A prompt library has a predictable adoption problem: most people will not read the guide that explains how to use it well. A folder is a reference, not a learning experience.

The Daft Gemini Playbook includes a second way in. Load a focused set of its documents into NotebookLM and the library becomes something you can question. Why is this prompt Red? What would make the output unreliable? What should I check before I act? Those questions are far easier to work through in a conversation than by reading a specification end to end. The initial setup takes about 90 minutes; the learning can happen at your own pace.

Curation matters more than coverage. NotebookLM answers from the sources you give it, so those sources become the syllabus. Put everything in one notebook and the answers tend to get broader and less useful.

Start with five foundation documents: the README, operating guide, risk-and-approval guidance, prompt standard, and review rubric. Then add 15 to 25 prompt files that fit your role, plus the eight recommended starting prompts. Do not load all 112 by default. The generated index, CI files, and changelog can stay out too. They are useful in the repository, but they do not make a better tutor.

Build your own notebook rather than passing around one shared version. A notebook gradually reflects the sources you chose and the questions you had to work through. That makes it more useful when it is built around your role.

Keep live work out of it. Do not add current reports, exports, leases, registers, or correspondence. A study notebook is not a production workspace. Once real operational data enters the mix, a learning aid becomes a records-management and sharing problem.

The study paths matter less than the habit behind them. Ask about failure modes, not just features. “What is prompt 011 for?” is fine. “How could prompt 011 be wrong in a way I would miss?” is much more useful. Ask the notebook to challenge your interpretation, too. A well-supported disagreement teaches more than an easy confirmation.

For the first week, follow the citations and read the underlying source. That builds the judgment to tell the difference between an answer that is grounded and one that is reaching. The notebook is not just teaching someone how to run the library. It is teaching them how to review the result before it matters.

There is a limitation here. This is not a centrally managed training asset. There is no single notebook to hand a new hire, and a large group will create many individual notebooks that need to be refreshed as the library changes. Audio Overview can be helpful for a quick introduction, but it is not a substitute for working through the source material.

The goal is modest and valuable: build better reviewers. If the notebook finds a conflict between two sources, it has found something worth fixing. And if you cannot tell whether an output is right, or would not notice if it were wrong, that is the point to stop prompting and ask the person accountable for the decision.