NotebookLM in 2026: Gemini 3, Video Overviews, and Gemini App Integration

Hello, this is Hamamoto from TIMEWELL.

About six months ago, I wrote about the news that NotebookLM's audio summaries had gained Japanese support. At the time, the simple fact that an AI could discuss your own documents like two radio hosts — in Japanese — was impressive enough on its own. But between then and today (June 2026), NotebookLM has changed not just one step but two. The model underneath was swapped for Gemini 3, video joined the audio-only summaries, and you can now call NotebookLM notebooks directly from inside the Gemini app.

Honestly, few tools reinvent themselves this fast. In this article, I'll lay out where NotebookLM stands in the first half of 2026, with an eye toward actually fitting it into your work. I've written it so the full picture is clear whether you're someone who "keeps hearing the name but isn't sure what it does" or someone who "tried it once and hasn't been back."

What Changed in Six Months: Then and Now

Let me capture the scope of the change in a single table. Putting the state of things when I wrote the previous article next to where we are now makes the gap obvious.

Item	Then (around January 2026)	Now (June 2026)
Base model	Gemini 2.5 Flash	Gemini 3[^1]
Summary format	Audio (Audio Overview) centered	Audio plus video (Video Overview)[^2]
Video Overview languages	English only	Expanded to 80 languages[^3]
Relationship to Gemini	Separate app	Notebooks usable as a source inside the Gemini app[^4]
For enterprises	Limited	NotebookLM Enterprise generally available (GA)[^5]

Of these, the quietest and most impactful is the model change. NotebookLM switched its base from Gemini 2.5 Flash to Gemini 3 in late 2025[^1]. From the user's side the screen barely changed, but the comprehension when you feed it long documents — and the accuracy of answers to questions that span multiple files — is clearly higher. When I loaded about ten investment-review PDFs and asked cross-cutting questions, queries that previously got a curt "this isn't mentioned in the document" now came back with answers that cited the relevant passages in other files. Model generation changes rarely make for splashy headlines, but they translate directly into this kind of everyday feel.

The "then vs. now" point worth holding onto is that NotebookLM is no longer a standalone convenience tool. It's becoming part of the Gemini mothership — one of its entry points. Miss that, and the features I'll describe next only make half their sense.

The Basics: It's Trustworthy Because the Sources Are Constrained

Before the new features, let me confirm why this tool has a different character from other AI. The plain chat in ChatGPT or Gemini builds answers from the whole internet and its trained knowledge. That's convenient, but the provenance gets fuzzy and plausible-sounding falsehoods (hallucinations) slip in.

NotebookLM takes the opposite approach. It constrains the material for its answers to only the sources you upload. You bring in PDFs, text, website links, YouTube videos, and Google Docs, and the AI answers only within that range. On top of that, the generated text carries citations indicating "this sentence came from page X of this source." Because you can verify provenance instantly, it pairs well with documents where accuracy is everything — internal materials, contracts, and the like. The fact that uploaded data is not used for model training is also a non-negotiable condition when handling internal documents.

The range of supported sources widened in 2026 as well. You can now upload EPUB e-books directly[^6], which makes book-based research much easier. The basic workflow hasn't changed. Drop your materials into "Sources" on the left, uncheck what you don't need, and just ask your questions. That's it. Once you're used to it, loading the first book and getting an answer takes less than a minute.

The Upgraded Studio: Audio, Video, Mind Maps, and Reports in One Place

The biggest visual change in NotebookLM is the "Studio" panel on the right. Previously you could create only one of each summary type per notebook, but now you can save multiple outputs of the same type[^2]. The top of Studio lines up four tiles — Audio Overview, Video Overview, Mind Map, and Report — and from there you convert your materials into different formats[^2].

The headline is Video Overview, the video summary. It automatically assembles a narrated slide video from your materials. The English version launched in July 2025[^2] and later expanded to 80 languages[^3]. Because it pulls diagrams, graphs, quotes, and figures from your sources into the video, it lands more easily than reading text. It works well as warm-up material before an internal study session, or for announcing a new service across the company. There's also a more ambitious "cinematic" video overview, where Gemini chooses from hundreds of structural and stylistic patterns to present your content like a story[^3].

Audio Overview was also overhauled, now speaking in a deeper, more comprehensive way[^3]. It supports Japanese, and the format where two speakers explain your material in a back-and-forth is still there. Skimming the key points at 2x speed during a commute, then replaying just the parts that caught my attention at normal speed — that's a habit I still keep.

A subtle but effective change: you can now call up these formats straight from the chat screen. After going a few rounds of questions about your material, you can convert that conversation into an audio summary, a video, or a report with one click[^7]. Conversation history is saved automatically, so you can close a session and pick up where you left off[^7]. The "I just dug into something — let me turn that into a video to share" moment comes up more often than you'd think.

The genuinely hard part of adopting an AI tool isn't memorizing features — it's figuring out where in your workflow to insert it so it actually delivers. TIMEWELL's WARP helps embed the latest tools like NotebookLM and Gemini into real operational workflows. If you want to avoid the "we deployed it but nobody uses it" trap, let's talk.

Report output and infographics were also strengthened. The infographics that visually summarize your sources can now be chosen manually from ten predefined styles[^8]. Generated slide decks can be exported as PPTX in addition to PDF[^7], so you can fine-tune in PowerPoint and feed it straight into a proposal.

Integration with the Gemini App: NotebookLM Becomes an "Entry Point"

The structurally largest change in 2026 is the integration of the Gemini app and NotebookLM. Since late January 2026, you can designate a NotebookLM notebook as a "source" inside the Gemini app[^4].

Why is that valuable? Until now, NotebookLM's strength — "grounding answers strictly in your own materials" — was self-contained within NotebookLM. Now you can combine it with Gemini's powerful toolset. Concretely, you can run Gemini's Canvas (collaborative document and code editing), Deep Research, and Guided Learning (interactive learning support) while keeping your answers grounded in the materials in your notebook[^4]. You can also build a notebook into a "Gem," which saves your frequently used setups[^4].

For example, consolidate your internal regulations and manuals in NotebookLM, tie that to a Gemini Gem, and when you casually ask Gemini "under our work rules, how is this case handled," it answers based on your own regulations rather than internet content. The way to put it: NotebookLM has started functioning as the "trustworthy source of truth" for Gemini, the AI you use day to day. Note that using this integration requires your administrator to have NotebookLM enabled, controlled in your organization's Admin Console[^4].

I see this as quite important personally. Many companies are debating "which AI should become our internal standard," and a setup that centers on Gemini while using NotebookLM as the grounding layer looks like a realistic landing spot for businesses already on Google Workspace.

Considerations for Enterprise Adoption: Enterprise GA and Security

So far I've focused on individual use, but if you're serious about using this as a company, you need to look at the enterprise options. In 2026, NotebookLM Enterprise became generally available (GA) for the Gemini Enterprise Business edition[^5]. Studio features like slide deck and infographic generation are available in the Enterprise edition too[^5].

What matters for enterprise use is connecting to external data. Gemini Enterprise provides a mechanism to connect your own MCP (Model Context Protocol) server, securely giving the AI access to your private internal data, custom internal tools, and MCP-compliant third-party systems[^5]. Previews for connecting external service data stores like Notion and Crossbeam have also started[^5], greatly widening the range of internal information the AI can reference. On the model side, Gemini 3.1 Pro — aimed at more complex tasks — is available in preview via Vertex AI and Gemini Enterprise[^9].

That said, having all the features doesn't mean a company-wide rollout will succeed right away. From what I've seen in the field, the stumbling block is almost always operations, not technology. Who is allowed to upload which documents? How do you draw the line on handling confidential information? Can you send a generated summary or video outside the company as-is? Roll it out without settling these rules, and a convenient tool ends up floating untethered.

This is exactly the area TIMEWELL focuses on. As a foundation for enterprise AI, we offer ZEROCK, a high-security AI agent that runs on domestic Japanese servers, chosen by companies that prioritize domestic data sovereignty and knowledge management. How to divide responsibilities between general-purpose tools like NotebookLM and Gemini and a dedicated in-house foundation like ZEROCK — that design is, I believe, where the difference will show in 2026 AI adoption.

Summary: A Good Time to Revisit It

If you tried NotebookLM briefly six months ago and thought "interesting, but still a toy," you're exactly the person I'd want to open it again. With the base now on Gemini 3, video summaries added, and the tool embedded as part of the Gemini app, the range of what you can do has gone up a level.

Finally, if you want to test the current NotebookLM at work, I'd recommend this order:

• First, load about five of your own documents and check the accuracy of cross-cutting questions (this is where Gemini 3's strength shows)
• Next, make just one Video Overview and see whether it holds up for internal sharing
• If you use Google Workspace, try the integration that lets you designate a notebook as a source from the Gemini app
• If you're considering a company-wide rollout, settle your upload and disclosure rules before the features

Tools turn into different products in six months. But the judgment of where to insert them into your work isn't something a tool does for you. Working that out together is what I consider our job. If you're unsure how to fit the latest AI into your operations, feel free to reach out through a WARP consultation.

→ WARP consultation

WARP service details

Related Articles

• The Future of Software Development Powered by AI — The New Development Revolution Shaped by Anthropic and Cursor
• Google AI Studio: What It Can Do for Your Business
• Google Cloud Next Recap: TPU, Gemini, and Agent Advances Ushering in a New AI Era

Footnotes

[^1]: Android Central, "NotebookLM is now powered by Gemini 3 for better reasoning and multimodal understanding" (Dec 22, 2025) https://www.androidcentral.com/apps-software/ai/notebooklm-is-now-powered-by-gemini-3 [^2]: Google, "What's new in NotebookLM: Video Overviews and an upgraded Studio" (Jul 29, 2025) https://blog.google/innovation-and-ai/models-and-research/google-labs/notebooklm-video-overviews-studio-upgrades/ [^3]: Google, "Google's NotebookLM updates Audio and Video Overviews" https://blog.google/technology/google-labs/notebook-lm-audio-video-overviews-more-languages-longer-content/ [^4]: Google Workspace Updates, "Take your notebooks further by adding NotebookLM as a source in the Gemini app" (Jan 27, 2026) https://workspaceupdates.googleblog.com/2026/01/take-notebooks-further-notebooklm-gemini.html [^5]: Google Cloud, "Gemini Enterprise release notes" https://docs.cloud.google.com/gemini/enterprise/docs/release-notes [^6]: Google Workspace Updates, "New ways to customize and interact with your content in NotebookLM" (Mar 2026) https://workspaceupdates.googleblog.com/2026/03/new-ways-to-customize-and-interact-with-your-content-in-NotebookLM.html [^7]: Google, "Dive deeper into Google I/O 2026 with NotebookLM" https://blog.google/innovation-and-ai/products/notebooklm/notebooklm-google-io-2026/ [^8]: NotebookLM Help, "Generate Audio Overview in NotebookLM" https://support.google.com/notebooklm/answer/16212820?hl=en [^9]: Google, "Gemini 3.1 Pro: A smarter model for your most complex tasks" https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/ [^10]: Jeff Su, "NotebookLM in 2026: What Changed and What Matters" https://www.jeffsu.org/notebooklm-changed-completely-heres-what-matters-in-2026/