NotebookLM training quiz: create and check a first draft

A NotebookLM training quiz starts with one source, a short Studio quiz and a check of the proposed answers. After this exercise, you can generate a practice set from six fictional handover rules and save review decisions for its first three questions. The result is a checked draft, not a certification of employee competence.

Google's current help pages use the name Gemini Notebook, while many readers still search for NotebookLM. This guide follows the desktop browser route. AI means artificial intelligence. Its role here is to propose questions from supplied material, not to invent your organisation's operating policy.

Content last updated: 15 October 2026. The studio, source rules and review examples below are fictional teaching material. No live account test, participant assessment or measured training result is reported. The fifteen-minute plan assumes an existing working Google account with access to the product. Registration and access troubleshooting are outside that allowance.

What should the first practice set actually test?

Imagine a small design studio teaching a new coordinator how to hand over approved files. The immediate learning goal is narrow: choose the next step when an approval or version is missing. Asking an assistant to build an entire onboarding course would make it harder to inspect whether the first questions are useful.

Choose a low-stakes training example for the first run. The rules below are not a real studio's instructions and do not cover every situation in client delivery. Do not use this sample as an operational policy. A real procedure needs its owner's approval before you assess anyone against it.

For the source-writing stage, the standard operating procedure guide explains how to document an agreed process. Here, we begin with a supplied fictional procedure and create practice questions. That separates writing the rule from checking whether a question reflects it accurately.

Which source should you add to the notebook?

Copy the following block as one source. Keep its title, version and N labels. Those labels make it possible to point to a particular rule during review. You do not need client names, unpublished design files or employee records to practise this workflow.

FICTIONAL TRAINING SOURCE, version 01. N1: Check that the delivery request identifies the approved file version. N2: If approval is missing, pause delivery and ask the project owner. N3: Label the delivery note with the project code and version. N4: If a client requests a new change, record the request for the project owner; do not silently alter the approved file. N5: If two versions are both labelled approved, ask the project owner which governs this delivery. N6: After delivery is completed, record its date and the person who performed it. This is training material only.

The source deliberately omits file formats, delivery software and response deadlines. A question should not require a trainee to guess those missing details. If an answer introduces a two-hour approval deadline, you should be able to reject that addition without arguing about whether two hours sounds reasonable.

Open notebook.google.com in a desktop browser and sign in. Choose the option to create a new notebook. In the source-adding window, use the copied-text route, paste the block and give it the title Delivery practice 01. Open the imported source and verify that N1 through N6 are present before generating anything.

A notebook is a project-specific collection of sources. Use only this source for the exercise, rather than adding old policies or unrelated files. The source maintenance guide covers a different task: updating an existing document and rerunning an answer. This exercise creates a new learning artifact from a deliberately bounded source.

How do you create the quiz in Studio?

Check that you can edit the notebook. Google's quiz instructions require edit access to generate a quiz. In the Studio panel, find Quizzes and open its pencil customization control before generation. Paste the following short brief into the available instruction field.

Create a short practice quiz using only Delivery practice 01. The audience is a new project coordinator. Use plain English scenario questions about missing approval, conflicting approved versions and a new client change request. Ask for the next action, not just memorised labels. Do not invent software, deadlines, file formats or new policies. Treat this as a training draft, not a competence assessment.

Choose a suitable starting difficulty and a smaller amount if the interface offers it. Google documents fewer, standard and more choices; none is a promise of exactly three questions. Start generation and open the result when it is ready. Our review scope is the first three questions, whatever the complete set contains.

If Quizzes is missing, inspect the account, notebook and your editing access. Do not continue as if an invisible control had worked. Do not purchase a plan merely because a button is absent. Your account may differ from a help screenshot, and this exercise makes no subscription entitlement or generation-time guarantee.

Record whether generation actually finished. A source added successfully is not a quiz generated successfully. A finished artifact is not a reviewed artifact. Keeping those stages separate prevents an optimistic status note from becoming an unsupported claim that the training is ready for the team.

How do you inspect the first three questions?

Read a question and locate its governing source rule before accepting the answer. Check whether only one option follows that rule. If two options express the same correct action in different words, the question is ambiguous. Record the problem rather than using its score to judge a trainee.

This is an author-created checking key, not output copied from a generated quiz. Your assistant may produce different questions. Match each of the first three to its actual N rule. When the source does not establish an answer, write unsupported by source instead of supplying a new policy yourself.

Use Explain when the answer offers that control, then compare the explanation with the source. Google's launch material describes explanations that point back to source information. A visible reference is still an invitation to inspect the evidence, not automatic approval. Read the sentence and any condition that changes its meaning.

For a very short source, Google's import guidance notes that references can point to the whole document rather than individual text. In that case, locate the N label manually. Do not invent a page number or claim that a citation highlighted a passage you did not actually see.

Give each reviewed question one of three statuses: accept, clarify or unsupported. Inspect the wording of the wrong options as well as the selected answer. A distracting option can accidentally state another valid source rule, leaving the learner unsure which part of the scenario matters. The question needs revision when that ambiguity affects the answer.

Short material can produce repeated or trivial questions. That is a limitation of the draft to record, not evidence that a colleague is unskilled. The first goal is to check question quality. A real assessment needs a broader method than counting correct answers in one automatically generated practice set.

What can you save and reopen within fifteen minutes?

1. Minutes one to three: define the learning decision and add the source. Expected output: a notebook containing all six N rules.
2. Minutes four to seven: customize and generate a short Studio quiz. Expected output: an artifact you can open.
3. Minutes eight to twelve: check the first three questions against the rules. Expected output: three review decisions with source labels.
4. Minutes thirteen to fifteen: save the review note and reopen the quiz. Expected output: a findable artifact and traceable checks.

This is a suggested work allocation, not a measured completion promise. If generation is still running, record source ready, quiz pending. If access fails, leave that stage blocked. Neither situation justifies marking a test reviewed or replacing missing work with an invented screenshot.

Open Notepad or another plain-text editor. Paste the notebook name, source title and version, the first three question texts, their proposed answers, supporting N labels and your decisions. Press Ctrl+S, choose Documents and save training-quiz-01.txt. Close the editor and reopen the file to confirm that those details remain.

Then reopen the notebook from its list and find the generated quiz in Studio. This is not a claim that the quiz was exported as a separate file. The text document you saved is your review note; this article does not promise CSV export for quizzes. If you inspected only three questions, label the rest unreviewed. If fewer than three were generated, record the incomplete review scope.

What remains before someone else uses the test?

Ask the procedure owner to approve the real source and the questions before team use. Decide what the practice is for: discussion, self-study or a supervised learning activity. Do not silently convert a small practice score into a hiring, promotion or certification decision.

In my earlier writing about effectiveness and perfection, I described projects delayed by pursuing every detail. Applied here, that principle means checking one short source and three questions before attempting a complete course. It is not a claim that I tested this workflow in a design studio or measured a training improvement.

For related text-task framing, see the study prompt guide. A spoken summary is a different artifact covered by the AI podcast guide. The artificial intelligence guide provides broader context. Today's completion check stays narrow: the quiz opens, three questions are checked, and unresolved problems are visible.