What is NotebookLM?

NotebookLM is a Google research notebook. Unlike a general AI that can pull from anywhere on the web, you upload a fixed set of sources — papers, articles, PDFs, transcripts, slides, web pages, YouTube videos — and the AI answers only from those.

That sounds like a small distinction. It isn’t. It changes the whole tool from a chatbot you argue with into a research assistant you trust.

What you actually upload

A notebook holds up to 50 sources on the free plan, 300 on Pro. Sources can be:

• PDFs (research papers, reports, manuals)
• Google Docs and Slides
• Plain text and Markdown
• Web URLs (NotebookLM fetches the page)
• YouTube videos (it pulls the transcript)
• Audio files (interviews, recordings — it transcribes them)

You then chat with the combined material. Ask for a summary, a study guide, a timeline, a list of contradictions between sources, an FAQ, a briefing doc. Every answer comes with inline citations — click any sentence and NotebookLM jumps to the exact paragraph in the exact source. That citation behaviour is the load-bearing feature for serious work. It’s an AI you can audit.

The audio overview moment

NotebookLM had been quietly available for over a year before it broke into the mainstream. The thing that did it was Audio Overviews, released in late 2024 — a feature that turns your sources into a synthetic two-host podcast.

The hosts banter. They riff. They ask each other questions. They stumble over phrases the way real podcasters do. People uploaded their resumes, their emails, their LinkedIn profiles, even single tweets, and got back ten-minute conversations about themselves. It went viral because the output didn’t sound like AI — it sounded like two coworkers prepping you for a meeting.

Video Overviews followed in 2025 — short narrated videos generated from your sources, suitable for quick explainers or social posts. The same trick: ground the output in your uploads, then render it in a different medium.

How it compares to other tools

If you’ve used Claude Projects or ChatGPT’s custom GPTs, NotebookLM occupies overlapping ground but with a different bias:

• Claude Projects — also lets you upload sources, and Claude is generally stronger at long-form writing and code. But Projects doesn’t generate audio or video, and its citations are less prominent.
• Custom GPTs — designed more around persona and tool use. Good for building a small assistant with a specific job. Less optimised for grounded research.
• NotebookLM — leans hardest into read-only research over a corpus. It’s the one you reach for when the goal is “help me understand this stack of material.”

None of these will replace the others. They’re different shapes of the same idea — give an AI a defined context window of trusted material, then talk to it.

Real use cases

The kinds of jobs people actually use NotebookLM for:

• Researchers — drop in twenty papers from a literature search and ask “what do these disagree on?” The contradictions list alone saves hours.
• Teachers and course creators — turn a textbook chapter plus three supplementary readings into a study guide and an audio overview students can listen to on the bus.
• Journalists — pull a politician’s last fifty speeches and ask where their position on a topic shifted.
• Lawyers and analysts — load a contract, related case law, and an explainer doc, then ask the questions you’d otherwise ask an associate.
• Founders and product teams — drop in user interview transcripts and ask “what’s the most-mentioned frustration?” (NotebookLM will quote you the exact line.)
• Personal use — upload your own writing and ask it to summarise your worldview, or upload a long manual and chat with it instead of reading the whole thing.

The catch: no official API

NotebookLM has no official public API. There’s no programmable way for one program to ask another to create a notebook, upload sources, or kick off an audio overview. Everything happens through the Google web app at notebooklm.google.com.

For an individual that’s fine — you click a few buttons and you’re done. For anyone trying to batch-generate dozens or hundreds of notebooks (a course, a podcast series, a research portfolio) it’s a real wall. The community has worked around this with unofficial wrappers and browser-automation tools. Here’s how I used those tools to ship 39 cinematic AI history videos.

Pricing

Free tier: 100 notebooks, 50 sources per notebook, 3 audio overviews per day. Enough for almost any individual workflow. NotebookLM Plus (bundled into Google AI Pro and Google Workspace) raises the limits to 500 notebooks, 300 sources per notebook, and higher daily generation caps — that’s the tier you reach for if you’re building course material or doing professional research.

When to reach for it

Use NotebookLM when you have too much source material to read all of it, and you need an AI whose answers you can verify against the originals. Skip it when your task is open-ended creative writing, code generation, or anything that needs the AI to bring in outside knowledge — that’s not what it’s built for.

External reference: Grokipedia: NotebookLM.

Related

• What is Claude Projects? — the closest analog in the Anthropic ecosystem.
• What is a Context Window? — the underlying mechanism that makes uploading sources work.
• What is MCP? — the emerging standard for connecting AI to your own tools and data.
• What is an API? — and why NotebookLM not having one matters.
• What is Playwright? — the browser-automation tool people use to script NotebookLM in the absence of a real API.
• How I used an unofficial NotebookLM API to ship 39 cinematic AI history videos — the deeper story.