Gemini Notebook (Formerly NotebookLM): 2026 Guide

What Is Gemini Notebook? The Rename, Explained

Gemini Notebook is NotebookLM. Google renamed the product on July 16, 2026, and nothing else about your account changed: same notebooks, same sources, same saved chats, and notebooklm.google.com still resolves. It's Google's free, source-grounded research and study tool, where every answer is generated from documents you supply rather than from the open web.

The rename wasn't cosmetic housekeeping. It arrived with native code execution, and in the two months since, Google has folded notebooks into Google Search, replaced the entire usage-limit system, and added voice. The product still runs on Gemini 3.5, still has a genuinely generous free tier, and most people still type "NotebookLM" into the search bar, so this guide uses both names.

When NotebookLM launched its Audio Overviews in late 2024, the pitch was simple: upload your documents, and an AI grounded in those documents answers questions and makes a podcast. That tool no longer exists in any meaningful sense. The product today researches the web on its own, generates animated videos, edits slide decks on command, writes and runs code inside a private cloud container, holds a spoken conversation about your sources, and suggests new sources mid-conversation. Google says 30 million people and more than 600,000 organizations now use it.

That pace creates a real problem for learners. Half the NotebookLM advice published in 2025 is now wrong, and the update stream is so fast that it's hard to tell which changes matter for an actual study workflow and which are demo material. A new infographic style is nice. An agent that builds your bibliography for you changes how you start every project.

This article is the sorting exercise. Every date and number below was independently verified against Google's announcements and primary reporting, and re-checked on September 17, 2026. If you're deciding whether NotebookLM should be your main research tool, or whether to pay for it, this is the current state of play. If you've already decided to leave and want options, that's a different article: see our NotebookLM alternatives comparison.

---

The Verified Update Timeline

Here's what shipped between late 2025 and September 2026, with the dates confirmed. Skim the third column if you only care about impact.

Five threads run through this list: ingestion got agentic, outputs multiplied, the chat itself got a much stronger brain, in July the whole thing was folded into Google's Gemini brand, and in September the economics of using it were rewritten. The next sections take each thread in turn.

---

Research Ingestion: Deep Research and Sources That Find Themselves

The single most important change in this entire timeline is the November 13, 2025 Deep Research launch. Before it, NotebookLM had a cold-start problem: the tool was only useful after you'd done the work of collecting sources. Deep Research inverts that. You type a question, it drafts a research plan, browses the web for several minutes, and returns a cited report plus a list of sources you can import into the notebook with one click. A lighter "Fast Research" mode does a quicker scan when you don't need depth.

The June 8, 2026 update pushed this further. You can now start a plain chat about a project, and NotebookLM will suggest sources as the conversation develops, using its research skills and Google Search to build the knowledge base around you. Google's framing is that you no longer need to bring your own sources at all. As of September 17, 2026 that capability still sits on Ultra and Workspace business tiers, with a promised wider rollout.

The same period quietly fixed the format gaps. NotebookLM now accepts Google Sheets, Word documents, PDFs straight from Drive, images, and as of March 2026, EPUB files. At I/O 2026, Google added automatic Drive sync, so a notebook built on living documents updates as those documents change. For a student, that means the course notebook tracks the shared lecture notes folder without manual re-uploads.

What does this mean in practice? The honest answer is that Deep Research is excellent for breadth and mediocre for judgment. It will surface fifteen plausible sources on spaced repetition in five minutes, which beats an hour of manual searching. It will not tell you which three of those are worth reading closely, and it inherits all the usual weaknesses of automated source-finding: SEO bait ranks, paywalled gems don't. Treat its output as a draft bibliography, not a finished one. For how Deep Research-style agents compare across ChatGPT, Gemini, and Perplexity, see our deep research tools comparison.

---

Output Formats: Audio, Video, and the Studio Explosion

In mid-2025, NotebookLM's Studio panel produced an audio overview and a few text artifacts. Today the menu reads: Audio Overview, Video Overview, Short Video Overview, Cinematic Video Overview, Mind Map, Reports, Interactive Learning Overview, Flashcards, Quiz, Infographic, Slide Deck, and Data Table.

Audio Overviews matured rather than changed. The headline move was August 25, 2025, when non-English audio finally got the full-length, genuinely conversational treatment instead of the shorter summaries non-English users were getting before. If you tried a Japanese or Hindi Audio Overview in early 2025 and found it thin, give it a second look.

Video Overviews now come in three lengths. The standard version, which is a narrated slideshow, is available on every tier including free. Short Video Overviews, added at the end of June 2026, are roughly 60-second clips that pair a narrative summary with educational animation, built for phone-sized review. They launched on Ultra and Pro and, as of the September 15, 2026 update, run in more than 80 languages and can be shared with classmates. The Cinematic Video Overviews, launched March 4, 2026, are a different animal: Gemini 3 acts as a creative director while Nano Banana Pro and Veo 3 generate animated, film-style visuals from your sources. They're impressive, they're English-first, and they went out Ultra-only before reaching Pro later the same month. For most learners, the cinematic version is a curiosity, not a study tool. A narrated slideshow conveys the same information.

The study artifacts are the underrated story. Flashcards and quizzes gained progress tracking in March 2026, so cards you miss come back and cards you've got stay gone. September 2026 addressed the quizzes' biggest weakness. Google says short answer, multiple select, and fill in the blank are coming to all users in the next few weeks, alongside the existing multiple choice. That matters more than it sounds. Multiple choice tests recognition. Short answer and fill in the blank force retrieval, and retrieval is the mechanism that builds durable memory. The same announcement covered Interactive Learning Overviews, which live under Reports on the web and stitch a summary together with the infographics, quizzes, and flashcards generated from the same sources instead of leaving them as separate artifacts you have to assemble yourself.

Data Tables, added December 2025, turn a pile of qualitative sources into a structured comparison table you can export to Sheets, which is genuinely useful for literature reviews and product comparisons alike. Slide decks became editable by prompt in February 2026 and export to PowerPoint, which moved them from toy to usable for actual class presentations.

A ranking for learners, most to least valuable: quizzes and flashcards with tracking (now that they test recall), Interactive Learning Overviews, Data Tables, Audio Overviews for commute review, standard Video Overviews, infographics, and Cinematic videos last. The pattern is boring but consistent: the formats that make you retrieve information beat the formats you passively watch. That's the same finding behind active recall, and it holds no matter who generates the cards.

---

Smarter Chat: Gemini 3.5, Saved History, and Personas

The model under the hood changed twice in seven months: Gemini 3 in December 2025, and Gemini 3.5 as the default on June 8, 2026. Benchmarks aside, the practical effect is that cross-source reasoning got noticeably better. Questions like "where do these five papers disagree about working memory" now produce answers that cite the actual points of tension rather than summarizing each paper in sequence.

Two quieter changes matter as much for a study workflow:

Saved conversation history, rolled out on October 29, 2025, ended one of NotebookLM's most annoying behaviors: chats that vanished when you closed the tab. Conversations now persist, resume, and can be deleted. Google enabled the full 1M-token context window across all plans in the same update, so a semester-long dialogue with your course materials is now actually possible on the free tier.

Custom personas and goals let you tell each notebook how to behave, with up to 5,000 characters of instruction since November 2025. A persona like "act as a tutor preparing me for a closed-book exam, always quiz me before explaining" turns the same notebook from a summary machine into something closer to a teaching assistant. Most users never touch this setting. It's the highest-leverage five minutes of setup the product offers.

You can now talk to it. On September 15, 2026, Google added real-time voice conversation with a notebook from the Android and iOS apps, in nearly 100 languages. You ask out loud, it answers only from your sources, and you can interrupt mid-answer to redirect it, which is closer to how you'd actually question a tutor. It went to Google AI Ultra subscribers aged 18 and up first, with Pro and other tiers promised soon after. For anyone who revises while walking or commuting, this is the first version of the product that works without a screen.

The June 2026 update also added what Google calls Antigravity-powered skills: you can give detailed instructions for output formats and the agent executes multi-step work, including writing code and producing downloadable reports. That's aimed more at professional research than studying, but it signals where the product is going: less notebook, more agent.

---

The Rename, Code Execution, and the API Question

On July 16, 2026, Google renamed NotebookLM to Gemini Notebook. The logo picked up the blue and purple Gemini gradient, and Google's framing was that the product stays "a standalone product focused on being your premier research tool" while doing "more across the Google ecosystem." Nothing you saved disappeared, and none of your notebooks broke. If you've bookmarked notebooklm.google.com, it still works. For most learners the rename is cosmetic, but it signals Google folding the tool into the Gemini family rather than letting it drift as a side project.

The substantive change shipped in the same update: every notebook now gets a secure cloud computer that can write and execute code natively, grounded in your sources. Ask it to chart the trend across a Data Table, run a quick statistical test on a dataset you uploaded, or clean up a messy CSV, and it does the work in a private container instead of hand-waving an answer. This is the Gemini 3.5 plus Antigravity upgrade reaching general availability. It went to Google AI Ultra and Workspace AI Ultra and Expanded Access users first, with a rollout to all Pro users on the web over the following weeks. For anyone doing quantitative coursework or light data analysis, this is the most useful thing to arrive since Deep Research.

Two ecosystem moves came with the rename, and the second one has now landed. Notebooks sync across the Gemini app and the standalone Gemini Notebook site, so a notebook you build on the web shows up in the app for quick chat on your phone. And on August 19, 2026, notebooks went live inside Google Search's AI Mode, in English, across every country Gemini Notebook supports except the European Economic Area. You can create, edit, and chat with a notebook straight from the search bar, and sources, renames, custom instructions, and deletions sync both ways.

Chats don't sync, and that catches people out: an AI Mode conversation stays in Search, and a Gemini Notebook conversation stays in Gemini Notebook. And notebooks shared with you by someone else don't appear in AI Mode at all, though you keep normal access to them in Gemini Notebook.

Does NotebookLM have an API? This is one of the most common questions, and the answer is split. There is no public, self-serve consumer API: you cannot grab an API key from a settings page and script your personal notebooks. What does exist is an enterprise API, now branded Gemini Notebook Enterprise, sold standalone or as part of Gemini Enterprise (the platform Google folded Agentspace into). Its docs currently carry a Preview, Pre-GA banner despite an earlier release note calling notebook management generally available. It exposes REST endpoints to create, retrieve, share, batch-delete, and list recently viewed notebooks, manage and upload sources, and generate an audio overview, with multi-regional US, EU, and global endpoints and enterprise controls like VPC Service Controls and customer-managed encryption keys. The gap developers keep hitting is stark: the product's own overview sells it as somewhere you "query, chat, or generate summaries," and query and chat are the two things the API cannot do. For a student or solo researcher, the practical takeaway is simple: automation is an enterprise feature, and unofficial community clients that drive the web app exist but can break whenever Google ships a change, which lately is often.

---

What Still Frustrates

A verified changelog should also verify what didn't change. Four structural problems survived the entire update wave.

Notebooks are still silos. Your knowledge lives in per-notebook buckets, and chat works within one notebook at a time. There's no "search everything I've ever saved" view and no graph of connections across notebooks. If you study four subjects, you maintain four walled gardens, and insights from one don't surface in another.

There's still no capture layer for reading. Google is closing part of this gap: the mobile app can already receive a website, PDF, or YouTube video shared to it from your phone, and an audio recorder for lectures was announced on September 15, 2026 to start rolling out the following week, English output only. Both are real improvements for anyone who learns by listening. The reading half is untouched. There's no highlighting on live web pages, no annotation while you watch a video, no way to mark the paragraph that mattered rather than filing the whole document. Everything still enters as a whole source, so the gap between "this one paragraph changed my mind" and "it's in my notebook" stays manual.

The limits got harder to predict, not just tighter. Fifty sources per notebook on the free tier sounds generous until a real literature review hits it. The bigger change is September 2's move to a compute budget: instead of knowing you have 50 chats today, you now have an allowance that drains faster when you ask for something expensive. Gemini Notebook warns you as you approach the wall, but you can no longer plan a study session around a number you know in advance.

The best features land on expensive tiers first. Cinematic videos launched Ultra-only and reached Pro weeks later. Free and Plus still don't get them, and never got native code execution either. The source-repository-from-chat feature launched on Ultra and Workspace business plans. The free tier today is more capable than the free tier of 2025, but the distance between free and Ultra widened, and Google's rollout pattern consistently rewards the $99.99-and-up crowd first.

None of these are dealbreakers. They just define what a notebook is good for: deep work on a bounded set of sources you assembled on purpose. So where does that set come from?

---

Pairing NotebookLM with a Capture Layer

Every NotebookLM workflow has the same dependency: the quality of the notebook is the quality of its sources. Deep Research helps you cold-start a topic you know nothing about. But the most valuable sources for your learning are usually the ones you already found yourself, scattered across months of reading, watching, and browsing. Gemini Notebook can receive a whole page you share to it, but it has no way to capture the one paragraph that stopped you, or to build that habit across months of reading. That's not a missing feature so much as a different job, and it's the job a highlighter does.

The pairing works like this. As you read during the day, Glasp's web highlighter captures the passages that made you stop, on any web page, with your notes attached. For video, YouTube Summary gives you the transcript alongside the player so you can highlight the moments that matter with timestamps instead of vaguely remembering "somewhere in that two-hour lecture." Over weeks, this builds a record of what actually caught your attention, which is exactly the raw material a notebook needs.

When a project takes shape, you export those highlights as markdown or text and drop them into a notebook as a source. Now NotebookLM's chat, quizzes, and audio overviews run over the things you personally found worth keeping, not just whatever Deep Research scraped. The two tools are honestly complementary: Glasp is the wide-open capture net across your daily reading, NotebookLM is the deep synthesis engine for a bounded project. One is free and social, the other is Google-scale and private. Using both costs nothing.

We've written up the full pipeline, including how this fits with Deep Research-style agents, in our AI research workflow guide, and the video-specific version in turning YouTube videos into study notes.

---

Free vs Paid: Limits, Pricing, and Which Tier You Need

On September 2, 2026, Google threw out the way Gemini Notebook counts. The old system gave you fixed daily allowances: 50 chats a day, 3 audio overviews a day, 10 Deep Research runs a month. Those are gone. In their place is a compute budget that, in Google's wording, factors in "the complexity of your prompt, the models and features you use, the length of your chat and specific features."

Three practical consequences:

• It refreshes every five hours, not every midnight, and it runs against a weekly ceiling. Hitting a wall at 2pm no longer costs you the rest of the day.
• Cheap things are now genuinely cheap. A short factual question about one source barely moves the meter. A multi-source Cinematic Video Overview moves it a lot. Under the old caps both cost you "one generation."
• You can queue work instead of losing it. When you're out of budget, Gemini Notebook will offer to generate a Video Overview or Slide Deck later and notify you when it's ready, which Google warns "may take a couple hours." It also suggests a cheaper output format when your first choice won't fit.

You can check what you've used and your weekly limit from the profile menu, the chat warns you as you near the wall ("You're almost at your AI usage limit. Limit resets at 3:00 PM"), and Studio fills in a bar showing the expected cost before you generate. That helps. What doesn't help is that Google has never published what one unit of compute buys, so you can't plan against a number you know in advance. The web is also the only place this works today.

What survived the change are the structural caps, the multipliers between tiers, and the features that stay tier-gated. Fair warning if you go looking: Google's upgrade page still publishes the old daily-caps table under a "subject to change" header alongside the September 2 notice, so the live documentation lags its own announcement. One thing that has not changed at all: you can't buy the tool on its own, since every paid tier is bundled with a Google AI subscription.

Pricing context, because it moved twice recently: Google restructured Ultra at I/O 2026 in May, adding a $99.99 entry tier and cutting the top tier from $249.99 to $199.99. Then on June 8, 2026, Google cut the Plus plan from $7.99 to $4.99 a month and doubled its storage to 400 GB, effective at each subscriber's next renewal.

Students get a year free, and it's the best deal in the product. Announced September 15, 2026, eligible US college students get 12 months of Google AI Pro, normally $19.99 a month. Students in more than 140 other markets get Google AI Plus instead, normally $4.99 a month, with 400 GB of storage. You have until December 31, 2026 to redeem it. Read the fine print before celebrating: it requires a valid payment method up front and rolls into the paid rate automatically unless you cancel, so put a reminder in your calendar for month eleven.

Who should pay what:

Stay free if you're a casual user or working at course scale, and check the student offer first. A hundred notebooks at 50 sources each covers a typical course load with room to spare, and under the new compute budget the ordinary work of studying, asking questions of your own readings, is the cheapest thing you can do. Expensive generation is what drains the meter, so be deliberate about video.

Plus at $4.99 is the new sweet spot. Doubled source caps and twice the compute allowance for the price of a coffee. If you hit the five-hour wall more than once a month, this is the rational upgrade, and the June price cut made it an easy call.

Pro is for thesis-scale work: 100+ sources per notebook, study materials generated daily. Four times the baseline allowance is effectively unlimited for one human doing text work.

Ultra is for teams and video production, not learners. Unless Cinematic Video Overviews are part of your job or you need thousands of queries a day, the $99.99+ tiers buy you very little studying.

---

Frequently Asked Questions

Did NotebookLM change its name to Gemini Notebook?

Yes. On July 16, 2026, Google renamed NotebookLM to Gemini Notebook. It's the same product with the same notebooks, sources, and features, plus a new logo and native code execution that arrived in the same update. Google kept it as a standalone research tool rather than merging it into the main Gemini app, though the two now sync. Most people still type "NotebookLM" into search, so both names point to the same tool.

What are Gemini Notebook's usage limits?

Since September 2, 2026 there is no single number to quote. Gemini Notebook runs on a compute budget that depends on what you ask for: a short question about one source costs very little, a Cinematic Video Overview across ten sources costs a lot. The budget refreshes every five hours until you hit a weekly ceiling, and an indicator under the chat box shows what you've used. The caps that are still fixed numbers are structural: 100 notebooks and 50 sources per notebook on the free tier, rising to 500 notebooks and 500 to 600 sources on Ultra.

Is Gemini Notebook free in 2026?

Yes. NotebookLM, now called Gemini Notebook, still has a free Standard tier with 100 notebooks and 50 sources per notebook, and it has the full 1M-token context window. What changed on September 2, 2026 is how the AI limits work: instead of fixed daily counts, the free tier gets a baseline compute allowance that refreshes every five hours against a weekly ceiling. Paid tiers multiply that allowance (Plus 2x baseline, Pro 4x, Ultra 5x to 20x Pro again) rather than unlock fundamentally different core features. The exceptions are native code execution and Cinematic Video Overviews, which need Pro or Ultra, and real-time voice chat, which started on Ultra with the other tiers promised next.

Is Gemini Notebook free for students?

Better than free. On September 15, 2026, Google opened a student offer giving eligible US college students 12 months of Google AI Pro at no cost, normally $19.99 a month, and students in more than 140 other markets 12 months of Google AI Plus, normally $4.99 a month. Redemption closes December 31, 2026. It requires a valid payment method at signup and renews at the full price automatically unless you cancel first.

When did Gemini Notebook come out?

It isn't a new product with a new launch date. Google previewed it as Project Tailwind at I/O in May 2023, launched it in limited testing as NotebookLM in July 2023, opened it generally in mid-2024, and renamed it Gemini Notebook on July 16, 2026. So if you're looking for a "Gemini Notebook release date," July 16, 2026 is the date of the rename, not of a launch.

What's new in NotebookLM?

The most recent changes, from September 2026: compute-based usage limits that refresh every five hours, real-time voice conversation with a notebook in nearly 100 languages, an audio recorder in the mobile app for capturing lectures, Interactive Learning Overviews, quizzes that now include short answer, multiple select, and fill in the blank, notebooks inside AI Mode in Google Search, and a free year of Google AI Pro for eligible US college students. Before that: the July 2026 rename with per-notebook code execution, Deep Research, support for Sheets, Word, EPUB, images, and Drive-synced files, a Gemini 3 then Gemini 3.5 engine upgrade, saved chat history, Data Tables, and editable slide decks with PowerPoint export. The full dated timeline is in the table near the top of this article.

Can I use Gemini Notebook inside Google Search?

Yes, since August 19, 2026. Notebooks are live in AI Mode in Google Search, in English, across every country Gemini Notebook supports except the European Economic Area. You can create, edit, and chat with a notebook from the search bar, and sources, renames, and custom instructions sync with the standalone app. Two catches: chats don't sync between the two, and notebooks that someone else shared with you won't show up in AI Mode.

NotebookLM vs ChatGPT for studying: which is better?

They solve different problems. NotebookLM grounds every answer in sources you (or its Deep Research agent) put into the notebook, which keeps it honest and citable, and its quiz, flashcard, and audio outputs are built for studying. ChatGPT is a generalist: better at open-ended explanation, coding, and conversation, but you have to manage grounding yourself. For coursework built on a defined set of readings, NotebookLM's source-grounded design wins. For "explain this concept five different ways until one clicks," ChatGPT remains stronger. Many students sensibly use both.

What are Cinematic Video Overviews?

Launched March 4, 2026, they're animated explainer videos generated from your sources, with Gemini 3 directing the narrative and Nano Banana Pro and Veo 3 producing the visuals. They look dramatically better than the standard slideshow-style Video Overviews. They reached Google AI Pro in March 2026 and remain English-first, with Pro capped at 2 a day and Ultra at 10 to 20. For learning value per dollar, the standard Video Overviews on the free tier deliver most of the same information.

Does NotebookLM work with web pages and YouTube videos?

You can add URLs and YouTube links as sources, and Deep Research can pull web sources in for you. What NotebookLM still can't do is capture while you read or watch: there's no live highlighting on pages, no transcript annotation during a video, no save-as-you-browse flow. If your learning starts on the open web, a capture tool like Glasp fills that gap, and you can export your highlights into NotebookLM as a source when a project firms up.

Can NotebookLM read Kindle books?

Not directly. NotebookLM accepts PDFs, Google Docs, Slides, websites, YouTube transcripts, and, since March 2026, EPUB files, but it can't open Amazon's proprietary Kindle format. The workaround is to bring your reading in as text. If you highlight as you read on Kindle, Glasp's Kindle import pulls those highlights out of Amazon and lets you export them as markdown, which you then add to a notebook as a source. That way the notebook runs on the passages you actually marked, not the entire book. It's the same capture-then-synthesize pattern that makes NotebookLM stronger with any source.

---

Conclusion

Gemini Notebook in late 2026 is a genuinely different product from the one most people tried as NotebookLM in 2024 and 2025. The changes that matter for learners are the unglamorous ones: Deep Research killing the cold-start problem, saved chat history, quizzes that now demand recall instead of recognition, a 1M-token context on the free tier, a free year for students, and a compute budget that stops punishing you for asking small questions. The cinematic videos will get the headlines; the retrieval practice features will get you through the exam.

What hasn't changed is the architecture of the thing. Gemini Notebook synthesizes what you give it, notebook by notebook, and the work of noticing, capturing, and collecting good material across your daily reading still happens upstream of every notebook you'll ever build. September's audio recorder proved Google understands the problem, but it solved it for lectures, not for the forty browser tabs you closed this week. Get that layer right and every Gemini Notebook feature gets better, because it's running on sources you actually chose.

That's the workflow worth building this year: capture as you read with Glasp's web highlighter, highlight lectures and talks with YouTube Summary, then export the best of it into a notebook when a project takes shape. Gemini Notebook keeps changing every quarter, right down to its name and how it counts. A well-built source library is the part of your system that compounds no matter what Google ships next, or what it calls the notebook.