Gemini Notebook Prompt Lab

1What it does

A prompt laboratory for NotebookLM — a maintained collection of tested prompt patterns organized by job rather than by domain. The lab's structure mirrors how NotebookLM actually gets used: chat prompts that interrogate your sources, research flows that chain questions into an investigation, and format recipes that steer the built-in generators (study guides, briefing docs, table outputs) toward specific shapes. Each prompt is written to be pasted into the notebook's chat or a generation dialog as-is, with the source-grounding already baked into the phrasing — "based only on the sources in this notebook" style constraints that keep the model honest. The lab framing matters: prompts are presented as experiments with notes on what they were built to elicit, not as a listicle of magic phrases.

2When to use it

Reach for it when your NotebookLM sessions produce mushy output — the difference between "summarize this" and a prompt that names the audience, the output shape and the source boundary is the difference between a paragraph of filler and a usable artifact. It fits researchers and students who want a working prompt vocabulary fast, and anyone who keeps retyping the same source-grounding clauses. Skip it if your use is already template-driven elsewhere (you maintain your own prompt file) — the lab's value is breadth and testing notes, not a single canonical pattern you'll copy forever.

3How to get started

Browse the repository's README and pick one prompt family that matches your immediate job — research chat, study material generation, or briefing output. Copy the prompt verbatim into a notebook that already has sources loaded; the patterns assume source grounding, so an empty notebook will just produce generic model output. After the first result, edit the audience/format clause rather than the grounding clause — the grounding language is the load-bearing part. The repo is plain markdown; no install, no tooling.

4Configuration

Configuration is prompt-level: the audience, format, length and source-scope clauses at the tail of each pattern. Keep the source-restriction phrases intact ("only from the uploaded sources", "cite the source document") — they are what separates NotebookLM behavior from a general chat. For repeat use, maintain a personal fork or notes file with your edited versions; upstream evolves as the author tests against current NotebookLM behavior. Language adaptation is safe — the patterns translate cleanly as long as the grounding clause survives translation.

5Things to know

• Prompt libraries age with the product — NotebookLM's chat has gotten better at default grounding, so some "make it cite" prompts matter less than they did; the lab notes which patterns still earn their keystrokes.
• Pairs with any structured workflow: use it to fill the chat half while a template repo covers the artifact half.
• MIT-licensed markdown — forking into a team wiki is the intended use.
• Prompts can't read unindexed sources — if a prompt seems to fail, check the source actually parsed (PDF scans without OCR are the usual culprit).
• The lab treats prompts as hypotheses — several entries carry notes on when the pattern stopped being necessary as NotebookLM's defaults improved, which is the honest way to run a prompt library.
• Because it is a lab, entries sometimes carry counter-examples — prompts that look right but underperform, kept with notes on why; reading those is as instructive as the patterns that work.
• Fork-and-trim is the intended pattern — most users lift five or six patterns into a personal prompt file and tune the audience clauses to their own sources.
• The lab's strongest section is format steering — prompts that reshape the built-in generators (briefing doc tone, study-guide sectioning) are harder to write blind than chat prompts, and that is where most of the tested phrasing earns its place.

6Related on the map

References

• Repository
• NotebookLM help — chat with sources