Support content has a timing problem: the FAQ gets written from the ticket queue, which means the queue has to fill before the deflection exists. Every recurring ticket answered manually while the docs catch up is cost that better timing would have deleted.

The tickets are forecastable. How-to videos about your product — official and third-party — are a map of what users need explained: every tutorial exists because a question recurs, every chapter marks a sub-question, every comment under it asks the follow-up. The ecosystem has effectively pre-written your FAQ's table of contents; support just needs to read it before the queue does.

library_import_youtube_video collects the forecast corpus: the how-to and troubleshooting videos about your product, imported with transcripts. run_youtube_tool processes them into support assets: how-to-extractor pulls the procedural answer from each video into step form; document drafts the article; comment-summarizer on the same videos surfaces the follow-up questions the video didn't answer — the second layer of your future queue.

The output is a deflection library drafted from demonstrated demand: articles for the questions tutorials already prove people ask, written before those questions become tickets. Post-launch, the same pass on first-look videos (see the launch-triage use case) keeps the FAQ ahead of the queue instead of behind it.