Can You Use NotebookLM to Write a Literature Review? (2026)

NotebookLM, defined: NotebookLM is Google’s source grounded research assistant. A user uploads their own documents, and NotebookLM answers questions using those documents, with in-line citations pointing back to the passage each answer came from. Google documents the supported uploads as "PDFs, websites, YouTube videos, audio files, Google Docs, or Google Slides" (Google, Learn about Gemini Notebook).

What does NotebookLM actually do with the PDFs I upload?

NotebookLM reads the uploaded files and restricts its answers to them, attaching an in-line citation to each claim so the researcher can click back to the sentence it came from. That is the whole mechanism, and it is the reason NotebookLM behaves differently from a general chatbot: the answer is bounded by the sources sitting in the notebook rather than by the open web.

Two caveats come from Google itself. The first is accuracy: the product "can make mistakes and its answers don’t reflect Google’s views". The second is a quiet one about citation granularity, from the product FAQ: "If your source content is too short, Gemini Notebook references the entire document without a cited individual text from your source" (Google, Gemini Notebook frequently asked questions). A short source, in other words, gets cited as a whole document, not as a line you can check in ten seconds.

One naming note, because it causes confusion when searching for help. As of September 2026 Google’s help centre for this product is published as Gemini Notebook, with the overview article titled "Learn about Gemini Notebook" and the FAQ written throughout in the same name, while the help URLs and the name most researchers still use remain notebooklm. If a support page calls it Gemini Notebook, you are in the right place.

How many papers can I load into one NotebookLM notebook?

Fifty, on the free tier, as of September 2026. Google’s limits page sets sources per notebook at 50 on the free plan, 100 on Plus, 300 on Pro and 500 or 600 on the two Ultra tiers, with 100 notebooks per user on free and 500 on Pro (Google, Gemini Notebook plan limits).

Fifty sources is a comfortable ceiling for a taught masters review, where the reading list is usually in the twenties or thirties. It is tight for a systematic review, where the screening set runs into the hundreds before exclusion, and the screening work has to happen somewhere other than NotebookLM first. Litrevu has a separate walkthrough of how to screen a large abstract set down to a readable pile.

Has anyone actually tested NotebookLM on a literature review?

Yes. Shor, Greene, Sumberg and Weingrad published an evaluation in the journal Psychiatry in 2026 (volume 89, issue 1, pages 82 to 91, doi:10.1080/00332747.2025.2541531). The team uploaded 22 papers from the Army Study to Assess Risk and Resilience in Servicemembers and put a series of literature review questions to NotebookLM about a hypothetical research paper.

Their finding is the useful part for anyone deciding whether to lean on the tool. The authors report that "the variability and utility of responses" were "determined in large part by the ability to write meaningful prompts and the extent to which new prompts provided additional information", and conclude that the utility of NotebookLM "will likely vary by the quality of source material uploaded into the program and the researcher’s familiarity with prompt generation". That is a study of 22 papers and one research team, not a benchmark, and it is worth reading as a description of the workflow rather than a score.

Where does NotebookLM stop and a literature review start?

NotebookLM answers one question at a time. A literature review is a single piece of connected prose that groups studies by theme, states where they agree and where they contradict each other, and names the gap the new research will fill. Those are different units of work: twelve good answers about twelve papers are not a themed synthesis, and assembling them into one is the part a marker is actually grading.

The 2026 evaluation of NotebookLM by Shor and colleagues in the journal Psychiatry describes the same shape from the inside, reporting that the team reached a usable mock review through "the iterative refinement of output" across repeated prompts rather than through a single request. If you want the mechanics of the synthesis step itself, Litrevu has a step by step guide to synthesising sources rather than summarising them one by one.

The other thing NotebookLM does not produce is a reference list in your required style. In-line citations inside a chat window point at your uploaded file, which is exactly what you want while reading, and it is not a Harvard or APA reference list you can paste into a chapter.

Is it safe to upload papers and unpublished drafts to NotebookLM?

Your own published PDFs, generally yes. Someone else’s unpublished manuscript, often no, and the rule is not Google’s. Elsevier’s generative AI policy states that "Reviewers should not upload a submitted manuscript or any part of it into an AI tool as this may violate the authors’ confidentiality and proprietary rights", and applies the same sentence to editors (Elsevier, generative AI policies for journals). If you are peer reviewing while doing your own postgraduate work, that manuscript stays out of any notebook.

On Google’s side, the product FAQ states "Your data will never be shared by Gemini Notebook". Note what that sentence covers and what it does not: it is about sharing, and your institution may still have its own rule about uploading supervisor drafts, participant data or anything under an ethics condition. Elsevier’s policy also says authors "should disclose the use of AI tools for manuscript preparation in a separate AI declaration statement", so the disclosure habit is worth building early, and Litrevu has a template for writing an AI-use disclosure.

NotebookLM, Elicit, Consensus, Scite, Research Rabbit or Litrevu: which one do I need?

Pick by the stage you are stuck on, because these tools do not overlap as much as their marketing suggests. Litrevu is an AI literature review assistant that turns the papers a researcher has already gathered into a cited first draft, with every citation traceable to the uploaded source.

Read that table as capability rather than price, and note where the honest losses are. NotebookLM beats Litrevu on interrogating a mixed pile of material, including videos and web pages, and it is free to start with a Google account. Elicit beats Litrevu at the search and screening end, since Litrevu starts from papers you have already chosen. Research Rabbit finds adjacent work that neither of the others will surface.

What does a cited first draft actually look like?

It looks like prose with the citations already attached, grouped by theme rather than by paper. The block below is a format illustration, not a real Litrevu output, so the citations are placeholder tokens rather than invented authors:

Theme 2: Measurement of adherence

Self-reported adherence remains the most common measure across the reviewed studies (Author, Year; Author, Year), although both papers note the tendency of self-report to overstate compliance. Pill-count methods appear in a smaller group of studies (Author, Year), where the authors argue that the discrepancy with self-report is large enough to affect conclusions. No study in this set compares the two methods in the same cohort, which is the gap this review takes forward.

Every (Author, Year) token in a real draft resolves to one of the PDFs you uploaded.

The shape is the point: themed heading, claim, citation attached to the claim, and a stated gap at the end of the theme. Litrevu also has a longer worked walkthrough of turning a folder of papers into a review, and a comparison of which AI tools cite sources that actually exist.

How do I use NotebookLM and still write the review myself?

Use NotebookLM as the reading and interrogation layer, then write from your own notes. A workable sequence: upload the papers you have already selected, ask one question per theme rather than one question per paper, click through every in-line citation to the passage it names, and keep your own note of which paper supports which claim before any drafting starts.

The verification step is not optional, and Google says so: the tool "can make mistakes and its answers don’t reflect Google’s views". The check that catches the most is the dullest one, which is opening the cited PDF and confirming the sentence says what the summary claims it says, page by page. Litrevu has a longer guide to spotting fake or mismatched AI citations before you submit.

Frequently asked questions

Can NotebookLM write my whole literature review for me?

No. NotebookLM answers questions about documents you upload and returns those answers with in-line citations, as Google’s help centre describes it. NotebookLM does not produce a themed synthesis or a reference list in a required style, and anything it does produce needs checking against the source before it goes near a chapter.

Does NotebookLM make up citations?

NotebookLM answers from the sources sitting in your notebook rather than from the open web, which removes the most common form of invented citation, the reference to a paper that does not exist. NotebookLM can still misread what a passage says, and Google states plainly that the tool "can make mistakes", so each in-line citation needs a click through to the passage.

How many PDFs can I upload to NotebookLM?

As of September 2026, Google’s limits page lists 50 sources per notebook on the free tier, 100 on Plus, 300 on Pro and 500 or 600 on the Ultra tiers. Free users also get 100 notebooks. For a screening set in the hundreds, the screening has to be narrowed before the papers reach a notebook.

Is it against the rules to use NotebookLM for my dissertation?

That depends on your institution, and most universities now ask for disclosure rather than banning AI tools outright. For journal submissions, Elsevier says authors should disclose AI use in a separate declaration statement, and separately tells reviewers and editors not to upload submitted manuscripts into AI tools at all.

What is the difference between NotebookLM and Litrevu?

NotebookLM answers questions about a mixed set of uploaded material, including web pages and videos, and gives you a grounded chat history. Litrevu takes the papers you have already chosen and drafts a structured, cited literature review from them, with each citation traceable to the PDF you uploaded.

Can I use NotebookLM for a systematic review?

Partly. The 50 source ceiling on the free tier sits well below a typical systematic screening set, so the search, screening and exclusion stages have to happen elsewhere and be documented in a PRISMA flow diagram. Once the included set is final and small enough, a notebook is a reasonable place to interrogate it.