Claude Projects vs NotebookLM: Which Fits Your Research?

If your research question is “what do these sources say?”, NotebookLM is the closer fit: its chat answers only from the sources you add, cites the passage behind each claim, and turns the same sources into Audio Overviews, reports, mind maps and flashcards. If your question is “what should we do with this?”, a Claude Project is: it starts every chat from your files and standing instructions, but Claude also brings its own knowledge, can search the web and connect to other tools, and helps you draft, analyse and decide. Many people use both, one to understand the material and the other to work with it.

One naming note first. Google renamed NotebookLM to Gemini Notebook in July 2026. Its announcement, NotebookLM is now Gemini Notebook↗ (opens in a new tab), calls it the same standalone product, now available from inside the Gemini app as well, with existing notebooks and links carried over. This piece uses both names; the help centre now says Gemini Notebook.

At a glance

• Built for. NotebookLM: understanding a fixed set of sources. Claude Project: ongoing work that draws on a set of files.
• Where answers come from. NotebookLM: your sources only. Claude Project: your knowledge files plus Claude’s general knowledge, and the web or connected apps when you allow them.
• Citations. NotebookLM: quotes from your sources on each answer. Claude Project: citations when Claude searches the web; the project pages do not describe passage-level citations for your own files.
• Outputs. NotebookLM: chat, notes, Audio and Video Overviews, reports, mind maps, flashcards, quizzes, slide decks, infographics, data tables. Claude Project: chat, with files and artifacts Claude creates in it.
• Sharing. NotebookLM: Viewer or Editor by email, a chat-only view, or public on consumer accounts. Claude Project: Can view or Can edit, inside an organisation on Team and Enterprise.

Sources and their limits

A notebook is a collection of sources for one project. Google’s page on adding sources↗ (opens in a new tab) lists PDFs, Google Docs, Slides and Sheets, Word and PowerPoint files, web pages, public YouTube videos, audio files, images, ebooks and pasted text. Each source can hold up to 500,000 words or 200MB, and a free account can put up to 50 sources in a notebook, with higher limits on paid plans. Sources imported from Google Drive stay in sync with the original. You can also ask it to find sources for you, with a quick web or Drive search or a longer Deep Research run.

A Claude Project holds knowledge files and instructions. Anthropic’s page on uploading files↗ (opens in a new tab) puts project files at 30MB each, with no fixed count, as long as the total fits Claude’s context window; for non-PDF files Claude extracts the text only. On paid plans that ceiling moves: when project knowledge approaches the context limit, Claude switches to searching it rather than loading all of it, which the help centre says expands capacity by up to ten times. How to choose what goes in is covered in Claude Projects best practices.

The structural difference is scope. Both keep each container separate: Google says a notebook cannot read across other notebooks, and Claude keeps each project’s knowledge and memory to itself. But NotebookLM treats the sources as the whole world, while a Claude Project treats them as the briefing for a colleague who also knows other things.

Citations and grounding

This is where NotebookLM is most distinctive. Google’s page on chat in Gemini Notebook↗ (opens in a new tab) says chat responses only use data from your sources, may decline when a question falls outside them, and come with direct quotes, text and images from the sources as citations so you can check them. That makes it well suited to work where every claim must trace back to a document: a literature review, a policy reading, a set of interview transcripts.

A Claude Project grounds its chats differently. Your files and instructions are in every chat, but Claude answers with its general knowledge too, which is what lets it explain a concept your sources assume, or spot what they leave out. Anthropic’s help centre says web search responses include citations to their sources. The project pages do not describe the same passage-by-passage citation for your own files, so if you need it, say so in the instructions: ask Claude to name the file and quote the line it relied on, and to say plainly when an answer does not come from the knowledge.

What each one produces

NotebookLM’s Studio panel turns sources into finished formats. The best known is the Audio Overview; Google’s page on generating an Audio Overview↗ (opens in a new tab) describes four formats: Deep Dive, a conversation between two AI hosts; The Brief, a single speaker in under two minutes; The Critique; and The Debate. You can choose the language and length and steer it with a prompt. Beside it sit Video Overviews, mind maps, reports such as a briefing document or study guide, flashcards, quizzes, slide decks, infographics and data tables, with reports exportable to Google Docs and tables to Sheets.

A Claude Project produces what a conversation produces: analysis, drafts, rewritten sections, comparisons, code, and files or artifacts Claude creates when you ask. There is no Studio panel of preset formats. The output is shaped by your instructions and your question, which is less guided and more open. If the deliverable is a memo in your house style or a decision with the trade-offs written out, that openness is the point; if it is a podcast-style summary for commuting, NotebookLM does it in one press.

Instructions, memory and tools

• Instructions. A Claude Project has project instructions applied to every chat. A notebook lets you configure chat with a style, such as Learning Guide or a custom one, and a response length, and each Studio output takes its own prompt.
• Memory. Anthropic says each Claude project has its own memory space and summary where memory is on. In a notebook, your chat history is saved and stays private to you, and you can save a response as a note.
• Tools. Claude in a project can use web search, Research and, in private projects, connectors to other apps. NotebookLM reaches outward mainly to find new sources, and Google has announced a secure cloud computer that runs code over your sources, rolling out by plan.

Sharing

Anthropic’s overview of what projects are↗ (opens in a new tab) makes project sharing a Team and Enterprise feature: members of your organisation get Can view or Can edit, and each person’s chats stay their own. The detail is in can Claude Projects be shared?

Google’s help centre describes wider options for a notebook. You can add people by email as Viewer, who can read the sources and notes, or Editor, who can add and remove sources and share further. A chat view hides the sources for a focused experience, though Google notes it does not revoke a viewer’s access to them. On consumer accounts a notebook can be made public, and its owner can share individual outputs such as an Audio Overview; public sharing is off for Workspace Enterprise and Education accounts. Personal accounts can share a notebook with up to 50 people.

Where each fits

• Reading a stack of papers, reports or transcripts and needing every answer tied to a passage: NotebookLM.
• Studying or onboarding, where flashcards, a quiz or an audio summary are the goal: NotebookLM.
• Turning research into a recommendation, a draft or a plan in your organisation’s voice: a Claude Project.
• Work that mixes your documents with live information or other systems, through web search or connectors: a Claude Project.
• Both at once: build the notebook to understand the sources, then carry the conclusions, with their citations, into the project where the work continues.

If the files you are researching live in a repository rather than a folder of documents, neither is the natural home; see Claude Projects vs Claude Code.

What happens after the research

Research ends in things to do: check this figure, ask the client about that clause, rewrite section three. Neither a notebook nor a project records who has picked those up or whether they are done. On fenbs they become tasks on a board, each a feature, enhancement or bug, moving through To Do, Next Up, In Progress and Completed, with every change recorded under who made it. Claude connects over MCP with the Claude connector and can file the follow-ups straight from the project chat where they came up, under your role.