Félicien Rops, Notebook with sketches and notes. The Metropolitan Museum of Art, public domain.
NotebookLM is a free Google tool that answers questions about documents you upload, citing which document each answer came from. It runs in a browser with no IT deployment and no procurement cycle.
The two kinds of knowledge a new hire needs. Procedural knowledge is how meetings run, which channel is used for what, who signs off on what. It can be documented, and in most firms it partly is. Contextual knowledge is why the strategy changed in Q2 2024, what happened with the client everyone goes quiet about, what the founder means by an internal shorthand. It is almost never written down, and the traditional way to acquire it is six weeks of bumping into it.
What we did. Each new hire got a notebook on day one holding four quarters of internal memos, the two most recent board packs, all product documentation, the org chart with role descriptions, and six months of transcribed All Hands recordings. Their first task was to write down every question they had and put it to the notebook instead of to a colleague.
The result. Across three hires, time to first meaningful deliverable went from an average of 42 days to 9. All three felt up to speed by the end of week two, against six weeks before.
What three people does not prove. No control group, and participants who knew they were in an experiment, which is the condition under which people work harder. This is an anecdote with numbers attached. The cost of being wrong is close to zero, which is a different standard from the one a study has to meet.
The finding that outlasts the number. The contextual layer was recoverable all along, sitting in memos and recordings nobody would think to search. Much of the six weeks went on interrupting colleagues for reference lookups, a cost carried by the people interrupted, which is why no onboarding measure ever showed it.
The last three people who joined us got a notebook before they got a laptop bag. On day one, before any introductions, each was asked to write down every question they had about the firm. They spent the afternoon putting those questions to a NotebookLM notebook instead of a colleague, and built a personal reference document from the answers, each with a citation to the document it came from.
Time to first meaningful deliverable across those three hires averaged 9 days. Our prior average was 42. All three said they felt up to speed by the end of week two, against six weeks before.
What was in the notebook
The upload list matters more than the tool, so here it is in full:
- the last four quarters of internal memos;
- the two most recent board packs;
- all product documentation;
- the full org chart, with role descriptions rather than just names and titles;
- six months of All Hands recordings, transcribed.
The org chart with role descriptions did more work than I expected. Many new-hire questions are about routing: who decides pricing, and who has to agree before it changes. Documents can answer those, and they are the questions people are most reluctant to ask twice.
Procedural knowledge is written down; contextual knowledge sits in old memos
Onboarding a knowledge worker at a mid-sized firm has taken about six weeks for as long as I have been hiring, and the time splits into two parts.
The first part is procedural: how meetings run, which channel is used for what, who signs off on what. Most firms have some of this documented, usually in a handbook nobody reads because it is organised by topic, and new hires arrive with questions.
The second part is contextual: why the strategy changed in Q2 2024, what happened with the client everyone goes quiet about, what the founder means when she says "Aunt Millie's rule." Nobody writes that layer down, and new hires usually pick it up over six weeks of coffee conversations and half-overheard Slack threads.
What surprised me is how much of the contextual layer was recoverable. The memos, board packs and All Hands recordings where those decisions were first explained still held the reasoning. It was spread across forty documents that no new hire would think to search.
The experiment, and its limits
- 3 hiresTotal sample. No control group, and the participants knew they were in an experiment
- 42 to 9 daysAverage time to first meaningful deliverable, before and after
- 2 weeksWhen all three reported feeling up to speed, against roughly six weeks previously
- 5 source typesMemos, board packs, product documentation, annotated org chart, transcribed All Hands
- $0Cost. The free tier covers this use, and there is no deployment step
What three hires can and cannot show
Three hires at one firm, with no control group, is an anecdote. The people involved knew they were in an experiment, which tends to make people work harder. Our 42-day baseline comes from a small number of earlier hires in different roles. Any one of those would disqualify this as evidence in a paper.
I keep doing it because the direction was the same for all three and the cost of being wrong is close to zero. That is a lower bar than a study has to clear, and the right bar for an operating decision at our size. I would not present this to a board as a productivity finding. I would present the method and ask them to run it on their own next hire.
Lookups can be handed to a notebook; judgement cannot
Most of the six weeks went on asking colleagues basic questions while they tried to do their own jobs. The people being interrupted carried that cost, so it never showed up in any onboarding measure. NotebookLM did not make anyone smarter. It removed a coordination cost nobody had counted.
The same applies wherever one person interrupts another to retrieve context: client history before a renewal call, earlier decisions before a design review, precedent before a pricing conversation. The test is whether the question is a lookup or a request for an opinion.
It breaks on the opinion half. A document-grounded tool answers a judgement question with the same confidence as a factual one, and a citation does not make the judgement sound. Keep it on lookups.
What to do next
- Hiring manager, before the next start date: define in writing what "first meaningful deliverable" means for the role, and record how long it took your last two or three hires, so you have a baseline.
- Operations or HR lead, one week before day one: build the notebook at notebooklm.google.com from the five source types above. Remove or date-label any superseded memo. Allow about half a day.
- New hire, day one: write the full question list before anyone answers anything informally, then put each question to the notebook. Once questions are asked out loud, you lose the measurement.
- Hiring manager, end of week one: spend an hour going through what the hire learned. Correct the confident wrong answers, which come from superseded documents and arrive with a citation that makes them persuasive.
- Hiring manager, same meeting: list the questions the notebook could not answer. That list shows where your firm's undocumented knowledge sits; assign an owner to write down the top five.
Frequently Asked Questions
Is it safe to put board packs and internal memos into NotebookLM?
Use a Google Workspace work account, not a personal Gmail account, because Workspace accounts come under your organisation's data protection terms with Google, which exclude your content from model training. Your Workspace administrator may need to switch NotebookLM on. Read Google's current privacy terms for NotebookLM before uploading, and leave out anything containing personal data you have no basis to share, such as salary files or medical notes.
How many documents can one notebook hold?
At the time of writing the free version allows around 50 sources per notebook, with a size limit on each source, and the paid tier raises both. Four quarters of memos can exceed that if every memo is a separate file, so combine them into one PDF per quarter. Check Google's help pages for the current limits before you build.
Do the All Hands recordings need to be transcribed first?
Not necessarily. NotebookLM accepts audio files and YouTube links as sources and transcribes them itself. We uploaded transcripts because they were easier to check and to cite, and a transcript lets you remove confidential segments before upload.
How do we keep the notebook from going stale?
Give it an owner and a quarterly review date. At each review, add the latest memos and board pack, delete or mark superseded documents, and add any answers written up from the unanswered-questions list. A notebook that nobody maintains starts giving outdated answers with citations attached within a year.
Can several new hires share one notebook?
Yes. NotebookLM lets the owner share a notebook with named people, so one maintained notebook can serve every new hire. Each person should still keep their own reference document from their questions, because writing it is part of how they learn.