How fast do top YouTube Shorts actually talk?
Across 132 YouTube Shorts with at least a million views each, the median speaking rate is 193 words per minute and the median first sentence ends at 3.1 seconds. Both numbers are well past the advice that circulates in short-form guides — including the defaults this tool shipped with.
Every number here is measured from the audio, not estimated. Method and limitations are at the bottom, and they matter — one metric had to be thrown out entirely.
Speaking rate
The middle half of these clips runs 156–221 WPM. The 10th percentile is 118 and the 90th is 246. This is pace while actually speaking — words divided by the span from first word to last, so silence at the start or end does not deflate it.
For comparison, the rates usually quoted for narration are 140–160 WPM. The measured median is 193. A script written to a 155 WPM estimate and then delivered at 193 finishes roughly 20% early — which, on a 60-second slot, is about 12 seconds of unplanned airtime.
The hook
The common rule is a hook of seven to ten words landing inside three seconds. Measured, the median first sentence is 11 words over 3.1 seconds, and the middle half runs 1.9–4.8s.
Fewer than half of these clips obey the three-second rule. 48% of first sentences end by 3.0s (60 of 125), 56% by 3.5s (70), and 77% by 5.0s (96). The rule describes roughly half of what works, not all of it.
Two related numbers. By the 3-second mark a median of 9 words have been spoken — more than the 7–10 the rule implies, because the pace is faster than the rule assumes. And 20% of clips (27 of 132) contain a pause longer than 0.4 seconds inside the first three seconds, so a beat early on is not disqualifying.
By niche
The spread between the fastest and slowest niche median is about 39 WPM — real, but smaller than the spread within any single niche. Pace is more a property of the individual creator than of the subject.
Table: all niches, including those too small to chart
| Niche | n | Median WPM | Median hook (s) |
|---|---|---|---|
| business (thin) | 5 | 234 | 1.64 |
| cooking | 23 | 192 | 2.56 |
| education | 18 | 180 | 4.8 |
| fitness | 18 | 219 | 3.24 |
| health (thin) | 12 | 187 | 4.1 |
| home-diy (thin) | 7 | 162 | 3.88 |
| parenting (thin) | 11 | 138 | 1.89 |
| personal-finance (thin) | 6 | 193 | 2.75 |
| storytime | 15 | 202 | 2.86 |
| tech | 17 | 187 | 3.68 |
Niches marked "thin" have fewer than 15 clips and are not charted; they remain in the overall figures.
Does pace predict views? No.
Within this sample there is no usable relationship between how fast someone talks and how many views the clip has. Spearman rank correlations against view count:
| Measure | rho | p | Verdict |
|---|---|---|---|
| First sentence length (s) | -0.162 | 0.0716 | No effect |
| Speaking rate (WPM) | -0.249 | 0.004 | Weak, and confounded — see below |
| Words spoken by 3s | -0.124 | 0.1571 | No effect |
The speaking-rate correlation is statistically significant and we are still not going to claim it. The reason is composition: the 10M+ group holds 74 of the 132 clips and is dominated by cooking, education, fitness and tech creators, while the 1–3M group has a different mix entirely. The correlation tracks which creators landed in which bucket, not a relationship between pace and reach. A sample built to answer that question would need to hold the creator constant; this one was not.
The honest summary: pace and hook length do not separate a 1-million-view Short from a 10-million-view one. Both are table stakes, not levers.
What this means if you are writing one
- Write to roughly 193 WPM, not 155. If you time a script at 155 and deliver it at your natural short-form pace, you will come in early and have to pad.
- Don't panic at three seconds. Half of these clips take longer than that to finish their first sentence, and anything up to about five seconds is inside normal practice. Write the sentence the hook needs, then cut words — not tenths.
- The 7–10 word hook rule holds up. A median of 9 words land before the 3-second mark, right inside it. What is wrong is the pace assumed underneath it: those 9 words arrive at roughly 193 WPM, not 155, so the sentence you fit in three seconds is longer than the old arithmetic suggests.
- A pause in the opening is fine. One clip in five has one.
- Do not tune pace hoping for reach. Nothing here supports that.
Time your script against these numbers →
Methodology
Ten niches. Channels were reached three ways: YouTube search; a hand-picked list of talking-head and voiceover creators; and a speech-oriented search top-up for three thin niches. Within each channel, candidates are the highest-viewed among that channel's 80 most recent Shorts, capped at five per channel, every one with at least 1,000,000 views as displayed on the channel page at discovery time (2026-09-03).
Discovery surfaced 312 candidates. Of those, 104 were removed editorially before any download, covering 32 channels: kids' entertainment, ASMR and silent clips, music and fancam channels, product advertisements, television news and newspaper channels, conference keynotes, podcast interview excerpts, non-English channels, and creators filed under the wrong niche. The reasons are recorded per channel. That left 208 to attempt.
The first 30 seconds of each was downloaded as 16 kHz mono audio and transcribed with
Whisper medium using word-level timestamps. Download outcomes: 152
fetched, 51 rejected for running longer than 60 seconds, 5 no longer available.
Filters applied to the 152 transcribed clips, each with its count:
| Filter | Removed |
|---|---|
| No speech detected at all | 2 |
| Not English | 5 |
| Language confidence below 0.8 | 11 |
| Fewer than 10 words | 2 |
| Longer than 60s | 0 |
| Analysed | 132 |
Hook figures use the 125 clips where Whisper found a sentence boundary in the first 30 seconds; in 7 clips it found none, and those are excluded from hook statistics rather than counted as one long sentence. Clips span uploads from 2022-07-28 to 2026-08-28, drawn from 42 channels, no more than five from any one.
Limitations, including one that cost us a metric
- Lead-in time is not reported, because it could not be measured. Whisper placed the first word at exactly 0.00 seconds in 128 of 132 clips. That is the transcriber snapping to the segment start, not 132 creators beginning to speak on the exact frame. Any "median silence before the first word" we could print would be an artefact, so there isn't one.
- Newest-first sampling. A channel's Shorts tab is ordered by recency, not views. Reading 80 entries therefore covers a few weeks for a daily poster and several years for an occasional one, which biases the sample toward low-frequency creators and, for prolific channels, toward recent rather than best work.
- Sentence boundaries are Whisper's. Punctuation defines where the first sentence ends. In about one hook measurement in eight the model punctuated sparsely, producing first "sentences" longer than 10 seconds. The median absorbs this; the upper quartile is softer than it looks.
- The 60-second cut discards a quarter of the pool. Shorts now run to three minutes, and 51 qualifying clips were dropped for length. These findings describe Shorts of a minute or less.
- Two niches are too thin to stand alone — business (n=5) and personal finance (n=6). Clips over a million views are genuinely scarce there: one creator had a single such Short in his 400 most recent, another's 80 most recent topped out near 51,000 views, and nine more returned none. What does clear a million in those niches is mostly podcast clips, advertisements and news, all of which were excluded. They stay in the overall figures and are not charted.
- YouTube only, English only, first 30 seconds only, and view counts as displayed at discovery time.
What is not published
Aggregates only. No transcripts, no word timings, no per-video rows, and no list of the videos or creators measured. The audio was fetched to a local machine, analysed, and deleted. Publishing a ranked list of other people's videos was never the point — the numbers are, and the numbers are all here.