Docs / Grow and learn
Retention to script loop
Where viewers leave each episode and what they rewatch, mapped onto its own lines and captions, and turned into rules the next scripts follow.
Views say how many people started an episode; retention says where they stopped. The retention loop maps each posted episode's retention onto the episode itself — the line being said, the caption on screen, the scene — finds what repeats across a channel's episodes, and writes it down as rules. Switch the rules on for a channel and its writer and its critic get them for every new episode.
What it does#
Each episode, on its own timeline#
Once a day, at an hour of the workspace's own, ChannelForge takes each posted episode's retention and lays it over the episode's own timeline: its narration lines, its captions, its beats (the title card, a caption on screen, no caption on screen, the end card), how much the picture moves, the cuts, and the segments of its 90-second edit. It marks the moments that matter:
| Moment | What it is |
|---|---|
| A steep drop | viewers leaving faster than the channel's usual decay at that point |
| A rewatch | a bump: viewers going back over a stretch |
| The leave point | where the average viewer stops watching |
Each moment carries what was on screen and which line was playing, so "people leave at 0:41" becomes "people leave during the line about the spring, while no caption is on screen".
Where the numbers come from#
| Platform | What the platform reports | What the loop can show |
|---|---|---|
| each Reel's average watch time | where the average viewer leaves, placed on the episode's timeline | |
| each Reel's average time watched | the same | |
| YouTube | the share still watching at 100 points along the video | the whole curve, against the channel's usual |
Instagram and Facebook give one number a video, divided by the length that was posted (the 90-second edit where that is what went out). YouTube gives a full curve.
Rules, once there is enough#
A channel gets rules once at least five of its episodes have been measured on a platform. Each rule is written by code from the measurements, with the numbers it rests on, so it can never say more than they do:
| Rule | From | What it tells the writer |
|---|---|---|
| The hook | YouTube's curve | half the audience is gone by the channel's usual half-life, so land the first consequence a few seconds before it; and, tested, whether episodes with a longer title card held fewer viewers to 10 seconds |
| Still pictures | YouTube | the steep drops came while the picture barely moved, far more often than still stretches fill the running time |
| Silent stretches | YouTube | the steep drops came while no caption was on screen |
| Rewatches | YouTube | what viewers went back over in two or more episodes — make more of that |
| The ending | YouTube | the end card loses a large share of who is still there (a quarter or more, over five or more episodes) — an ending that loops keeps them (not said where endings loop already) |
| Where they leave | Facebook, Instagram | where the average viewer leaves, and which narration line usually plays then; and, tested, whether episodes with a longer title card kept a smaller share watched |
A rule that rests on a test is labelled by how sure it is — the same statistics as Insights, with Benjamini–Hochberg q-values across the channel's tests. A rank-correlation test (the title card) needs at least eight episodes; the still-picture and silent-stretch tests need at least four steep drops across three or more episodes, with five or more measured.
| Label | Means |
|---|---|
| Tested | the test is strong (q below 0.05) |
| Worth trying | suggestive (q below 0.20): try it on purpose |
| Measured | a measurement, with no test behind it |
A test that does not clear the bar drops its sentence; a rule with nothing measured behind it is dropped.
Predicted against measured#
Before an episode is posted, the critic scores it, including for entertainment. Predicted vs measured puts those scores beside the share of the video its viewers actually watched, so you can see whether the critic's sense of an episode matches your audience's.
What the writer and the critic do with the rules#
With the channel's switch on, every new episode's brief carries the channel's rules — at most eight — and the critic reviews the episode knowing them. The rules inform each new episode; they never override the house rules on entertainment, facts and variety. Episodes already made are not changed.
Turn it on#
Editors, admins and owners can change it, per channel.
- Open Retention in the sidebar (or the Retention card on a channel's page).
- If you have several channels, choose one at the top.
- Under the rules, tick Give these rules to the channel's writer and critic and press Save.
From the channel's next episode, its writer and its critic get the rules. Untick and save to stop; the next episode's brief is then exactly what it was before. The weekly coach may also recommend Use the retention rules for a channel, as one click.
Settings#
| Option | Default | Effect |
|---|---|---|
| Give these rules to the channel's writer and critic | off | each new episode's brief and its critic carry the channel's rules (at most eight) |
Credits and plans#
Mapping retention and writing the rules run on our servers in code: they cost no credits. Giving the rules to the writer does not change what an episode costs. The retention loop is on the Creator plan and above; on the Trial the page explains it and links to the plans.
Limits and good to know#
- Your channel's own numbers. Each workspace's retention is mapped and learnt on its own; nothing is pooled with another customer's.
- A link, never a cause. A rule says what went with viewers leaving; it does not prove why.
- Kept fresh. Maps and rules not recomputed for 30 days are deleted.
Troubleshooting#
| You see | Why, and what to do |
|---|---|
| "Nothing measured yet" | No measured retention for the channel's posts yet — see the note above, or the channel has posted too little. |
| "No rule yet: they come once at least five of the channel's episodes have been measured on a platform" | Keep posting; rules appear by themselves. |
| "Retention is in the Creator plan and above" | The workspace is on the Trial. An owner can upgrade in Settings → Billing. |
Related#
- Insights
- Hook lab — the first seconds, tested directly
- Captions and endings — endings that loop
- Quality gates — the critic and the house rules
- The coach
Something missing or unclear? Tell us and we will improve this page.