was building towards |
| 0:50 to 0:58 | It leaves naturally and the Reel ends. No call to action, no talking, no outro |
What to do next
- Ask what the voiceover is carrying. If it narrates what is already visible on screen, this corpus says you are adding words and losing a little reach for them.
- If you do narrate, the lighter bands did better on the median. In practice that is closer to one clear line per beat than to continuous talking.
- Keep the size of this in proportion. Quietest band to densest is 0.65 against 0.42, which is about 1.5 times. The measured quality of a Reel's opening separates 0.35 from 1.37 on the same corpus, which is about 3.9 times. Speech density is a real effect and a small one.
- Three honest limits. These are Reels from accounts somebody chose to analyse, so they skew towards content that already worked. The transcript covers spoken words only, so a silent Reel carrying heavy on-screen text reads as quiet here. And this is an association across Reels, not a test: nobody took one Reel and published it twice.
- If you want to know which of your own Reels carry their weight without narration, run your handle through Scrptly. It returns the transcript, the hook score and the beat order for each one. The first analysis does not need an account.