What a NotebookLM prompt pack is
A NotebookLM prompt pack is a set of chat prompts tuned for the notebook's source-grounded model — not for a general chatbot. The difference matters. Prompts written for ChatGPT or Claude lean on the model's general knowledge; in NotebookLM that habit backfires, because the product's whole point is answering only from loaded sources. The packs that work here are the ones written against that constraint: they name the sources, ask for citations, and treat missing information as a finding rather than something to paper over.
Most packs arrive as a markdown file or a short doc you keep beside the notebook. There is nothing to install — you paste prompts into the chat box one at a time, usually after loading sources but before generating any of the built-in outputs (mind map, audio overview, study guide). The good ones are sequenced: a scoping prompt first, then extraction prompts per source, then synthesis prompts that reconcile disagreements across sources.
How to evaluate a prompt pack before you adopt it
- Ask whether it references sources. A prompt that says "summarize the key arguments" produces generic output; one that says "summarize the argument in Source 3 and cite the section" produces something checkable. The second kind is what this page lists.
- Check whether it handles the empty case. Source-grounded models refuse gracefully when the answer is not in the documents. Packs that include a "if the sources do not cover X, say so" clause save you from confident inventions.
- Prefer packs organized by task over packs organized by tone. "Extraction / quiz / synthesis" is a workflow you can reuse; "friendly / formal / punchy" is not.
- Watch for prompts that only work on one source type. A pack tuned on dense PDFs can produce thin output on transcript sources — the better ones say which input they were built against.
Official versus community material
Google publishes very little prompt material for NotebookLM — the product is designed to be used through its built-in buttons, so most prompt knowledge is community-discovered and reverse-engineered. That makes verification the whole job of this page. Every pack here was run inside a real notebook with real sources before listing; the entry notes what output it produced and where it fell short. Packs that only work on a narrow source type say so on their entry rather than letting you discover it yourself.
One last distinction worth internalizing: prompts here shape the chat, not the generated artifacts. If your goal is a better mind map or audio overview, prompts help indirectly at best — the levers that matter there live in how your sources are structured, which is what the templates hub is for.