How VidCC is backed by research
A retention graph shows where viewers leave. Working out why takes the video itself. VidCC's models are grounded in research that studies retention graphs together with the videos behind them, across many kinds of channel.
Why research matters here
Advice about retention is easy to find and hard to check. A tip that helps a cooking channel can hurt a gaming channel, and a graph on its own can't say which one you are. VidCC's models are built on research that looks at retention graphs and the videos behind them together, so its suggestions are grounded in what tends to happen on screen when a curve changes shape.
What the models learn from
The research covers a large number of retention graphs, each studied next to the video it came from, across many niches and formats. Every niche has its own normal, so a curve is read against the kind of video it belongs to rather than against a single universal benchmark.
Across that material, a small set of shapes comes up again and again. Each one is treated as a prompt to go and look at the footage, not as a verdict.
- Early drop
- A steep fall in the first seconds. The models check what the opening shows against what the title and thumbnail promised.
- Plateau
- A stretch where viewers mostly stay. The models note what is on screen and in the audio, so the pattern can be repeated.
- Slow decline
- Gradual loss over time is normal. The models look for changes in slope rather than the slope itself.
- Spike
- A moment viewers skip to or replay. It can be the best part of a video or a part that was hard to follow.
How findings are tested
Before an analysis approach ships, it is tested against real videos. Suggestions are compared with what the footage shows at the flagged time, and approaches that point at the wrong moment or explain a drop poorly are revised or dropped.
Testing continues after release. When the analysis misses, that feedback goes back into the next round of research.
What it can and can't tell you
VidCC describes what happens in the video at each flagged moment and offers a possible reason viewers left. It can be wrong, and no edit guarantees more views. Treat it as a second opinion on your own review, and change one thing at a time so you can compare the next upload's curve.
See it on your own videos
Connect a channel and open any video with retention data. VidCC draws the curve beside the playable video, marks the moments worth reviewing, and lets you ask Viddy about any of them.