How to Use NotebookLM for a Literature Review
NotebookLM is now Gemini Notebook. Fit 40 papers under the free 50-source cap, name your sources, and see where Google's docs go quiet.
Forty PDFs, one deadline, and a free Google account that stops you at 50 sources per notebook. That ceiling shapes every decision below, so start there rather than at the upload button.
A naming note first, because it will confuse you otherwise. On 16 July 2026 Google renamed NotebookLM to Gemini Notebook. The announcement post is headed "NotebookLM is now Gemini Notebook", the help center masthead now reads "Gemini Notebook Help", and the app sits at notebook.google.com. Your notebooks and sources carry over. Old help-centre URLs under support.google.com/notebooklm/ still resolve, and they serve pages that say Gemini Notebook throughout. This guide uses Google's current on-screen wording, since that is what you will be clicking.
Forty papers fits under the free 50-source cap with ten slots spare
Google's Upgrade Gemini Notebook page publishes the tier limits in a table. The numbers that decide whether a literature pile is workable:
| Limit | Standard (free) | Plus | Pro | Ultra (30 TB) |
|---|---|---|---|---|
| Notebooks | 100/user | 200/user | 500/user | 500/user |
| Sources per notebook | 50/notebook | 100/notebook | 300/notebook | 600/notebook |
| Chat queries | 50/day | 200/day | 500/day | 5K/day |
| Reports | 10/day | 20/day | 100/day | 1K/day |
| Deep Research | 10/month | 3/day | 20/day | 200/day |
Fifty sources is a per-notebook figure, not an account figure, and free accounts get 100 notebooks. So the constraint is not "how many papers can I hold" but "how many papers can I hold in one conversation". One notebook per research question beats one notebook per project, and it is the difference between a chat that finds the right paragraph and a chat that averages forty papers into mush.
Per source, the FAQ is specific: "The current limit is 500,000 words per source or up to 200MB for local uploads. There's no page limit." A journal article will not come close to either. The same page lists exactly three documented import failures: the source exceeds 500,000 words, the file exceeds 200MB, or "Your original PDF file is copy-protected." That third one catches people with library-DRM downloads, and no setting inside the notebook fixes it.
The chat cap matters more than the source cap once you are working. Fifty queries a day across forty papers is roughly one question per paper, per day. Plan the questions before you spend them.
Load the pile: local upload once, or a Drive folder that keeps syncing
The sources documentation lists what Gemini Notebook accepts: "Microsoft Word (docx), Text (txt), Markdown (md), PDF files (pdf), CSV (csv), and PowerPoint (pptx) files", plus web URLs, ePub files, public YouTube URLs, copy-and-pasted text, images, audio, and Google Drive files. Google Slides are capped at 100 slides and Google Sheets are "limited to 100k tokens".
Two routes in, and they behave differently after import.
- Local upload. Open notebook.google.com, select Create new notebook, then Upload a source in the pop-up. Add the rest later with + Add in the Sources panel. A locally uploaded PDF is a snapshot. Google's documentation describes no sync path for it.
- Google Drive import. Put the forty PDFs in one Drive folder first, then import them. The docs are explicit about what you get: "Sources imported from Google Drive are auto-updated and will sync every few minutes. Changes to your original document will automatically update when you open your Notebook. If needed, you can manually update a source by opening the source in Notebook and clicking 'Click to sync with Google Drive.'"
Drive import has a failure mode worth knowing before it bites. "If you lose access to a file in Google Drive or if the file is deleted, the source will be inaccessible", and dead sources are not free: "Inactive sources will count towards source limits but will not be referenced throughout your notebook." A shared folder that someone reorganises can silently eat slots in your 50.
One limitation lands hard on academic reading. Gemini Notebook "does not import footnotes or comments from Google files." In a lot of humanities and legal scholarship the footnotes carry half the argument. If your sources are Google Docs, convert to PDF before importing, or accept that you are reading a version of the paper with its apparatus stripped out.
Rename every source before you ask a single question
This is the five minutes that pays back hardest in the whole workflow, and the reason is in Google's own guidance: "When multiple sources are selected, mentioning source names in your query helps Gemini Notebook narrow its search. For example, instead of 'Summarize this source,' try 'What are the key findings about dog training in the "Dog Training 101" document?'"
You cannot name a source in a query if all forty sources are called 1-s2.0-S0140673620311806-main.pdf. Rename each to something you would actually type: Mehra 2020 hydroxychloroquine registry, Kording 2017 microprocessor critique. Author, year, three words of subject.
The Sources panel also carries a checkbox per source. The docs describe it plainly: "In your notebook, you can use the checkbox on each source to include or exclude certain sources the model should use to answer your question." Deselecting thirty-five papers to interrogate five is the closest thing the product has to a scalpel.
Run one triage pass across all forty, then stop
The mistake is opening the notebook and free-associating until the daily query cap runs out. Spend the first pass building a comparison grid instead, with all sources selected. One prompt, worth pasting verbatim:
For every source in this notebook, give me one row with: (1) the exact research question the authors state, in their words; (2) study design and sample size, or "not stated"; (3) the single headline finding, with the effect size or statistic if the paper gives one; (4) the limitation the authors themselves name first; (5) the strongest claim this paper makes that a later paper could contradict. Use "not stated in the source" wherever the source does not say. Do not fill gaps from general knowledge.
Two things make that prompt work rather than the vague version. It asks for the authors' own framing, which forces quoting over paraphrase. And it supplies an explicit escape hatch, so the model has somewhere to go other than inventing a sample size. The Sources documentation notes the same split between summary modes: you can "Ask for a summary of specific topics from your source directly in chat" or "Find an auto-generated summary of the entire source in the Source Guide." The Source Guide summary is free of your query budget. Read it for orientation, then spend queries on comparison, which is the thing forty papers are for.
Save the grid. In the Studio panel, choose Save to Note on the response. Google's notes documentation confirms what survives: "When you save a response as a note, the original format, including tables and clickable inline citations, gets saved." You get up to 1,000 notes per notebook, and saved response notes "aren't editable once created."
An in-line citation proves location, not support
Gemini Notebook attaches citations to chat answers, and the mechanism is documented: "Gemini Notebook uses direct quotes, text, and images straight from your sources as citations to answer your questions and perform actions." Hovering shows the quoted text. Selecting a citation "automatically navigates to the location of the quote, so you can easily view it in context."
Read that carefully, because it is narrower than it sounds. A citation demonstrates that a passage exists at that location in a paper you uploaded. It does not demonstrate that the passage supports the sentence the model built around it, and Google claims no such thing anywhere in the help centre.
There is also a documented case where you get no per-passage citation at all: "If your source content is too short, Gemini Notebook references the entire document without a cited individual text from your source." A one-page conference abstract can produce a whole-document reference that looks like a citation and locates nothing.
Four things Google's documentation does not tell you
Which model is answering. The Upgrade page has a row labelled "Gemini Model" whose cells read "Access to Gemini models", "Higher access to Gemini models" and "Highest access to Gemini models". No version number, anywhere in the help centre. Which means you cannot map behaviour you observe to a model card, and you cannot know when it changes under you.
How much of your corpus reaches the model per question. The FAQ confirms retrieval happens: "When your notebook contains many sources, Gemini Notebook retrieves the most relevant information based on your question, then builds a response from it." It publishes no chunk count, no passage budget, no ranking method. A question that returns nothing may have missed retrieval rather than missed the evidence, and there is no documented way to tell the two apart.
How the citations are chosen. The docs say citations are direct quotes from your sources. They do not say whether the quote was retrieved and then written around, or selected after the sentence was drafted. That distinction decides how much weight a citation carries, and it is not on the page.
Whether any of this is stable. The Upgrade page footnotes its own table "Usage Limits (Subject to Change)", and the whole product changed name three weeks ago.
Google's own hedge is on the overview page, so take it at face value: "Gemini Notebook can make mistakes and its answers don't reflect Google's views. Always consult a qualified professional for medical, legal, or financial advice."
Forty papers in one notebook will get you a defensible map of a literature in an afternoon. Every claim on that map still has to be opened, read at the cited location, and checked against the paper it came from before it reaches your draft.
Changelog (1)
- July 31, 2026 โ First published.