Using NotebookLM to Synthesize a Reading List

What this tutorial covers

You have 30 (or 80, or 200) PDFs from a literature search and need to understand the landscape before you can write a coherent review section. Reading everything fully before you know what matters is inefficient. NotebookLM lets you upload your papers and query across them conversationally — asking “what do these papers say about X?”, “which papers disagree on Y?”, “what methods are most common?” — with cited responses pointing back to specific passages.

This tutorial covers: uploading papers, writing effective queries, building a structured synthesis, and knowing when to go back to the source.

Step 1: Curate your input set

NotebookLM works best with a focused, curated set rather than everything you’ve ever downloaded.

Target: 20–50 papers on a specific question or sub-topic. If you have hundreds of papers, break them into thematic batches.

Get your PDFs from:

Semantic Scholar — download PDFs directly for open-access papers
Zotero — export a collection folder as PDFs
Your institution’s library access for paywalled papers

Before uploading: Skim abstracts to remove papers that are clearly off-topic. NotebookLM’s answers are only as good as what’s in the notebook — including irrelevant papers introduces noise.

Step 2: Create a notebook and upload sources

Go to notebooklm.google.com and sign in with a Google account
Click New notebook
Upload your PDFs via Add sources → Upload
- NotebookLM accepts PDFs, Google Docs, Google Slides, and pasted text
- Free tier: up to 50 sources per notebook, 25MB per source
Wait for processing (usually under a minute per source)

Tip: Name your notebook specifically (“Federated Learning in Clinical AI — July 2026”) so you can return to it later.

Step 3: Start with orientation queries

Before asking specific questions, orient yourself with broad queries:

“What is the main argument or contribution of each source?”
“What are the most common research methods across these papers?”
“What questions do these papers leave unanswered?”
“Which papers are most frequently cited by others in this collection?”

NotebookLM will respond with cited answers — each claim links back to a specific passage in a source. Always click through to verify the citation actually says what the summary claims.

Step 4: Ask targeted synthesis questions

Once you have a sense of the landscape, move to more specific questions:

For finding consensus:

“What do these papers agree on about [topic]?”
“What is the most commonly reported finding about [outcome measure]?”

For finding disagreement:

“Which papers disagree on [mechanism/finding]? What explains the disagreement?”
“Are there papers that challenge the dominant view on [topic]?”

For methodological mapping:

“What study designs are used? Are any methods dominant?”
“What datasets or benchmarks are most commonly used?”
“What limitations do authors most often acknowledge?”

For gap identification:

“What do the authors say about future research directions?”
“What questions are raised but not answered in these papers?”

Step 5: Build a structured synthesis document

Use NotebookLM’s Note feature to save key responses as you go:

For each major theme you identify, ask a targeted question and save the response as a note
Label notes by theme: “Methods,” “Key findings,” “Gaps,” “Definitions”
Use the Studio feature to generate a structured summary document from your notes

The result is a working synthesis you can paste into your literature review draft — with citations attached at the passage level, so you can trace every claim back to its source.

Step 6: Identify gaps and go back to the source

NotebookLM synthesizes across what you uploaded — it cannot tell you what’s missing from your collection. After building your synthesis:

Return to Semantic Scholar or your database of choice to search for papers covering gaps you’ve identified
For any finding that will appear in your manuscript, read the original passage in the source PDF, not just the NotebookLM summary
Cross-reference key claims against a second source — NotebookLM can mischaracterize a paper’s position, especially on nuanced points

Common mistakes to avoid

Uploading too broadly. A notebook with papers on five different sub-topics produces muddled answers. One focused question per notebook works better.

Trusting summaries without checking citations. NotebookLM hallucinates less than a raw LLM, but it can still misrepresent what a paper says. Always verify claims that matter before putting them in your manuscript.

Using it as a substitute for reading. NotebookLM is excellent for orientation and for deciding which papers to read carefully. It is not a substitute for reading the papers that become central to your argument.

Not saving your queries. Good queries are reusable. Keep a running list of the prompts that produced useful responses — you’ll want them when you update your review.