YouTube Script Statistics 2026: Word Counts, Speaking Rate, Hooks and Flops From 1,290 Videos
Across 1,290 videos from 132 channels in two Prepublish studies, delivery speed did not separate hits from flops. Speaking pace landed at a median 181 words per minute for larger-channel hits against 180 for flops, and time to the first concrete reason to watch landed on word 14 against word 15. What separated the groups on larger channels was what the opening says: hits engaged the title's promise in 87 percent of openings against 78 percent for flops, and giving no reason to keep watching was 60 percent more common among flops. The median flop did 0.28 times its own channel's normal views. On channels between 1,000 and 20,000 subscribers, the preregistered replication found no separation on any opening pattern it tested, so the null results sit on this page next to the positive ones. Each table below names its source, and every figure can be checked against the public CSV rows. These are view-performance findings, not retention findings, and the analyzer output on this site is a relative script-hold index over a draft, never a claim about published viewers.
Last verified 2026-09-21. Next refresh: the first business day of each month.
Speaking rate percentiles and group medians
| sample | videos | p25 (wpm) | median (wpm) | p75 (wpm) |
|---|---|---|---|---|
| All videos with a valid rate | 349 total | 160 | 181 | 201 |
| Education channels, at least 20 videos each | 89 total | 153 | 179 | 204 |
| Finance channels, at least 20 videos each | 40 total | 166 | 190 | 224 |
| Larger-channel overperformers | 177 of 349 studied | not published | 181 | not published |
| Larger-channel underperformers | 172 of 349 studied | not published | 180 | not published |
| Small-channel hits | part of 941 studied | not published | 169 | not published |
| Small-channel flops | part of 941 studied | not published | 170 | not published |
How to read this: The percentile rows describe every video with a valid rate in the 349-video study, and the four group rows show that pace separated hits from flops by one word per minute at most.
Source: lib/seo/wpm-data.ts, overallWpm, wpmByGenre and wpmHitsVsFlops (349 videos, 36 channels, collected July 2026); lib/seo/guides.ts, does-talking-faster-help-youtube, for the 169 and 170 small-channel medians.
Scripted word counts by video length at the planning default
| video length | scripted words at 150 wpm with a 15 percent buffer |
|---|---|
| 60 seconds (Shorts) | 130-150 |
| 3 minutes | 380-460 |
| 5 minutes | 630-760 |
| 8 minutes | 1,020-1,220 |
| 10 minutes | 1,275-1,500 |
| 12 minutes | 1,530-1,800 |
| 15 minutes | 1,910-2,250 |
| 20 minutes | 2,550-3,000 |
| 30 minutes and up | 3,800+ |
How to read this: These are planning targets at 150 words per minute with the 0.85 buffer the word count post uses, and the measured median in the hook study ran higher, at 181 words per minute.
Source: app/blog/blog-data.ts, youtube-script-length-word-count, table 'Word count targets by video length'.
Hook signals in the first 45 seconds, hits against flops
| signal in the first 45 seconds | larger-channel hits | larger-channel flops | small-channel hits | small-channel flops |
|---|---|---|---|---|
| Engages the title's promise | 87.0% | 77.9% | 79.2% | 80.4% |
| Contains a specific number | 63.3% | 52.3% | 47.2% | 47.4% |
| Never gives a reason to stay | 10.2% | 16.3% | 21.0% | 19.8% |
| Opens with a concrete claim | 33.3% | 25.0% | 22.2% | 19.8% |
| Opens with a context dump | 18.6% | 26.2% | 30.6% | 29.3% |
| Opens with a greeting | 7.3% | 11.6% | 16.6% | 23.3% |
How to read this: Read each row across: the larger-channel gaps run 4 to 11 points, and on small channels five of the six gaps collapsed to 3 points or less.
Source: app/blog/blog-data.ts, youtube-hook-study and small-channel-hook-study; raw rows in public/hook-study-data.csv (349 rows) and public/small-channel-study-data.csv (948 collected rows).
Words to the first concrete reason to watch
| sample | median words to the first reason to watch |
|---|---|
| Larger-channel hits | word 14 |
| Larger-channel flops | word 15 |
| Small-channel hits | word 15 |
| Small-channel flops | word 15 |
How to read this: Both groups reach a point on the same schedule, so the schedule is table stakes and the content of the point is the variable left to test.
Source: app/blog/blog-data.ts, youtube-hook-study (word 14 and word 15); lib/seo/guides.ts, how-fast-get-to-the-point-youtube (word 15 in both small-channel groups).
How a flop is defined and how deep it runs
| measure | larger channels (349-video study) | small channels (941-video study) |
|---|---|---|
| Definition of a flop | each channel's bottom five videos by views, compared against its own top five | same design, top and bottom five per channel |
| Median overperformer views | 4.7x channel normal | 4.67x channel normal |
| Median flop views | 0.28x channel normal | 0.31x channel normal |
| Mildest flop in the study | 0.79x channel normal | 0.88x channel normal |
| Flops below half of channel normal | 140 of 172 | 78 percent |
| Middle 50 percent of flops | 0.14x to 0.43x | not published |
How to read this: Each column measures its own sample, so the two flop medians describe two groups of channels rather than two estimates of one number.
Source: app/blog/blog-data.ts, youtube-hook-study (0.28x, 0.79x, 140 of 172, 0.14x to 0.43x); lib/seo/guides.ts, why-did-my-youtube-video-flop (4.67x, 0.31x, 0.88x, 78 percent).
Retention benchmarks by video length
| video length | strong retention | exceptional retention |
|---|---|---|
| Under 5 minutes | 65-75% | 75%+ |
| 5-10 minutes | 50-60% | 60%+ |
| 10-15 minutes | 40-50% | 50%+ |
| 15-30 minutes | 35-45% | 45%+ |
| 30-60 minutes | 25-35% | 35%+ |
| 60 minutes and up | 20-30% | 30%+ |
| Shorts, under 60 seconds | 70-85% | 85%+ |
| Live streams | 10-20% | 25%+ |
How to read this: These brackets are aggregated from public benchmark studies and creator surveys rather than measured by Prepublish, so treat them as starting anchors and not as targets.
Source: app/blog/blog-data.ts, youtube-retention-benchmarks-2026; the length brackets rest on Backlinko's analysis of 1.3 million videos at https://backlinko.com/youtube-ranking-factors.
Methodology
This page compiles numbers Prepublish already published, covering 132 channels and 1290 videos collected between 2026-07-01 and 2026-07-31. The population is two fixed samples of public YouTube uploads, so treat each number as descriptive of those videos, and not as a universal benchmark. Source files:
lib/seo/wpm-data.tslib/seo/guides.tsapp/blog/blog-data.tspublic/hook-study-data.csvpublic/small-channel-study-data.csv
Compiled from two preregistered Prepublish studies of 1,290 videos from 132 channels, plus public retention benchmarks. Every number names its source.
Method
Both studies compared each channel against itself: its top five videos by views against its bottom five. Channel size drops out of that comparison, so a 40,000 view video from a 60,000 subscriber channel can sit in the same group as a 4 million view video from a large one. The 349-video study covers 36 channels and 254 million combined views. The 941-video study covers 96 channels between 1,000 and 20,000 subscribers and 31.8 million combined views.
The small-channel study was preregistered before any hook was collected. Its predictions, enrolled channel list, classifier prompt, statistical tests, and verdict rules are frozen and timestamped at https://osf.io/2zq9p, so the outcome could not be chosen after the fact.
Row-level data is public. The 349 rows are at /hook-study-data.csv and the small-channel rows are at /small-channel-study-data.csv, which holds 948 collected rows and 941 analyzed rows. Each row carries the video URL, the view count, the channel median, the over or under ratio, the words per minute, and every classification, so any row can be checked against the video on YouTube.
Classification was AI-assisted against fixed definitions and blinded to the view count, the channel, and the group. Both datasets were collected in July 2026. The videos inside them were uploaded before that month, so the date range on this page covers collection time, not upload dates.
Limits
These are view-performance studies, not retention studies. Views and channel-normal ratios describe which videos got watched, and nothing on this page measures how long a viewer stayed.
Null results are published as null results. The preregistered replication returned its registered "can't tell" verdict on all three confirmatory opening patterns, with 94 to 98.5 percent power to detect the original effect sizes, and that outcome is reported here next to the patterns that did separate hits from flops on larger channels only.
The two flop medians on this page, 0.28x and 0.31x, come from two different samples, so neither contradicts the other. Where a figure appears in both a study write-up and a guide, this page keeps the study write-up that links the public CSV.
A medians check run over the raw CSV can move by a point or two from the published figure, because the published analysis dropped rows, for example the two small channels that fell below three usable videos per group. This page reports the published figures and links the raw rows so any reader can recompute them.
The greeting signal leaned toward flops in both studies, at 23.3 percent of small-channel flop openings against 16.6 percent of hits, and it was registered as exploratory in the replication, so it gets no verdict and this page gives it none.
Every number is an association, not a cause. Outlier status reflects the whole video and its packaging, and a video can flop with a strong opening because the topic was dead on arrival.
The analyzer on this site returns a relative script-hold index over a draft, not a forecast of published viewers. It reads text only, and it cannot account for delivery, editing, topic, thumbnail, or distribution.
The retention brackets in the last table are aggregated from public benchmarks and creator surveys rather than measured by Prepublish, and no single study covers the niche level at scale.
Cite this
Canonical URL: https://prepublish.ai/research/youtube-script-statistics
Source: Prepublish script statistics, prepublish.ai/research/youtube-script-statistics, verified 2026-09-21