NotebookLM Start From Scratch: you no longer need to upload anything
Since June 8, 2026, you can open a blank notebook, type an idea, and let NotebookLM find the sources. It deploys Google Search on its own, gathers relevant pages, and pulls them into your notebook as cited, active sources — you never had to assemble them first. This inverts NotebookLM's founding rule ("bring your own documents") and turns the blank notebook into the starting line for research.
Yes. Since June 8, 2026, you can open a blank NotebookLM notebook, type a question or idea, and it uses Google Search to autonomously find, cite, and add relevant web sources for you — you approve which stay. The upload-first prerequisite is gone, live first on Ultra and Workspace tiers.
TL;DR — NotebookLM's June 8, 2026 update removes the upload prerequisite. Open a blank notebook, enter a question or loose idea, and it uses Google Search to autonomously find, suggest, and cite relevant web sources — including related works and primary sources in other languages — adding them as active sources you stay in control of. Live first for Google AI Ultra and Workspace (AI Ultra / AI Expanded) on the web, rolling out to other tiers over time. *Google's internal benchmark: 78.2% win rate on web research and source discovery vs the prior version.
Verified June 8, 2026. Maintained by a small team of power users focused on multi-AI research and learning workflows — no affiliate relationship with Google. See the full June update →
What changed on June 8, 2026
The prerequisite that defined NotebookLM for two years is gone
NotebookLM began life as a retrieval tool: it worked best when you arrived with your documents already assembled, and it answered strictly from those sources. The June 8 update breaks that barrier. Powered by Gemini 3.5 and the Antigravity agentic framework, NotebookLM can now build the source repository for you from a loose question typed straight into the chat box.
Someone exploring an unfamiliar topic — who doesn't yet know which sources to trust — can now start inside NotebookLM instead of doing preliminary research somewhere else first. It's the natural companion to code execution: one feature finds the material, the other analyzes it.
The start-from-scratch workflow
Four steps from a blank notebook to a cited source set
Open a blank notebook
No uploads, no PDFs, no links. Start with an empty project and the chat box.
Type a question or loose idea
Describe what you're trying to understand in plain language — a topic, a claim to investigate, an author to explore.
NotebookLM searches and gathers
It deploys Google Search autonomously, compiles relevant pages into a repository, and brings them in as active, cited sources — surfacing related works and primary sources in other languages where useful.
You curate, then work
You decide which discovered sources to keep; attribution stays visible. From there it's a normal notebook — ask, summarize, or run code on what it found.
How the autonomous sourcing works
Search integration, not a static index
When you enter an idea, NotebookLM uses Google Search to query across the open web, evaluate relevance, and assemble a secure repository of documents it judges useful — then imports them as sources you can cite and query. Google frames the use cases plainly: find primary sources in other languages to understand new perspectives, or track down related works by an author you just discovered.
Crucially, you stay in control: NotebookLM proposes and adds sources, but you choose what stays in the notebook so your work remains grounded in material you trust, and every source keeps clear attribution. This is broader than Deep Research (the November 2025 mode that autonomously combs hundreds of sites to produce a report) — here the point is to build your source library from chat, not just return a one-off report.
Upload-first vs start-from-scratch
When each entry point wins
| Situation | Best entry point | Why |
|---|---|---|
| You already have the documents | Upload first | Grounded in exactly the material you trust from the start. |
| New topic, no sources yet | Start from scratch | Skip the manual hunt — let it assemble a starting set. |
| Exploring an unfamiliar field | Start from scratch | Surfaces related works and primary sources you didn't know existed. |
| Need non-English primary sources | Start from scratch | Can surface primary sources in other languages. |
| Regulated / confidential corpus | Upload first | Keep the source set closed and controlled; don't pull in web pages. |
Steering it — and watching source quality
The upside comes with an obvious risk
Autonomous web access is powerful, but broader reach means more chances for weak sources, junk pages, or plain-wrong information to slip into the mix — and a polished report built on shaky inputs is still shaky. Treat start-from-scratch like a very fast junior researcher: excellent at gathering, still in need of a human editor.
Practical guardrails: be specific in your opening prompt (name the angle, the time frame, the kind of source you want); review and prune the discovered set before you build on it; verify citations against the original pages; and remember NotebookLM's own limits — paywalled URLs pull in the gate page, not the content, so a source that looks added may be hollow. Full caps and gotchas are in the limits & benchmarks sheet →
Which plans have it
Same rollout as the rest of the June agentic upgrade
Start-from-scratch sourcing shipped with the June 8 agentic bundle, live first — globally, on the web — for Google AI Ultra subscribers and Workspace business customers with AI Ultra Access or AI Expanded Access. Free-tier users weren't included at launch; Google said access will widen over time but gave no date, and the mobile app wasn't named as a launch surface. The numeric caps on sources and chats are unchanged — see the spec sheet for those.
Free prompt: research a topic from zero
Paste into a blank Ultra/Workspace notebook to kick off autonomous sourcing with guardrails
The full Research-From-Scratch prompt pack
12 prompts for building trustworthy source sets fast: structured discovery, credibility triage, multilingual primary-source hunting, gap analysis, and hand-off prompts that turn a curated set into briefs, reports, and study guides.
The blank notebook went from a dead end to the starting line
- No cold-start problem. You don't need to know the literature before you begin — NotebookLM assembles a first draft of the source set for you.
- Grounded, not open-ended. Unlike a raw chatbot, it still cites and shows sources, and you approve what stays — research you can trust, not vibes.
- It pairs with code execution. Find the material here, then run the numbers on it — the two June features are one workflow.
Powerful gathering, human editing — curate before you build ↓
Frequently asked questions
Can NotebookLM work without uploading documents?
Yes. Since June 8, 2026 you can open a blank notebook, type a question or idea, and NotebookLM uses Google Search to find and add relevant sources for you. The old requirement to assemble your own sources first is gone.
Does NotebookLM search the web now?
Yes. It deploys Google Search autonomously to discover relevant, high-quality pages, suggest materials, surface related works, and even find primary sources in other languages, then imports them as cited sources in your notebook.
Are the auto-found sources cited?
Yes. Every source keeps clear attribution, and you stay in control of which discovered sources are added, so your work stays grounded in material you trust rather than unsourced generation.
How is this different from Deep Research?
Deep Research (November 2025) autonomously combs hundreds of sites to produce a one-off report. Start-from-scratch is broader: its purpose is to build your notebook's source library from a chat prompt, which you then keep working with — ask questions, run code, or generate outputs.
How do I control what it pulls in?
Be specific in your opening prompt, then review and prune the proposed set before building on it. You approve which sources stay, and you should verify citations against the originals — broader web access means weak or paywalled pages can appear.
Which plans and devices support it?
At launch it's web-only and limited to Google AI Ultra subscribers and Workspace business customers with AI Ultra or AI Expanded Access. Free-tier and mobile weren't included at launch; Google said access will expand over time without a published schedule.