How to use NotebookLM for research

Finding sources in Gemini Notebook

1. In the Sources panel, type your question in the search box.
2. Choose where to look: the web or your Drive.
3. Choose Fast Research or Deep Research, and press Enter.
4. Open the results, untick what you don't want and click Import.

Deep Research tells you that you can leave while it works, and it shows its progress step by step.

Google says Deep Research is for users over 18, and a large import can stop part-way at your usage limit. From Google's help page Add or discover new sources for your notebook, read October 30, 2026.

Fast Research vs Deep Research: one question, both ways

On October 30, 2026 we asked both for peer-reviewed evidence on whether retrieval practice improves long-term learning.

Read the list before you import. Two papers each appeared twice in Fast Research's 10 results, once from PubMed Central and once from the journal's site. So the list held 8 papers at most. Duplicates use up source slots and can make one study look like two.

One question on one day. We read the source lists and the opening of the report, but didn't check every claim in it, and you shouldn't take it on trust either.

Research papers and academic work

1. Make one notebook per question or literature review, not per course or project.
2. Add the papers as PDFs (up to 500,000 words or 200 MB each), from Google Drive, or by pasting a link to the PDF. Add the review articles first: they give the notebook the field's vocabulary.
3. Ask across the papers, then check every claim against its citation before you use it. For example: "Compare the methods and sample sizes of these studies in a table", "Where do these papers disagree, and why?", "Which findings rest on a single study?"
4. Save the answers you'll reuse as notes, and make a report or a data table in Studio when you need it in writing.

We tried this with two well-known machine learning papers, added as PDF links. Asked for a result that appears only in a table, the chat gave the right figures and separated the test score from the development score. Its citation opened the PDF with the table highlighted. A data table comparing the two papers came back with the right scores and a link to each source.

PDF links keep their web address as their name, so rename them. And keep unrelated material out: a notebook only knows what you gave it, but it will happily combine everything you gave it. In one test the chat added up figures from the wrong version of a course and gave a precise, wrong answer.

A gap in your reading also becomes a gap in its answers. KastorFlow's source audit shows which kinds of sources are missing, including academic ones, and suggests sources to fill them that you approve first (Pro+).

What the studies say about AI for research: AI for research statistics

A whole research notebook, with KastorFlow

1. In KastorFlow's dashboard, open Create and choose Build research notebook.
2. Type the topic. NotebookLM's own research engine finds the sources.
3. Review what it found and untick anything you don't want.
4. Import into a new notebook. NotebookLM's report becomes the briefing.

Free: 5 builds a month; Pro and Pro+: no monthly limit.

Going further

Learning a new field with it

1. Run Deep Research on the field and import the report with the sources you trust.
2. Ask for a map: the main questions, the schools of thought, the terms to know.
3. Audit it to see which kinds of sources are missing, and fill the gaps.
4. Make flashcards and a quiz in Studio, and review them over the following days.

Learning a new field with AI: what the research says

Questions