# How to Use Claude Projects with Gemini Notebook (formerly NotebookLM)
> How to use Claude Projects for research: write the instructions, choose what goes in project knowledge, and connect NotebookLM so every answer comes back cited.
By Juli · Published Oct 26, 2026 · https://www.kortex-notebooklm.com/blog/claude-projects-research
Claude Projects are the easiest way to stop re-explaining your research to Claude. You write the brief once, add a few reference files, and every chat in the project starts from them. The catch shows up when the reading pile grows: project knowledge shares Claude's working memory, and Claude is free to blend your files with what it already knows. This guide shows how to use Claude Projects for research without writing a line of code, and how to make Gemini Notebook (formerly NotebookLM) the project's source library, so the answers that matter come back with citations.
> **Key Takeaways**
> - A Claude Project holds instructions, knowledge files, its own chats and its own memory. Free accounts get five (Claude Help Center, 2026).
> - Put the brief, outline and style guide in project knowledge. Keep the papers in Gemini Notebook, which answers only from your sources and cites them.
> - Kortex MCP, added as a custom connector, lets the project ask up to three notebooks at once and save findings back. It's included with Kortex Pro and Lifetime.
> - Three prompts cover most research weeks: a literature scan, a weekly brief and a cited draft.
## What is a Claude Project, and what does it hold?
A **Claude Project** is a self-contained workspace with its own chat history and knowledge base, in Anthropic's words ([Claude Help Center](https://support.claude.com/en/articles/9517075-what-are-projects), 2026). It holds three things you control. **Instructions** shape every chat in the project, such as a tone, a role or a set of rules. **Project knowledge** is the documents, text and code Claude reads for context. And each project keeps **its own memory**: inside a project, Claude uses that project's memory plus your general memory, but never another project's ([Claude Help Center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context), 2026).
Projects come with every plan. Free accounts can create up to five. Paid plans (Pro, Max, Team and Enterprise) add retrieval, which switches on when knowledge nears the context window and expands capacity by up to 10 times ([Claude Help Center](https://support.claude.com/en/articles/11473015-retrieval-augmented-generation-rag-for-projects), 2026). Each project file can be up to 30 MB, and the number of files is unlimited, but the total still has to fit Claude's context window ([Claude Help Center](https://support.claude.com/en/articles/8241126-upload-files-to-claude), 2026).
One detail trips people up. Chats in a project don't share context with each other unless it's in project knowledge ([Claude Help Center](https://support.claude.com/en/articles/9519177-how-can-i-create-and-manage-projects), 2026). Memory and, on paid plans, chat search can recall some of it, but neither is a record you can cite. So the great finding from Tuesday's chat needs saving somewhere Wednesday's chat can quote it. The setup below gives every finding that home.
## Which Claude Projects do you have: chat, Cowork or the Code tab?
Claude has two kinds of project, and which one you see depends on your plan and the part of the app you're in. Most people have the first. Pro and Max accounts may also have the second, in the Code tab.
| | Projects in chat and Cowork | Projects in the Code tab |
|---|---|---|
| What it is | A workspace of chats that share instructions and knowledge | One ongoing conversation where Claude coordinates threads that do the work |
| Where it works | claude.ai, the desktop app and the mobile apps | claude.ai/code, the desktop app's Code tab and the Claude mobile app |
| Who has it | Every plan; Cowork features need a paid plan | Public beta on Pro and Max; not on Team or Enterprise |
| Files | Project knowledge, up to 30 MB per file | Library: up to 100 files and 2 GB per upload, one file up to 480 MB |
| Schedules | Scheduled tasks inside Cowork projects | Cloud scheduled tasks on the project's Scheduled tasks tab |
Sources: [Claude Help Center](https://support.claude.com/en/articles/15520349-use-claude-cowork-on-web-desktop-and-mobile), 2026; [Claude Code docs](https://code.claude.com/docs/en/claude-projects), 2026.
**Projects in chat and Cowork.** In chat, a project is a home for related conversations. **Cowork** is Claude's mode for carrying out multi-step tasks with your files and apps. It gives each project its own instructions, scheduled tasks, memory and context, which can be a local folder, a linked chat project or a URL ([Claude Help Center](https://support.claude.com/en/articles/14116274-organize-your-tasks-with-projects-in-claude-cowork), 2026). Cowork is on paid plans. On accounts where chat and Cowork are unified, there's no separate Cowork switch: you describe a task in any conversation and Claude takes it from there. Projects work on the web, the desktop app and mobile, though a project tied to a local folder runs Cowork sessions on desktop only (Claude Help Center, 2026).
**Projects in the Code tab.** The name suggests programmers, but no code is needed. Anthropic's docs say a project there "doesn't need a repository at all," and its threads can still research and write documents, delivering files to the project's Library (Claude Code docs, 2026). You send work to one coordinating conversation. Claude starts a thread for each task, the threads run in parallel in the cloud and keep going after you close your laptop, and you can check on them from your phone. Every thread starts with the project's instructions, files and memory, and instructions can run to 16,000 characters.
The trade-off is usage: a project draws on the same plan limits as your other Claude Code sessions and uses them faster. Not every Pro or Max account has it, either. If **Projects** isn't in the Code tab's sidebar or under **More**, yours doesn't. An existing chat project can be upgraded to this version, and Anthropic says its instructions, files, chat history and scheduled tasks come along (Claude Help Center, 2026).
The setup below works in both. Where the clicks differ, each step says so.
## How are Claude Projects different from chats, Cowork, skills and artifacts?
A project is where context lives. The other four are what you do with it: a chat is one conversation, Cowork carries out tasks, a skill is a reusable method, and an artifact is something Claude makes. Google autocomplete for "claude projects" completes to exactly these comparisons ("vs chats", "vs cowork", "vs skills", "vs artifacts"), so here they are side by side.
| Feature | What it is | Use it in research for |
|---|---|---|
| Chat | One conversation; inside a project it starts with the instructions and knowledge | Quick questions and one-off reading |
| Project | A workspace with its own instructions, knowledge, chats and memory | One research question you'll return to for weeks |
| Cowork | Claude doing multi-step tasks with your files, apps and schedules, on paid plans | The weekly brief and other recurring chores |
| Skill | A folder of instructions Claude loads when a task calls for it, anywhere in Claude | A method you reuse everywhere, like your citation style |
| Artifact | Anything Claude makes that you'd put in front of someone: a doc, deck or dashboard | The draft, the slides, the summary table |
Anthropic draws the skills line clearly. Projects "provide static background knowledge that's always loaded when you start chats within them," while skills load only the information a task needs ([Claude Help Center](https://support.claude.com/en/articles/12512176-what-are-skills), 2026). Artifacts are outputs, not inputs: they're "saved to the Artifacts tab in your Claude sidebar," available on every plan ([Claude Help Center](https://support.claude.com/en/articles/17153992-what-are-artifacts), 2026). In practice, a research project uses all five. The project holds the brief, chats and Cowork do the work, a skill keeps your house style, and the draft comes out as an artifact.
## Why use Gemini Notebook as the project's source library?
Because a research pile needs two things project knowledge doesn't promise: room and receipts. On room, a free Gemini notebook holds 50 sources, Google AI Pro raises that to 300 and Ultra to as many as 600 ([Gemini Notebook Help](https://support.google.com/gemininotebook/answer/16213268?hl=en), 2026). Each source can run to 500,000 words ([Gemini Notebook Help](https://support.google.com/gemininotebook/answer/16269187?hl=en), 2026). On receipts, Google says "chat responses in Gemini Notebook only use data from your sources," and it uses "direct quotes, text, and images from your sources as citations" ([Gemini Notebook Help](https://support.google.com/gemininotebook/answer/16179559?hl=en), 2026).
Claude works differently, on purpose. It reads project files as background, then brings everything else it knows. That's why it drafts so well. However, it's a risk when you need to prove a sentence came from one specific report. Grounding narrows that risk. In a 2025 study on a 300-document corpus, NotebookLM hallucinated in 13% of outputs against about 40% for ChatGPT and Gemini chat ([Hagar, Agustianto & Diakopoulos](https://arxiv.org/abs/2509.25498), 2025). Claude wasn't in that test, but the lesson carries: let the notebook find and cite, and let Claude reason and write. For the full side-by-side, see our [Gemini Notebook vs Claude Projects comparison](/blog/gemini-notebook-vs-claude-projects).
People ask for exactly this pairing. When we analyzed the 776 answers to the Kortex MCP waitlist survey (Sep-Oct 2026), 110 people wanted their AI assistant to ask their notebooks and get cited answers back. Separately, 91 wanted drafts written from their own sources, and 128 wanted jobs to run on a schedule. Researchers were the largest profession on the list, 202 of 776 (26%). One respondent summed up this whole guide: "I need an AI agent that, for a given topic, will find relevant literature, create a review outline, and write a draft."
People asking for each job, out of 776 (one person could ask for several). Highlighted: the jobs this guide's three prompts cover. Source: Kortex MCP waitlist survey, Sep 19 to Oct 1, 2026.
The problem is reach. Google publishes no public consumer API for Gemini Notebook, so a Claude Project can't open your notebooks by itself. The default, therefore, is copy and paste, and citations rarely survive the trip. In the same survey, 244 of 776 respondents (31%) described a manual copy-paste workaround they use today.
Build the library your Claude Project will read
Kortex files your Gemini notebooks into folders, and its button inside Claude sends any chat to a notebook with prompts and replies labelled. Free on Chrome.
## How to use Claude Projects for research, step by step
Name the project after one question, write a short brief as its instructions, put your working documents in project knowledge, keep the evidence in two or three notebooks, then connect Kortex. The example is a policy researcher preparing a council briefing on four-day work week pilots. Swap in your own topic.
### Step 1: Create the project around one research question
Scope it to a question you'll come back to for weeks, not a whole field. "Four-day week briefing" works; "Research" doesn't.
- **In chat:** open **Projects** in the sidebar (or claude.ai/projects), click **+ New Project**, and enter a name and description. Claude can't see the name or description, so the goal belongs in the instructions (Claude Help Center, 2026).
- **In the Code tab:** open **Projects**, select **New project**, and fill in the name plus an optional one-line **Goal**, which Claude works toward (Claude Code docs, 2026).
### Step 2: Write instructions that work as a research brief
Instructions are the brief every chat or thread starts from. In chat, click **Set project instructions**, paste, and **Save instructions**. In the Code tab, they live under **Project settings > Memory > Project instructions**. Here's a brief for the example project:
```
You are my research assistant for "Four-day week briefing": a briefing for a city council on whether to pilot a 32-hour week.
Evidence lives in three Gemini notebooks, reached through the Kortex connector:
- "4DW trial reports": pilot results and evaluations
- "4DW academic studies": peer-reviewed papers
- "4DW critics and press": opposing views and coverage
Rules:
1. For any factual claim, ask the right notebook first (ask_question, or ask_library for up to three notebooks) and keep its citations next to the claim.
2. Label anything that comes from you rather than a notebook as "my note".
3. If the notebooks don't cover it, say "not in my notebooks" and suggest a source to add.
4. Use project knowledge for the brief, the outline and the style guide, not as evidence.
5. Show me any text before you save it to a notebook.
6. Plain English, short paragraphs, every number with its source.
```
Rule 4 does the quiet work. Without it, Claude treats your outline as a source, and your own assumptions start coming back to you as findings.
### Step 3: Decide what goes in project knowledge and what stays in notebooks
Project knowledge tells Claude how to work. Notebooks hold what's true. That one split prevents most research mix-ups.
| Put it in project knowledge | Keep it in Gemini Notebook |
|---|---|
| The brief, outline and research questions | Papers, reports and evaluations |
| Your style guide or a template you write to | Interview transcripts and meeting notes |
| Your current draft | Web articles and YouTube videos you're tracking |
| A one-line map of each notebook and what it holds | Anything you'll need to quote with a passage |
Two reasons. First, project knowledge is a copy. In the Code tab, a change on your computer doesn't reach the project until you upload the file again (Claude Code docs, 2026). Cowork won't change a project's contents either, so you add anything worth keeping yourself (Claude Help Center, 2026). Second, notebooks scale and cite. Three notebooks of 50 sources each is a lot of reading, and every answer points back to a passage.
Keep the knowledge files tidy, too. Anthropic's retrieval tips say to use clear, descriptive filenames, group related documents in the same project, and name specific documents when you ask (Claude Help Center, 2026). On the notebook side, split the pile by strand of evidence rather than by date. Our guide to [organizing Gemini notebooks](/blog/organize-notebooklm-notebooks) covers folders, tags and naming.
### Step 4: Connect Kortex MCP so the project can ask your notebooks
**Kortex MCP** is the bridge we built between AI clients and Gemini Notebook. Claude calls one endpoint, Kortex relays the call to the Kortex extension in your own Chrome, and the extension asks Gemini Notebook with the Google session you're already signed in to. Kortex never stores your Google password or cookies ([Kortex MCP docs](/docs/mcp-server), 2026). It works the other way too: whatever Claude can reach, such as a web page or a file, Kortex can carry into a notebook. It's included with Kortex Pro and Lifetime.
1. Install the Kortex extension in Chrome and sign in with a Pro or Lifetime account.
2. In Claude, go to **Customize > Connectors**, click **+ Add**, then **Add custom connector** ([Claude Help Center](https://support.claude.com/en/articles/11175166-getting-started-with-custom-connectors-using-remote-mcp), 2026).
3. Name it `Kortex`, paste `https://mcp.kortex-notebooklm.com/mcp`, and continue. A Kortex consent page opens: sign in and choose what to allow. Reading is enough for answers; allow writes if the project should save findings back.
4. **In a chat project,** switch Kortex on in each chat from **+ > Connectors**. **In the Code tab,** every cloud thread gets your account's connectors automatically, but the coordinating conversation has none, so phrase notebook work as a task for a thread (Claude Code docs, 2026).
Leave Chrome running, even minimised. The first call after a quiet spell can take up to 30 seconds while the extension wakes. Our guide to [connecting Gemini Notebook to Claude](/blog/gemini-notebook-in-claude) covers tool permissions and sessions in more detail.
## Three research prompts: a literature scan, a weekly brief and a draft
These three prompts lean on the instructions above, so they stay short. Each names its notebooks, because a named notebook beats a guessed one.
### 1. A literature scan across notebooks
`ask_library` asks the same question across up to three notebooks and returns each answer separately, with its own citations (Kortex MCP docs, 2026). For example, a scan that compares strands of evidence looks like this:
```
Literature scan: use ask_library on "4DW trial reports" and "4DW academic studies" with this question: "What effects on productivity, burnout and staff retention do these sources report, and how was each one measured?" Show each notebook's answer separately with its citations. Then make a table: finding, notebook, citation, method or sample if the source gives one. End with the three places the notebooks disagree. Save the table as a note in "4DW academic studies" with create_notebooklm_note, titled "Lit scan YYYY-MM-DD".
```
Here's the shape of what comes back (made-up sources, trimmed):
> **"4DW trial reports":** Most pilot firms reported steady or higher output [1][2]. Sick days fell in two of three evaluations [3].
> **"4DW academic studies":** Burnout scores dropped, but the effect shrank after a year in one panel study [4].
> **Where they disagree:** the trial reports measure output by manager survey [1]; the panel study uses payroll data [4].
The "where they disagree" line is the part a blended answer would hide. Saving it with `create_notebooklm_note` turns Tuesday's finding into a native note that Wednesday's chat can find with `search_everything`. Then `convert_note_to_source` makes it something the notebook can cite (Kortex MCP docs, 2026). In a Code tab project, you can send three scan questions in one message, and Claude runs each as its own thread.
### 2. A weekly brief that keeps the library current
This prompt uses the bridge in both directions. Claude searches the web, Kortex carries what it finds into a notebook, and the notebook answers with citations.
```
Weekly brief:
1. Search the web for news and reports on four-day week pilots from the past seven days. Pick up to five credible ones.
2. Add their URLs to my "4DW critics and press" notebook with add_url_source.
3. Ask that notebook: "What do the newest sources add to or dispute about four-day week pilots?" Keep the citations.
4. Write a brief of up to 200 words: what's new, what it challenges in the draft in project knowledge, and which sources it rests on.
If Kortex doesn't answer, stop and say so. Don't fill the gap from memory.
```
To put it on a timer, describe the task and how often it should run in any conversation, then click **Schedule**. In Cowork, you can also go to **Scheduled**, then **New task**. Tasks run hourly, daily, weekly, on weekdays or manually, and they run in the cloud even when your computer is asleep ([Claude Help Center](https://support.claude.com/en/articles/13854387-schedule-recurring-tasks-in-claude-cowork), 2026). In a Code tab project, ask Claude to schedule it, and it appears on the project's **Scheduled tasks** tab. One honest catch: Kortex answers only while your Chrome is open, so schedule the brief for a time your computer is on. Alternatively, let Kortex run the notebook half itself: its schedules ask a notebook and save the answer hourly, daily or weekly while Chrome is open, as our [daily briefing guide](/blog/gemini-notebook-daily-briefing) shows.
### 3. A draft built only from cited findings
The draft is where Claude shines, so give it room to write but no room to invent.
```
Draft section 2 of the briefing, "What the pilots found", in about 400 words, following the outline and style guide in project knowledge. Build it from the lit scan note and the weekly briefs; for anything they don't cover, ask the notebooks with ask_library. Put each citation right after the sentence it supports and mark your own reasoning as "my note". Write it as a document I can edit. Then list every claim you couldn't source.
```
Claude writes the draft as a document, which lands in the Artifacts tab, or in the Library in a Code tab project. In short, the list of unsourced claims at the end is your reading list for next week.
A research project that cites its sources
Your Claude Project asks up to three Gemini notebooks at once, keeps each answer's citations, and saves the findings back for the next chat. Kortex MCP is included with Kortex Pro and Lifetime.
## How do you keep a research project accurate?
Keep the notebook and Claude in separate lanes, save every finding with its citation, and watch two sets of limits. Specifically:
- **Two lanes.** Notebook claims sit next to their citations; Claude's own thoughts carry a "my note" label. Rule 2 of the brief does this, but check the first few answers anyway.
- **Save findings with their receipts.** Because chats don't share context, a finding needs a home. `save_chat_to_note` keeps a cited answer as a notebook note; `add_text_source` saves a decision as a source the notebook can cite later (Kortex MCP docs, 2026). Put the date in the title and the original citation inside, since next month it will get cited like any other source. Our guide to [using Gemini Notebook as memory for Claude](/blog/gemini-notebook-ai-memory) goes deeper on what to save where.
- **Read-only when you only need answers.** Text hidden in a web page can try to give your assistant instructions, which is called prompt injection. A read-only connection can't change your library, whatever a source says.
- **Two sets of limits.** Claude's plan limits apply to the project, and Code tab projects use them faster (Claude Code docs, 2026). Gemini Notebook's own usage refreshes every five hours up to a weekly cap ([Gemini Notebook Help](https://support.google.com/gemininotebook/answer/17670842?hl=en), 2026), and `ask_library` asks each notebook separately. When you know which notebook holds the answer, ask just that one.
## Where should you start?
Start with one question, two or three notebooks and the brief above. Run the literature scan first, because it shows within one chat whether your notebooks hold enough to answer the question. Then add the weekly brief, and write the draft last. If the notebooks themselves need building, our guide to [using Gemini Notebook for research](/blog/notebooklm-for-research) covers source types, upload limits and synthesis prompts. The short version: the project holds how you work, the notebooks hold what's true, and Kortex MCP lets one ask the other.
Give your Claude Project a library that cites
One connector, and every chat in your project can ask your Gemini notebooks, show the passages behind each answer, and save what you learn. Kortex MCP is included with Kortex Pro and Lifetime, and Kortex itself is free on Chrome.
## Frequently asked questions
### How do you use Claude Projects?
Create a project, set its instructions, add the files Claude should read as project knowledge, then start every related chat inside it. Each chat begins with those instructions and files. Free accounts can create up to five projects, and paid plans add retrieval so a project can hold up to 10 times more content.
### What is the difference between Claude Projects and chats?
A chat is one conversation. A project groups chats around shared instructions, knowledge files and its own memory. Chats inside a project don't share context with each other unless it's in project knowledge, so save findings to the knowledge base or a notebook if a later chat needs them.
### What is the difference between Claude Projects and Cowork?
A project is where context lives: instructions, files and memory. Cowork is Claude carrying out multi-step tasks with your files, apps and schedules, on paid plans. Cowork projects add per-project scheduled tasks and a linked folder, chat project or URL as context, so you often use both together.
### What is the difference between Claude Projects and skills?
Anthropic describes projects as static background knowledge that's always loaded in that project's chats, while skills are folders of instructions Claude loads only when a task needs them, anywhere in Claude. Put one research question's brief in a project, and a method you reuse everywhere, like a citation style, in a skill.
### Can a Claude Project read my Gemini notebooks?
Not on its own, because Google publishes no public consumer API for Gemini Notebook. Add Kortex MCP as a custom connector and the project can ask up to three notebooks at once with citations and save findings back. Kortex MCP is included with Kortex Pro and Lifetime, and Chrome has to stay open.
### Are Claude Projects free?
Yes, with limits. Free accounts can create up to five projects. Retrieval for large knowledge bases, Cowork and scheduled tasks need a paid plan (Pro, Max, Team or Enterprise). Projects in the Code tab, where Claude runs tasks as parallel threads, are in public beta on Pro and Max.
---
*Editorial note: fact-checked against the official help pages and docs linked above on October 26, 2026. Spotted something out of date? [Contact us](/support) and we will fix it, or read more [about Kortex](/about).*