It reads your last thirty videos and says what follows from them.
Your last 30 published
Six, each with its source
Nothing, unless you keep it
It reads the channel, not the internet
The last thirty videos you actually published go in: their titles, their formats, how each one did against your own median. Six suggestions come back, and every one names which of your videos it follows from, so you can disagree with the reasoning rather than just the idea. A suggestion with no provenance is a guess with a confident font.
Where the line still is
This screen suggests what to make. It does not tell you how any of it will do, and it never will: the moment a model is allowed to say a video will work, every measured number beside it becomes marketing. So there is no predicted view count, no confidence score and no ranking by expected performance. Six things that follow from what you have already published, and the decision stays yours.
Nothing is kept unless you save it. A saved suggestion waits on the runsheet’s empty days; the rest are gone when you leave, because a backlog of generated titles nobody wants to film is the thing this screen replaced.
The same line holds everywhere in the product: a model writes copy here and never produces a finding. Every number Runsheet reports about your channel is measured. The methodology page has the thresholds.
What this feature will not do, stated where you can read it before paying.
- A suggestion is a starting point, not a prediction. Nothing here estimates views, and no finding anywhere in Runsheet is written by a model.
- It needs published videos to read. A channel with nothing on it gets nothing back, and the screen says so rather than inventing six.
- Suggestions are not stored unless you save one, and reloading the page discards the rest. A fresh set needs the button, which uses one drafting run: Free has 3 a day.