Commands
| Command | What it does |
|---|
/blog notebooklm <question> | Queries the active notebook and returns a source-grounded answer with file and page citations |
/blog notebooklm setup | Runs the one-time Google OAuth flow in a browser tab and caches the session locally |
Building a research notebook
NotebookLM accepts PDFs, Google Drive docs, web URLs, and YouTube transcripts as source inputs. The skill quality is bounded by the corpus quality, so curate deliberately. A typical topic notebook benefits from 10 to 30 trusted sources: peer-reviewed papers for the empirical claims, industry reports for benchmarks, primary documentation for technical specs, and a handful of expert blog posts for framing.
Once your notebook is populated, you can query it from Claude Blog with natural language. NotebookLM only answers from the uploaded corpus, so a sparse or off-topic notebook will produce sparse or off-topic answers. Treat the notebook as the research substrate, not the question scope.
Inside the writing loop
The typical flow: kick off /blog write with a topic; pause when you need a citable statistic; run /blog notebooklm against your topic notebook; paste the returned passage with its citation back into your draft. /blog factcheck later verifies the numbers against the same sources, so the trail stays clean from research to publish.