A Different Kind of AI Tool
NotebookLM is fundamentally different from ChatGPT, Claude, and Gemini.
General-purpose models like ChatGPT draw from their training data - everything they learned from the internet. They are broad but ungrounded. When they answer your question about nasopharyngeal carcinoma management, they are synthesizing from patterns across millions of web pages.
NotebookLM only references the documents you upload. That is it.
When you ask a question, it searches through your materials and gives you an answer with inline citations pointing to specific passages in your own documents. No hallucinated references. No information from outside your sources. Just your knowledge, organized and searchable.
What Makes It Different
NotebookLM only answers from the documents you upload. It does not hallucinate from training data. Every answer is grounded in your sources with inline citations you can verify.
This changes the game for building and maintaining a personal clinical knowledge base.
My Setup: Disease-Site Notebooks
I maintain separate notebooks for each major disease site I cover. Right now I have notebooks for nasopharynx, oropharynx, prostate, breast, CNS, lung, and GI - plus a few topic-specific notebooks for things like re-irradiation and brachytherapy physics.
Each notebook is a collection of documents related to that disease site. The content is intentionally broad:
- Key clinical trial papers (PDFs)
- NCCN and institutional guidelines
- My own clinical notes and summaries
- Lecture slides and textbook chapters
- Board review notes
- Interesting case discussions
I upload everything. NotebookLM does not care about formatting. Messy notes, annotated PDFs, copy-pasted text documents - it handles all of it. The AI parses whatever you give it and makes it searchable.
Chatting With Your Own Knowledge Base
This is where the real value lives. Once your documents are uploaded, you can ask NotebookLM questions in natural language and get answers grounded entirely in your materials.
Some examples from my actual use:
"What are the indications for postoperative radiation in nasopharyngeal carcinoma?"
NotebookLM gives me a synthesized answer pulling from the guidelines and trial papers I uploaded, with numbered citations pointing to the specific source documents and passages. I can click each citation and see exactly where the information came from.
"Summarize the key findings from the NRG HN002 trial."
If I have uploaded the paper, it gives me a focused summary drawn from the actual document. If I have not uploaded it, it tells me it does not have that information. No fabrication.
"What fractionation schemes have been used for re-irradiation in recurrent head and neck cancers based on my uploaded sources?"
It pulls from every relevant document in the notebook and synthesizes a comprehensive answer. This kind of cross-document synthesis is something that would take me thirty minutes of manual searching through PDFs. NotebookLM does it in seconds.
The key advantage over general AI: every answer is traceable. You can verify every claim against the source document. This is the grounding that general-purpose models lack, and it is what makes NotebookLM appropriate for clinical knowledge work in a way that ungrounded models are not.
AI-Powered Podcasts: The Feature I Did Not Expect to Love
NotebookLM has a feature called Audio Overview that generates a podcast-style conversation about your uploaded content. Two AI voices discuss the material in a conversational, accessible format - summarizing key points, highlighting interesting findings, and explaining complex concepts.
I was skeptical when I first tried it. Then I generated a podcast from my nasopharynx notebook and listened to it during my commute. It was genuinely useful.
Hearing the material discussed conversationally activated a different kind of processing than reading the same content, and it surfaced connections I had not noticed.
I now generate these audio summaries regularly, especially when:
- I am reviewing for tumor board and want a refresher on a disease site
- I have uploaded new papers and want to absorb the key points
- I am preparing for a lecture and want to hear the material organized differently
- I just want to stay sharp on a topic during my drive
Is it a replacement for reading the primary literature? No. Is it a remarkably efficient way to review and reinforce knowledge? Absolutely.
How to Get Started (10-Minute Setup)
If you want to try this yourself, here is the quickest path:
Step 1: Go to notebooklm.google.com and sign in with your Google account.
It is free. No subscription required.
Step 2: Create your first notebook.
Click "New Notebook." Name it after a disease site or topic you know well - you want to start with something where you can judge the quality of the AI's responses.
Step 3: Upload sources.
Click "Add Source" and upload documents. Start with 5-10 key documents: a few landmark trial papers, relevant guidelines, and any personal notes you have. Supported formats include PDFs, Google Docs, text files, web URLs, and more.
Step 4: Start asking questions.
The chat interface is intuitive. Ask a clinical question related to your uploaded content. Notice how the response includes numbered citations that link back to your source documents.
Step 5: Try the Audio Overview.
Look for the Audio Overview option (usually in the notebook's top menu or sidebar). Generate a podcast from your uploaded sources. Put on headphones and listen.
That is it. The whole setup takes less than ten minutes, and you will immediately feel the difference between this and a general-purpose chatbot.
Practical Tips From Regular Use
| Tip | Why It Matters |
|---|---|
| Keep notebooks focused | A notebook with 200 documents on every topic becomes less useful than 10 focused notebooks with 20 documents each |
| Upload liberally | Your quick notes from a conference talk, a screenshot of a key table, a copy-pasted guideline excerpt - all of it is useful |
| Use it for board prep | Upload your study materials for a disease site, then quiz yourself by asking questions and checking the cited sources |
| Combine it with general models | Use NotebookLM to get a grounded, cited answer from your own materials, then take that answer to Claude or ChatGPT for further reasoning or drafting |
| Refresh your sources | When new guidelines drop or a landmark trial publishes, add the document to the relevant notebook |
Why This Matters
The central challenge with AI in medicine is trust. General-purpose models are powerful but ungrounded - you never quite know where their information is coming from or whether it is accurate.
NotebookLM solves this for a specific and important use case: working with your own curated knowledge.
It is not trying to know everything. It is trying to help you work with what you already know and have collected. That constraint is its greatest strength.
The Real Value
For a radiation oncologist managing multiple disease sites, staying current on dozens of trials, and preparing for daily tumor boards, having an AI that can instantly search and synthesize your personal knowledge base is not a luxury. It is a genuine workflow improvement.
The Bottom Line
Try it. Start with one notebook, one disease site, ten documents. You will not go back to manually searching through PDFs.