NotebookLM is Google’s free AI research tool that answers questions only from the documents you upload, with a citation to the exact passage behind every answer. For lawyers, that means you can drop a whole case file into it and interrogate it like a paralegal who has read every page.
A lemon law case file at my firm is a pile of documents. Repair orders from three different dealerships. A purchase contract. Registration. Sometimes a deposition transcript and months of manufacturer correspondence. Somewhere in that pile sits the fact pattern that wins the case, and traditionally a paralegal spends hours digging it out.
NotebookLM does the digging in seconds. It’s free, it’s from Google, and most lawyers have never opened it.
What NotebookLM actually is
NotebookLM is Google’s research tool built on Gemini. You create a notebook, upload your sources (PDFs, Word docs, Google Docs, website links, even YouTube videos), and ask questions in plain English. Every answer is grounded only in what you uploaded, with numbered citations that link to the exact passage in the exact source.
That grounding is the whole product. A general chatbot answers from everything it absorbed in training, which is how lawyers end up sanctioned for citing cases that don’t exist. NotebookLM answers from your documents or it tells you the answer isn’t in them. It cannot cite anything you didn’t give it, so the fake-case-law failure mode is off the table.
For the task lawyers do most, extracting facts from a defined set of documents, that is a different risk profile than ChatGPT. Not a smarter model. A safer architecture.
What to do with it at a law firm
The core move: dump a case file into a notebook and interrogate it. Prompts my teams run, adapted for print:
“List every mention of the transmission issue, with the date and which document it appears in.”
“Compare the plaintiff’s deposition to the service manager’s deposition. Where do they contradict each other?”
“Build a timeline of repair visits with days out of service for each.”
Each answer comes back with citations. Click one and NotebookLM jumps to the passage in the source document. Verifying an answer takes seconds instead of another full read-through.
Past the single case file, three setups worth stealing:
Expert witness prep. Upload your expert’s report, their prior testimony, and the opposing expert’s report into one notebook. Ask where the two experts disagree and which assumptions are weakest. You walk into the deposition with a map.
A statute notebook per practice area. For lemon law that means Song-Beverly, Magnuson-Moss, and the state acts you file under most. Staff query the notebook instead of Googling, and the answer always cites the actual statutory text.
An onboarding notebook. Upload your firm’s SOPs, scripts, and training docs. New hires ask the notebook their first hundred questions instead of interrupting your best paralegal a hundred times. At 62 staff, this one pays for itself in a week, and it costs nothing.
Audio overviews: the case file as a podcast
NotebookLM can turn a notebook into an Audio Overview: two AI hosts discussing your material like a podcast episode. It sounds like a gimmick. It isn’t.
Upload a 200-page deposition transcript, generate the audio, and listen to a 15-minute discussion of it on the drive to court. The hosts surface themes and flag details you’d want to know cold. It also works for CLE-style learning: a new statute, a dense appellate opinion, a regulation you’ve been avoiding. Dead commute time becomes prep time.
The five-minute setup
- Go to notebooklm.google.com and sign in with a Google account.
- Create a new notebook and name it after the matter.
- Upload your sources. The free tier takes up to 50 per notebook.
- Ask questions in the chat panel. Citations appear inline.
- Pin useful answers as notes so they persist, and generate an Audio Overview from the studio panel if you want the podcast version.
No prompt engineering required. The questions you’d ask a paralegal are the questions you type. Save the prompt library for when you graduate to drafting.
Confidentiality and limits
Rules before client documents go anywhere near it:
Use a Google Workspace business account, not a personal Gmail. Workspace terms commit Google to not training on your content, and NotebookLM inherits those protections. A free consumer account gives you no contractual cover, so treat it the way you’d treat any unsigned vendor: fine for public statutes and test documents, wrong for client files.
Verify before you rely. Grounding cuts hallucination risk dramatically, but the model can still misread a document or merge two facts. The citation is right there. Click it.
Know what it isn’t. NotebookLM reads; it doesn’t draft. It will answer questions about your documents all day, but it won’t write a demand letter in your voice. For drafting, use Claude and feed it the facts NotebookLM pulled. It also can’t search the open web; when you need cited research beyond your own documents, that’s Perplexity’s job.
What it costs
Nothing, for most lawyers. The free tier covers 100 notebooks with 50 sources each, which is more case-file capacity than a solo will ever touch.
| Plan | Price | Notebooks | Sources per notebook | Extras |
|---|---|---|---|---|
| Free | $0 | 100 | 50 | Audio Overviews, citations, notes |
| Plus | Bundled with Google Workspace and Google One AI Premium | 5x | 5x sources and audio generations | Usage analytics, firm-wide notebook sharing |
Start free. Upgrade when a team is actually sharing notebooks daily.
The bottom line
- Answers come only from your documents, with a citation for each
- The free tier covers a solo firm’s entire caseload
- Workspace account for client files; personal Gmail for public documents only
- It reads, it doesn’t draft; pair it with Claude for the writing
Build one notebook today
Here’s the assignment. Pick your most document-heavy active matter. If you’re on a personal Google account, run a non-privileged test first: public filings, a statute, your own SOPs. Upload the documents, then ask the three questions you’d normally hand to a paralegal with a highlighter.
Compare the output to what an hour of manual review gets you. That comparison, on your own case, is worth more than anything I can write here. This tool sits early in the adoption roadmap for a reason: it works on day one, and it’s free.