Gemini Notebook (Formerly NotebookLM), Explained
If you bookmarked NotebookLM at some point and went looking for it recently, you may have noticed the name is gone. Google renamed it Gemini Notebook in July 2026, a few months after launching "Notebooks in Gemini" in April, which made your notebooks show up inside the Gemini app and sync both ways. It also still exists as a standalone product (the old NotebookLM address redirects), for anyone who wants it separate from their main Gemini conversations. Same tool, new name, and now two doors into it instead of one, though the two doors are not identical.
That's a small piece of trivia on its own, but it's worth clearing up because Gemini Notebook does something genuinely different from a normal Gemini conversation, and the rename has made it harder for people to find. If you already know the basics of chatting with Gemini from the Complete Beginner's Guide to Gemini, this article is about the specific case where a regular chat isn't the right tool and Notebook is.
What actually makes it different
A regular Gemini conversation draws on its general training and whatever you paste or attach in that chat. Gemini Notebook flips the emphasis: you feed it a defined set of source material first, PDFs, Google Docs, slides, web pages, audio, video, and everything it tells you afterward is grounded in and cited back to those specific sources. It's built to stay inside the boundary of what you gave it, rather than blending your material with everything else the model knows.
That boundary is the whole point. Ask a regular Gemini chat to summarize a set of research papers you paste in, and you'll get a reasonable summary, but you can't always tell which claim came from which paper, or whether a detail crept in from the model's general knowledge instead of your actual documents. Gemini Notebook answers with inline citations pointing back to the exact passage in the exact source, so you can click through and check.
Source-grounded
Answers are tied to the documents you uploaded, with citations back to the specific passage.
Built for a defined body of material
Works best with a bounded set of sources: a course's readings, a project's documents, a research stack, not an open-ended topic.
Multi-format input
Accepts PDFs, Docs, slides, web links, audio, and video as sources, not just pasted text.
Generates study aids from your sources
Can turn your material into an audio overview, a study guide, quiz questions, or a mind map without you writing a separate prompt for each.
Gemini Notebook has also picked up a secure cloud computer for running code (rolled out first to Google AI Ultra and eligible Workspace customers, with Pro following), which sounds like a developer feature but matters for anyone whose "sources" include a spreadsheet or dataset rather than only prose. It means Notebook can actually run calculations against uploaded data, not just describe it in general terms, which is a meaningfully different capability from summarizing a PDF.
What the notebook actually looks like
The interface is built around three panels working together, not one long scrolling chat.
Gemini Notebook
Gemini Notebook's layout, illustrated
A sources panel runs down the left side, listing every PDF, Doc, slide, web page, or recording you've added to this notebook, each one toggleable on or off if you want an answer to draw from only some of them. The chat panel sits in the center, where you actually ask questions and get back grounded answers with inline citations. A generated-artifact panel sits on the right, holding whatever you've asked Notebook to produce, a study guide, an audio overview, quiz questions, a mind map, as its own standing document rather than something buried partway down the chat log.
- 1
Sources panel: Every document, slide, page, or recording in this notebook, with the ability to include or exclude individual sources per question.
- 2
Chat panel: Where you ask questions and get answers grounded in, and cited back to, the sources you selected.
- 3
Generated-artifact panel: Where a requested study guide, audio overview, quiz, or mind map appears as its own document, separate from the ongoing chat.
Grounded versus guessing, a concrete comparison
Say you upload three sources to a notebook: a peer-reviewed paper on urban heat island mitigation, a city planning department's public white paper listing a five-year tree canopy target, and a set of your own reading notes summarizing four other studies on reflective roofing materials. Then you ask a specific, bounded question.
According to these sources, what is the most cost-effective way for a mid-sized city with a limited budget to reduce urban heat island effects, and which source does that recommendation come from?
”Gemini Notebook
A grounded, citation-backed answer, illustrated
Based on these sources, reflective ("cool") roofing looks like the more cost-effective option over tree planting for a limited budget. The city's own white paper [2] projects its tree canopy target to take over a decade to produce a measurable temperature effect, while your reading notes [3] summarize two studies estimating a 2 to 3 degree localized surface temperature reduction from reflective roofing within a single budget cycle. The peer-reviewed paper [1] doesn't compare the two costs directly, but its modeling supports the reflective roofing figure.
Ask the same question in a regular Gemini chat with nothing uploaded, and you get something that sounds similarly confident but isn't tied to anything you can check.
The same question with no sources uploaded, illustrated
That second answer isn't wrong exactly, it's a reasonable general summary of the topic. But it isn't about your city's white paper, your specific canopy target, or the two studies your reading notes actually cite, it's the model restating common knowledge on the general topic. There's no way to check it against a specific document, and no way to know whether it's quietly missing the one detail your own sources disagree on. The Notebook answer is verifiable: click the [2] citation and you land on the actual paragraph in the white paper that supports the claim. That's the real difference grounding makes, not that the ungrounded answer is necessarily false, but that you have no way to confirm it's true for your situation rather than true in general.
A real example: turning a stack of PDFs into a study guide
Say you're prepping for an exam covering six weeks of a course, and you have the syllabus, four assigned readings as PDFs, and your own lecture notes as a Doc. Dropping all of that into one long Gemini chat works, but the conversation gets messy fast, and you lose track of which reading a given answer actually came from.
- 1
Create a notebook and add every source at once
Upload the syllabus, the four PDFs, and your notes doc as sources in the same notebook. Gemini Notebook treats all of them as one bounded body of material it can search across.
- 2
Ask for a structured overview first
Prompt“Based on these sources, list the major topics covered across this unit and roughly how much material each one represents. Group related readings together.
”This gives you a map of the material before you commit to a study guide format, so you can catch anything that's underrepresented or missing.
- 3
Generate the actual study guide
Prompt“Create a study guide organized by topic, not by source document. For each topic, include a plain-language explanation, one example from the readings, and two likely exam question types based on how this material is usually tested.
” - 4
Use the audio overview for review on the go
Notebook can turn the same sources into a spoken-conversation-style audio summary, useful for a commute or a walk when reading isn't practical. It's a supplement to the written guide, not a replacement for actually reading the source material once.
The result is a study guide that traces back to your actual course material instead of a generic overview of the topic pulled from the model's general knowledge, which matters if your professor emphasizes something idiosyncratic that a generic summary would never include.
When a regular Gemini chat is the better call
Gemini Notebook is not the right choice for everything, and reaching for it out of habit slows you down for tasks it isn't built for.
Use Gemini Notebook when
- You have a defined set of documents and want answers grounded only in them
- You need citations back to a specific source for accountability
- You want the same material turned into multiple formats: guide, quiz, audio
- The material is stable, not something you're still actively writing
Use a regular Gemini chat when
- You're brainstorming or don't yet have finished source material
- You want the model's general knowledge blended with your input
- The task is a single quick question, not a body of research
- You need Canvas-style collaborative drafting, not summarization
Limitations worth knowing
Notebook is only as good as the sources you give it. If a topic isn't covered in your uploaded material, it won't quietly fill the gap with outside knowledge the way a regular chat would, which is usually the point but can feel like a limitation if you expected a broader answer. There's also a practical ceiling on how many sources one notebook comfortably handles before organization starts to matter more than volume: past a certain point, splitting a sprawling project into a few focused notebooks by topic works better than dumping everything into one.
Tip
If you're not sure whether to use notebooks inside the Gemini app or the standalone Gemini Notebook site, they share the same notebooks and sync changes to sources and instructions. The Gemini app is more convenient if you're already living there. Per Google's help page, though, generating artifacts like Audio Overviews, infographics, and slide decks is only available in the standalone Gemini Notebook, so use that door for the study aids in this article.
Official sources
- NotebookLM is now Gemini Notebook
- NotebookLM is now Gemini Notebook
- Try notebooks in Gemini to easily keep track of projects
- About notebooks in AI Mode in Google Search
Checked on September 21, 2026. Features, plans and names change often, so the vendor's own pages are the final word.