Grounded AI for literature reviews and meeting notes. It only ever answers from the sources you give it.
Immunology, cancer biology, infectious disease. Different questions, same two problems.
More papers than any of us has time to read, in every subfield.
Zotero, Mendeley and EndNote organise your library. They don't help you interrogate it.
Supervision and collaboration meetings fade into memory, or into scattered voice memos.
Gemini Notebook has become how I close both. It's not a replacement for reading. It's a way to question a curated set of sources.
A Google tool, built on Gemini, that answers only from the sources you upload into a notebook. Ask questions in plain language and it retrieves the relevant passages before answering (retrieval-augmented generation, or RAG).
PDFs · audio · video · docs
Upload what you trust. Caps scale with plan.
Retrieves the relevant passages
Not the open internet. Not its training data.
With inline citations
Every claim traceable back to a source.
Recently renamed. Google renamed NotebookLM to Gemini Notebook on 16 July 2026. Same product, same notebooks, same links. notebooklm.google.com still works.
Source caps per notebook: 50 (free) · 100 (Plus) · 300 (Pro) · 600 (Ultra). Screenshot taken 28/07/2026.
Why the risk drops
Every claim comes with an inline citation to the exact passage and source it used.
It's restricted to your uploaded sources rather than the open web or training data, so hallucination risk drops sharply versus a general chatbot.
Still on you
It can't tell a good source from a bad one. A flawed or biased paper gets reflected back faithfully.
Grounded ≠ correct. It's only as good as what you feed it, so click through the citations for anything you'll rely on.
Store and organise references.
Ask questions across them, grounded in the text.
What do these sources agree and disagree on? What methods have been used to measure X? What hasn't been shown yet?
Voice memo on your phone during the meeting.
Drop the audio file straight into a notebook.
Decisions, action items and topics, weeks later.
Confidential material: check what's appropriate to upload before you record sensitive conversations. More under Where to be careful.
A cross-cutting topic, on purpose.
A handful of real papers are loaded into a notebook ahead of time. The full reference list is in the appendix. I'll ask a couple of questions live and we'll look at where the citations land.
Rebuilding this notebook yourself takes about five minutes with the appendix list.
Two on the day. Hover any block to copy it.
Which bacterial genera are most consistently linked to response to immune checkpoint inhibitors across these papers?
Do the sources agree on whether antibiotics before immunotherapy hurt outcomes? Where do they disagree?
What's the proposed mechanism linking gut bacteria to anti-tumour immunity? Summarise the evidence on faecal microbiota transplantation for overcoming checkpoint-inhibitor resistance. Cite each claim. Based only on these sources, what's still unresolved in this field?
Data & privacy. Uploaded sources go to Google's servers under your Google account. Before uploading, check what's appropriate for unpublished data, patient information, or confidential meeting content, and check your institution's data-handling guidance.
notebooklm.google.com, with any Google account.
One notebook, 8–12 well-chosen sources. Not everything you own.
Audio overviews, mind maps and briefing docs, all generated from your sources.
Your next meeting recording, or the reading list for a talk you're prepping.
Presenting in this series yourself? Slots are bookable through the ECR Skill Swap Bookings page. Short and practical beats polished and long.
The six papers loaded into the demo notebook. Retrieved via PubMed.