The analytics loop
Click-through rate and retention come back from YouTube’s APIs. Hooks, titles and thumbnails of the next film learn from them, per channel.
Planned
What is read
With a connected channel, VidSwarm reads two YouTube APIs every night for each published film. Views, watch time, retention and subscribers come from the YouTube Analytics API. Thumbnail impressions and click-through rate come from the YouTube Reporting API’s daily reach reports (channel_reach_basic_a1), which arrive about two days late; the Analytics API does not deliver them reliably. What VidSwarm reads:
- Thumbnail impressions and click-through rate per day (Reporting API).
- Average view duration and average percentage viewed.
- The audience retention curve.
- Views and watch time at 48 hours, 7 days and 28 days.
- Estimated revenue and RPM (revenue per 1,000 views), only for monetized channels connected with the revenue permission. They appear on
/app/c/[channel]/analyticsonly when YouTube reports them. - The 20 newest top-level comments, read-only, shown on the film’s analytics page.
Without a connected channel, the loop does not run. You can still enter numbers by hand on the film’s analytics page.
Where to see it
Each film has /app/c/[channel]/videos/[id]/analytics with its numbers, the retention curve laid over the scene list, and the moments where viewers left. /app/c/[channel]/analytics compares films across a channel and across channels.
What the loop learns
After each film, VidSwarm compares its numbers with the channel’s earlier films at the same age and writes learnings, for example “Cold opens at the most dramatic moment held more viewers at 30 seconds than question openings” or “Thumbnails with a single object had higher CTR from browse”. Learnings are listed on the channel page with the films they are based on.
How learnings are used
When writing the next film, the hook, title options and thumbnail variants take the learnings into account. You can switch off or delete any learning. A learning that contradicts one of your channel skills is shown as a suggestion; your skill wins until you change it.
Small numbers
On new channels, single films are noisy. VidSwarm marks learnings with how many films support them and does not act on a pattern from one film alone.