15 NotebookLM tips and tricks power users swear by

I have sorted these by how much they change your results, not by how clever they sound. The first five are worth doing in every notebook you create. The rest depend on what you use NotebookLM for.

1. Scope questions with the source checkboxes

Every source has a checkbox, and almost nobody touches them. Untick everything except the two documents you want compared and the answer quality jumps: no bleed-through from unrelated files, cleaner citations, sharper comparisons. This is the single highest-value habit in the entire app.

2. Recycle notes into sources

Save a strong answer with Save to note, then use Convert to source from the note's menu. Now your own distilled synthesis is queryable material. Power users run whole projects this way: generate a briefing doc from twenty sources, convert it, then untick the originals and chat against the distillation. Answers get faster and tighter, and you claw back source slots at the same time.

3. Rename every source before you generate anything

Source titles feed the model context. A mind map built from document(3).pdf, final_v2.pdf and scan0001.pdf is measurably worse than one built from descriptive names. Thirty seconds of renaming beats most prompt engineering.

4. Set custom chat instructions once per notebook

Open the settings icon in the Chat panel and pick Custom. Whatever you write there applies to every future question in that notebook. A working example: "Cite the source and location after every claim. If something is not in the sources, say 'not in the sources'. Skip introductions." No more repeating yourself in every prompt. Full walkthrough in the chat settings guide.

5. Click citations like it's a habit

NotebookLM is grounded, not infallible. A journalism study measured roughly 13% of claims as unsupported, far better than open chatbots but not zero. The citation chips make verification a two-second act: hover to preview, click to jump into the source. Do it for anything you plan to reuse. More on this in citations and accuracy.

6. The "don't summarize" audio hack

The Audio Overview hosts are summarizers by default. Tell them not to be: "Do not summarize. Go through every section in detail, discussing every concept, example and number." Combined with the Longer length setting, people reliably get 40 to 60 minute episodes out of material that would otherwise become a 12 minute overview. The inverse works too: "stay within 5 minutes" tightens a rambling episode.

7. Give the podcast hosts jobs

The default hosts agree with each other to the point of parody. Assign roles in the customize prompt: one host explains, the other challenges the methodology and plays devil's advocate. The result has actual tension and surfaces weaknesses in your material. More prompt recipes in Audio Overview prompts.

8. Beat the source cap with Google Docs tabs

One Google Doc with ten tabs imports as one source, but each tab stays a self-contained section. People consolidating research this way report cutting their source count by well over half. Merging related PDFs into one file works on the same principle. The tradeoff and two other methods are in source limits.

9. Use mind map nodes as a navigation system

A mind map node is not a picture, it is a button: click it and chat opens pre-scoped to that branch. And read the map critically. Thin branches mean thin coverage, so a sparse branch is literally a list of what to study or source next. See mind maps.

10. Feed it YouTube playlists

Any public YouTube URL with captions becomes a source. Paste in five tutorials and ask "rank these from most foundational to most advanced" and you have a watch order. Ask a question and the citation jumps to the exact transcript moment. This quietly replaces hours of passive watching; workflow details in YouTube sources.

11. Interview the audio instead of just listening

Interactive mode (the Join button on a playing Audio Overview, English only for now) lets you interrupt the hosts with a spoken question. They answer from your sources and pick the episode back up. It turns a commute podcast into office hours.

12. Force contradictions into the open

Summaries average your sources into one smooth story, which is exactly wrong for research. Ask directly: "Find every disagreement between my sources. For each: the claim, who takes which side, and which has stronger evidence." This one prompt is the difference between an annotated bibliography and an actual analysis. More in best prompts.

13. Attach notebooks to Gemini for cross-notebook questions

Notebooks cannot talk to each other, which becomes painful once you follow the good advice of keeping them small and focused. The workaround: in the Gemini app, attach several notebooks to one conversation and ask across them. It is the only current way to synthesize between notebooks.

14. Keep a prompts notebook

Custom instructions do not transfer between notebooks. Keep one notebook containing a single note with your best-performing prompts, and copy from it when you start a project. Boring, effective.

15. Get your work out before you need it

The classic NotebookLM regret is doing two weeks of analysis and then discovering there is no clean way to hand it to anyone. Some reports export to Google Docs, but chats and most outputs do not. The NotebookLM to PDF extension adds a one-click export to PDF, Word or Markdown with citations intact, and a regular backup habit means a deleted notebook is an inconvenience instead of a disaster.

FAQ

What is the most useful NotebookLM trick?

Scoping questions with the source checkboxes. Unticking everything except the sources relevant to your question sharpens answers more than any prompt technique.

How do I make NotebookLM Audio Overviews longer?

Choose the Longer length setting, then add a customize prompt telling the hosts not to summarize and to cover every section in detail. Episodes of 40 minutes and more are common.

Can NotebookLM notebooks reference each other?

Not directly; each notebook is a silo. The workaround is attaching multiple notebooks to a Gemini conversation and asking questions across them there.