Changelog

We tested the scoring model against real Google AI Overview citations — videos we confirmed were actually cited, compared against similar videos on the same topic that weren't. Cited videos scored higher two-thirds of the time, a result that held up across multiple independent rounds of testing, not just a one-off lucky sample.

A few things came out of that testing:

• Videos with auto-generated captions (no punctuation) were being unfairly penalized for "unclear" transcripts, even when the content itself was perfectly clear. Fixed.
• Short, specific titles were being flagged as "not specific enough" just for being short. That check now judges specificity on more than length.
• Chapter structure carries less weight in the overall score now — in testing, it wasn't a reliable difference-maker between videos that got cited and videos that didn't.

Six dimensions — transcript clarity, chapter structure, description quality, title extractability, topic specificity, and freshness signals — weighted based on published research on what AI engines tend to cite from YouTube.