YouTube Script Benchmarks 2026: Words per Minute, Script Length and Openings

Key findings

1. Across 3,074 English YouTube videos from 1,310 channels, the median finished video carries 179 words per minute of runtime (middle half: 155 to 203).
2. Videos of 8 to 12 minutes run 1,786 words at the median (middle half: 1,502 to 2,068) across 470 videos from 320 channels.
3. Videos of 15 to 20 minutes run 3,208 words at the median (middle half: 2,810 to 3,635) across 568 videos from 360 channels.
4. Words per minute is lower at both ends of the length range than in the middle: 1 to 3 minute videos run 166 words per minute at the median, 3 to 5 minute videos 164, 5 to 8 minute videos 170 and videos over 60 minutes 166, against 180 to 185 in the 8 to 45 minute bands; videos shorter than 60 seconds are not in the sample at all, so the shortest band starts at 1 minute.
5. Science & Technology is the fastest category in the sample at 188 words per minute median across 523 videos, and Entertainment is the slowest at 175 across 324 videos.
6. Across the 3,035 videos long enough to measure, the first 30 seconds of speech contain 90 words at the median (middle half: 75 to 104).
7. 88.8% of the 3,074 videos have their first caption with speech within 5 seconds, at a median of 0 seconds (the first caption with words starts at the very beginning in at least half of the videos) and a 75th percentile of 1 second.
8. 19.0% of videos open with a greeting in their first 12 words of speech, and Gaming has the highest greeting share at 36.3% across 223 videos.
9. 69.8% of the 3,070 videos with an eligible title word say one within 15 seconds and 81.4% within 30 seconds, at a median of 5 seconds.
10. Manual caption tracks read 167 words per minute at the median across 322 videos and automatic tracks 181 across 2,752, and the tables above pool both kinds of track, so a manually captioned channel may read somewhat lower than its category median.

The whole sample at a glance

Every figure is a percentile over the 3,074 videos in the sample; words per video is only reported at the 25th, 50th and 75th percentiles.

Words and pace by video length

Read a row across to compare a runtime band with the whole sample, and down the median column to see where pace rises and falls with length. A band with fewer than 50 videos keeps its counts and prints an em dash instead of statistics.

Words per minute by YouTube category

Rows are YouTube's own category labels, ordered by video count. "Other categories" pools every category below 100 videos or 30 channels, which is why it has no single label of its own.

How the opening of the spoken track behaves

Time to the first title word uses the 3,070 videos whose title carries a word of four letters or more that is not a stopword; a title with no such word is left out of that row.

Manual and automatic caption tracks

This is a robustness check, not a correction: the manual and automatic rows are reported separately so a reader can see how much the caption source moves the median.

What the sample is

The sample is 3,074 English YouTube videos from 1,310 channels. Caption tracks were fetched between 2026-09-23 and 2026-10-24, and the videos themselves were published between 2008-08-22 and 2026-10-08.

The videos are the ones people and Prepublish's transcript indexer requested transcript pages for, so the sample leans toward popular, talking-head videos rather than a random draw of YouTube.

Each video contributes one public caption track, 322 of them manual and 2,752 generated automatically. Categories are YouTube's own category labels, fetched on 2026-10-30, not a taxonomy this page invented.

Method

Words per minute is caption words divided by the video's full runtime, including intros, pauses and any silent stretch. It is the number of words that fill a minute of finished video, not how fast the host talks.

Percentiles use linear interpolation, the same rule as PostgreSQL's percentile_cont. A median is the 50th percentile and the middle half is the 25th to the 75th.

At most 10 videos per channel are kept, chosen by the smallest sha1 digest of the video id so the choice is deterministic. Without the cap a channel that uploads constantly could fill a runtime band on its own; the cap removed 1,897 videos.

Exclusions, each counted once: 203 videos shorter than 60 seconds or longer than 3 hours, 21 with fewer than 100 caption words, 20 with no YouTube category on record, 55 in the Music category, and 2 whose caption words implied more than 400 words per minute.

Opening features: the first caption with speech is the first timed segment whose cleaned text contains a letter, after bracket tags, parentheses and music symbols are removed. Words in the first 30 seconds counts words in speech segments starting before 30 seconds, and is only computed for videos at least 2 minutes long. A greeting opening means a greeting word such as hey, hi, hello or welcome, or a phrase such as good morning, appears in the first 12 words of speech. The time to the first title word is the start of the first speech segment containing a title word of four characters or more that is not a stopword, matching exactly or after dropping one trailing s on either side.

The pipeline is read-only and reproducible: it exports public caption tracks, fetches each video's public category, and writes the aggregates with no timestamps, so two runs over the same inputs produce identical output. The code lives in ops/seo-data/benchmarks/.

The pipeline lives in ops/seo-data/benchmarks/ (export.sql, categories.sh, build.py, README.md) and runs read-only against the stored caption tracks.

Limitations

Every number here is descriptive of the sample. Nothing on the page says that a faster pace, a longer video or a particular opening causes better results, and no association is offered as a target.

No view counts, retention curves or engagement data were used. The table this page reads holds caption tracks, runtimes, publish dates and public category labels, so the page cannot speak to whether any pace or opening performed well.

Manual and automatic captions do not agree: manual tracks read 167 words per minute at the median across 322 videos and automatic tracks 181 across 2,752. Most of the sample is automatic captions, so the pooled median leans toward how a speech recognizer segments speech.

Category labels are chosen by uploaders, so two channels making similar videos can sit under different labels, and a label says nothing about a channel's subject beyond the uploader's own choice.

Caption timing is the uploader's or the recognizer's, so a median first-speech second is only as precise as the track it came from.

The sample is not a random draw of YouTube, and it is not a map of the platform. It is the set of videos someone asked Prepublish to transcribe, which over-represents popular videos with a caption track.

What we left out and why

Chapter and section counts are not reported. Prepublish does not store video descriptions or chapter markers, so there is nothing to count, and a guess would be worse than an absence.

Hook length, intro length and time to the first topic statement are not reported either. Each needs a human judgment of where the hook, the intro or the topic statement ends, and caption timing cannot mark that boundary reliably. Instead the page reports two proxies: words in the first 30 seconds, and the time to the first title word.

Niche-level rows are pooled rather than shown. A category with fewer than 100 videos or fewer than 30 channels sits in "Other categories" (311 videos from 165 channels) instead of appearing as its own row with a wide, unstable median.

Frequently asked questions

How to cite

Use these numbers

• Script runtime calculator — turn a word count and a pace into a video length, and back.
• Words to minutes converter — the same conversion with the pace presets written out.
• YouTube Partner Program calculator — check the current eligibility thresholds rather than an old habit about length.
• How many words should a YouTube script be? — the word counts behind the runtime bands on this page.
• Writing a documentary-length script — structure for long-form runtimes.
• The hook study — the opening compared between hits and flops, with performance data this page does not use.
• The first 30 seconds — what to do with the opening once you know how little room it has.
• The transcript hub — the video pages this sample was drawn from.

Example transcripts from the sample

• But what is a neural network? (Education) — an Education video, a category that runs 3,123 words at the median across 993 videos.
• A Science & Technology video — the fastest category in the sample at 188 words per minute median.
• A Howto & Style video — the shortest median video length of the named categories at 13.2 minutes.