NotebookLM vs Claude Projects vs ChatGPT: Which Should You Use

If you need an AI tool that works from your own documents instead of the open web, the right pick depends on what "your own documents" means: a pile of PDFs and transcripts you want summarized and quoted with citations (Gemini Notebook, the product Google spent years calling NotebookLM), an ongoing workspace where a model keeps context across many chats and tasks (Claude Projects), or a reusable, shareable chatbot you configure once and hand off to a team or customers (ChatGPT's Projects and Custom GPTs). All three ship today, all three are free to start, and each one solves a slightly different problem — this guide lines up their real, currently published limits, prices, and citation behavior so you can pick without guessing.

What each tool actually is

Google's Gemini Notebook — the product formerly and still informally known as NotebookLM — is a source-grounded research assistant. You upload documents, slides, audio, or web links into a notebook, and it answers questions, drafts summaries, and generates audio or video overviews using only what you gave it. Google renamed NotebookLM to Gemini Notebook on July 16, 2026, folding it fully into the Gemini brand; existing notebooks and shared links carry over automatically, and Google says more than 30 million people and over 600,000 organizations were using the tool at the time of the rename.

Anthropic's Claude Projects is a persistent workspace inside Claude: you upload files, paste text or code, and write standing instructions that apply to every chat you start inside that project. It's less a citation engine and more a way to keep Claude anchored to a consistent set of background material — a style guide, a codebase, a client's files — across many separate conversations.

OpenAI splits the job in two. Projects in ChatGPT are folders that group chats, files, and custom instructions for a single piece of work. Custom GPTs are a separate, older feature: a configured version of ChatGPT — with its own instructions, knowledge files, and optional actions — that you build once and reuse or share, including through the public GPT Store.

NotebookLM is now Gemini Notebook

The rename is the single biggest news item touching this comparison. Google's own announcement describes it as a name and logo change only — "the same standalone product," now integrated more tightly with the Gemini app and, eventually, Google Search's AI Mode. Admins don't need to do anything, and automatic redirects keep old notebooklm.google.com links and shared notebooks working. The underlying product also picked up a new capability alongside the rename: every notebook now gets a secure cloud sandbox so Gemini Notebook can write and run code against your uploaded sources, a feature Google is rolling out to Google AI Ultra and qualifying Workspace accounts first, with Pro-tier access following "over the coming weeks." For the rest of this piece we'll use "Gemini Notebook" with "NotebookLM" noted on first mention per section, since that's the name you'll actually see in the product today.

Source and document limits, plan by plan

This is where the three tools diverge most sharply, because they're solving different storage problems: Gemini Notebook caps how many discrete sources you can drop into one notebook, Claude scales a single project's knowledge base against its context window (then quietly switches to retrieval once you outgrow it), and ChatGPT caps the number of files per project folder and, separately, per Custom GPT.

Two things are worth flagging rather than smoothing over. First, Claude and ChatGPT both decline to publish a hard number for "how much can I actually put in here" — Claude because retrieval-augmented generation (RAG) makes the real ceiling a function of relevance-ranking rather than a fixed count, and OpenAI because its own Help Center articles on Custom GPT knowledge limits contradict each other (20 files in one, 10 in another). Second, Gemini Notebook is the only one of the three that publishes a clean, plan-by-plan number for exactly how many sources fit in one container — a side effect of its retrieval model being built around discrete, named sources from the start, as Google's own plan and usage-limits page lays out.

Claude Projects and automatic retrieval

Claude Projects are available to every account, including free ones, which get up to five projects to organize chats around. The ceiling on a free project is simply the model's context window — once your uploaded knowledge and conversation would exceed it, you can't add more.

On Pro, Max, Team, and Enterprise plans, Claude switches on retrieval-augmented generation automatically once a project's knowledge nears that context-window threshold — there's no toggle to find or setting to configure. Anthropic says this can expand effective capacity "by up to 10x" while keeping response quality roughly consistent with in-context processing, because Claude searches and pulls only the most relevant passages for a given question rather than loading every file into memory at once. A visual indicator in the project shows when RAG mode is active, and you can nudge retrieval by referencing a specific uploaded document by name in your prompt. Project sharing with teammates — as opposed to just using a project solo — is restricted to Team and Enterprise plans, and project memory, where supported, is kept separate from your regular chat memory.

ChatGPT Projects vs. Custom GPTs

ChatGPT Projects are the closer analogue to Claude Projects and Gemini Notebook: a container for chats, files, and instructions tied to one piece of work. OpenAI's published file-count tiers scale cleanly with plan — 5 files on Free, 25 on Go and Plus, 40 on Edu, Pro, Business, and Enterprise — though a separate Help Center note caps simultaneous uploads at a smaller batch size regardless of your total allowance, which has generated recurring confusion in OpenAI's own developer community forums about why a "25-file plan" sometimes balks partway through a bulk upload.

Custom GPTs solve a different problem: a reusable configuration — custom instructions, a knowledge base, and optionally external actions — that you build once, in the GPT Builder, and reuse or publish to the GPT Store. Building or editing a Custom GPT requires a paid plan (Plus, Pro, Team, or Enterprise); free accounts can use GPTs other people have published but can't open the builder themselves. This is a meaningfully different restriction from Gemini Notebook and Claude Projects, both of which let free users build their own containers from day one — OpenAI reserves creation itself for paying customers.

Citation behavior: who shows you the receipts

Gemini Notebook is the only one of the three whose own documentation makes grounding a hard guarantee rather than a feature you have to hope for: Google's help center FAQ states that responses are built from your uploaded sources, and if an answer isn't supported by those sources, it won't provide one. For long or dense sources it retrieves the most relevant passages rather than citing the whole document, though Google notes citations can occasionally reference a whole short source instead of a specific passage.

Claude's and OpenAI's consumer documentation are both thinner on this point for their project features specifically. Anthropic's support article on RAG for projects describes the retrieval mechanism in detail but doesn't spell out whether or how Claude visually attributes an answer to a specific uploaded file inside a project chat (Anthropic's developer-facing citations feature, which returns exact supporting passages, is documented for API use rather than for the consumer Projects experience). OpenAI's Projects and Custom GPT help articles likewise focus on file limits and setup rather than describing a citation format in detail. In practice, this means Gemini Notebook is the safer default if citation-backed, source-only answers are the actual point of the exercise — it's the one tool built around that guarantee from the ground up, which is consistent with how our own look at AI search assistants and their sourcing habits has generally found citation depth varies a lot by design intent, not just model quality.

Pricing, plan by plan

• Gemini Notebook: Free to start. Higher source/notebook limits come bundled with a paid Google AI Plus, Pro, or Ultra subscription, or with qualifying Google Workspace and Workspace for Education plans. Check Google's own Google AI plans page for current local pricing, since it renders by region rather than publishing one fixed figure on the Gemini Notebook help pages.
• Claude: Free (up to 5 projects); Pro is $20/month billed monthly or roughly $17/month billed annually; Max starts around $100/month; Team runs $20–25 per seat on the standard tier and $100–125 per seat on the premium tier (annual vs. monthly billing); Enterprise is custom-priced with API-rate usage on top, per Claude's pricing page.
• ChatGPT: Free; Plus is $20/month; OpenAI also sells a lower-cost Go tier and higher Pro and Business/Enterprise tiers, but exact current figures for those higher tiers are not something we're going to invent here — OpenAI's own pricing page is the place to check them before you buy, since third-party trackers disagree with each other on the Pro and Business numbers at the time of writing.

Which one should you actually use

If the job is "digest a stack of PDFs, lecture slides, or transcripts and get grounded, source-only answers back," Gemini Notebook is the purpose-built tool — it's the one with a documented, plan-by-plan source cap and a stated refusal to answer outside its sources. If the job is "keep an AI assistant anchored to a consistent body of reference material across months of ongoing work," Claude Projects is the better fit, especially once you're on a paid plan and RAG quietly removes the practical ceiling — the same persistent-workspace thinking that shows up in how Anthropic has been building out Claude Skills as reusable, portable instructions rather than one-off prompts. If the job is "build a configured assistant once and hand it to a team, customers, or the public," ChatGPT's Custom GPTs remain the only one of the three built explicitly for that distribution model, store listing included — and OpenAI's broader model lineup, including the model families discussed in our look at what powers ChatGPT's newest tiers, keeps shifting what a GPT can do without changing this basic structure.

None of the three is strictly better; they're answering different questions. A researcher summarizing fifty sources wants Gemini Notebook's hard source cap and citation guarantee. A consultant running a months-long client engagement wants Claude Projects' persistent, RAG-backed knowledge base. A team that wants to hand a configured assistant to non-technical colleagues wants a Custom GPT.

What next

Expect the file and source caps on all three products to keep moving upward rather than staying put — Google raised Gemini Notebook's ceilings as part of the July 2026 rebrand, and OpenAI has already revised its Projects file limits more than once since the feature launched. Before committing to a workflow built around any one tool's limits, check the live help page rather than this article or any other, since all three vendors reserve the right to change these numbers without much notice. If your work genuinely spans all three ecosystems, there's no rule against using Gemini Notebook for source-grounded research, Claude Projects for ongoing reference work, and a Custom GPT for anything you need to hand off — the subscriptions are not mutually exclusive, even if paying for three AI subscriptions at once is its own kind of decision.

Written byTheo Park

Theo Park runs the AI desk at Pandromeda. He follows model launches from the frontier labs and the open-weight community, tracks the assistants and developer tools built on them, and explains what each release changes on pricing, capability and safety. His reporting leans on primary sources: model cards, technical reports, API documentation and the companies' own announcements.

More from Theo ParkAll AI coverage