Why use transcript files instead of pasting video links?
NotebookLM can ingest a YouTube link directly, but one link at a time, and only the video's own transcript. For a whole channel — a course, a podcast archive, a technical back catalog — you want the transcripts as files, uploaded in bulk, so a single notebook can answer across hundreds of hours of material at once: "when did she change her position on X?", "list every tool he has recommended," "summarize the argument across the series."
How do you get a channel into NotebookLM?
1. On scripthaul.com, paste the channel or playlist link.
2. Choose TXT as the file type with timestamps Off — clean prose is what NotebookLM digests best; timestamp markers on every line just add noise to its retrieval. (If you want to jump back to moments in the video afterwards, run the job again with timestamps on — the second pass is instant from cache.)
3. Download the ZIP and unzip it.
4. In NotebookLM, create a notebook and add the .txt files as sources. Files upload in batches; each becomes a citable source named by date and title, so answers point at the specific video they came from.
How do you work within NotebookLM’s source limits?
As of mid-2026, a NotebookLM notebook accepts up to 50 sources on the free plan (more on paid tiers), each up to roughly 500,000 words — check Google's current limits, as they change. For channels bigger than the source cap, two honest strategies: curate the 50 most relevant videos (the manifest's date-and-title listing makes triage fast), or split the channel across several notebooks by era or topic. A typical 15-minute video runs ~2,000 words, so a single source slot can hold dozens of concatenated transcripts if you merge files — the trade-off is coarser citations.
How reliable are the transcripts?
Auto-generated captions carry YouTube's recognition errors and no punctuation; creator-written captions read like edited text. The ZIP's manifest.csv records which kind each video used (the caption_kind column — manual vs asr), so you know how much to trust a quoted line before citing it anywhere serious.
Related: download a whole channel, JSON for RAG pipelines (when you outgrow NotebookLM), or the channel downloader tool.