AI Learning Hub › Events › Past events › Identifying Trends with NotebookLM
Past session · 17 October 2025
Identifying Trends in a Corpus of Documents Using NotebookLM
Your next insight is buried somewhere in your documents. This is the do-it-yourself version of the workshop on turning a pile of them into a knowledge base you can question — what NotebookLM is for, where it fails, and the limits worth knowing before you start.
- Presented
- Friday, 17 October 2025, on Zoom
- Presented by
- Heather Joy (hnjoy@law.stanford.edu), Reference and Student Scholarship Librarian, Robert Crown Law Library
- Format
- Slides with speaker notes, plus the limits and cautions written out below
This was accurate when presented. NotebookLM changes often, and the plan limits below are the ones that applied in October 2025 — check the current ones before you rely on a number.
Three terms
What NotebookLM is
An AI research assistant from Google that is grounded in the sources you provide. Unlike a general-purpose chatbot it answers from the documents you upload and cites them, which cuts out one whole category of hallucination — though not all of them.
What we mean by trends
- A preemption check
- Judicial trends across a set of opinions
- Patterns across interviews
- How a term is most often defined
- Where viewpoints diverge
What we mean by a corpus
- Litigation documents
- Judicial opinions
- Interview transcripts
- Scholarly articles
- A set of statutes or regulations
What the session covers
1. Explain
Say what NotebookLM is, and how being grounded in your sources changes what it can be trusted with.
2. Evaluate
Weigh the pros and cons for analyzing a specific corpus — including whether to use it for this task at all.
3. Apply
Build a notebook from your own documents and work out the practices that suit the analysis you need.
The presentation
Open the slides and follow along; the explanations are in the speaker notes.
For the speaker notes, use the three-dot menu in the player and choose Open speaker notes.
Top tips
Reach for it when
- You have a specific task in mind, not a vague hope
- You have a defined set of material to work from, in one format or several
- The material is not confidential
- You want to lower the risk of hallucination
- You want output in more than one form
What you can upload
PDFs, Google Docs, Google Slides, plain text, URLs that contain text, audio files, and links to videos that have captions.
The plan limits as of October 2025
- Free: 100 notebooks, up to 50 sources each; each source capped at 500,000 words or 200 MB. Fifty chat queries, three audio and three video overviews, ten reports, ten quizzes, ten flashcard sets per day.
- Paid: 500 notebooks, up to 300 sources each. Five hundred chat queries, twenty audio and twenty video overviews, and a hundred each of reports, quizzes, and flashcard sets.
- A day means twenty-four hours, not a calendar day — you cannot make three videos at 11:30pm and three more after midnight.
Where it falls down
- It misses sarcasm and tone
- It cannot read facial expressions in video
- Hallucinations still happen — usually as wrong or confused data drawn from your own uploads, or as formatting problems
- It does not always do what you asked, so check the output every time
Where to go next
- Try NotebookLM (opens in a new tab) with your own sources.
- Google’s help center (opens in a new tab) for the official how-to and the current limits.
- Part eight of the beginners course covers preparing the documents you are about to upload, which is most of what makes this work.
Privacy and confidentiality. Grounded in your sources still means your sources went somewhere. Check how the tool treats your data before uploading, keep confidential material out, and if you would not want it on the front page of the newspaper, do not upload it.
Questions? Bring them to the AI Curiosity Corner, or email library@law.stanford.edu. All past sessions are in the archive.