NotebookLM Review: Gemini Notebook Features, Plans, and Privacy

This NotebookLM review examines Google's research assistant, now called Gemini Notebook, for people who need to understand documents and turn evidence into useful work. Its appeal is a persistent collection of sources, answers with citations, and tools for producing summaries, audio, and other outputs.

Google announced the rename on July 16, 2026, while keeping the product a standalone research workspace. The name has changed, and so have several capabilities and limits. This review uses current official documentation checked on October 14, 2026; it is a documentation-based assessment rather than a claimed hands-on test. Google's rename announcement

NotebookLM Review: Quick Verdict

Gemini Notebook is worth testing when your work starts with a defined body of evidence: technical documentation, course readings, project briefs, or research reports. The strongest reason to use it is the ability to return to the same source collection and inspect the passages behind an answer.

Start with the free version and one representative project. Upgrade when source capacity or compute limits interrupt useful work, rather than buying a larger plan before establishing a routine. Its weakest fit is a task where you expect an answer to be correct simply because a citation appears beside it.

The decisive question is whether it reduces the effort required to produce a checked result. An attractive summary that needs extensive correction can still be poor value.

What Is NotebookLM Now?

Google's July announcement describes Gemini Notebook as the continuation of NotebookLM, with connections across the Google ecosystem. It also announced a cloud computer for code execution, initially available to Ultra users and qualifying Workspace customers, with Pro rollout planned afterward. That historical rollout announcement is not proof that every account has identical access today. Check the features visible in your account. Rename and rollout details

The June upgrade announcement named Gemini 3.5 and Antigravity as the underlying system. Google reported improved performance against its previous system, but these are vendor evaluations, not evidence that NotebookLM beats every competing research assistant. Google's research upgrade

For readers, the product name matters less than the workflow: choose evidence, ask a bounded question, inspect support, and revise the output.

NotebookLM Features That Matter

Source Collections and Citations

Current source documentation lists PDFs, Word files, Markdown, CSV, PowerPoint, Google Docs, Slides, Sheets, images, audio, web links, and other supported formats. Imports have boundaries: a web page contributes text rather than its embedded media, and a YouTube source uses the transcript of a supported public video. Supported sources

For ordinary source chat, users can select the materials to include. Citations open supporting passages so you can check the answer in context. That makes the tool useful for questions such as which specification defines a requirement or where two reports disagree. Chat and citations

A useful prompt is: “Compare the release requirements in these two specifications. Cite each requirement and flag any conflict or missing evidence.” This asks for an inspectable result instead of a broad summary that hides disagreements.

Research and Downloadable Outputs

Gemini Notebook also documents an agentic chat experience that can search, run code, create files, and revise generated artifacts. Google explicitly calls these functions experimental. Distinguish this broader experience from ordinary chat over selected sources. Agentic chat documentation

The upgrade announcement describes outputs including PDF reports, Word documents, charts, structured data, Excel spreadsheets, and PowerPoint decks. Those formats can shorten the route from evidence to a deliverable, but generating a file does not establish that its calculations or conclusions are sound. Output formats

For a spreadsheet, ask for units, assumptions, and the calculation method. For a presentation, check that visual simplification has not removed a qualification that changes the meaning.

Audio for Orientation

Audio Overviews turn sources into AI-hosted discussions. Google documents formats including a deep dive, brief, critique, and debate, along with customization instructions. It also warns that generated audio can contain inaccuracies. Audio Overview guide

Use audio to get familiar with a subject before reading the evidence. A fluent conversation is easy to remember, so verify names, numbers, and disputed claims before repeating them in a meeting or article.

NotebookLM Pricing and Limits in 2026

NotebookLM pricing is now best understood through Gemini Notebook access within Google's plans. Standard access is free; higher capabilities and limits are available through Google AI Plus, Pro, Ultra, and qualifying Workspace or Cloud arrangements. Check the account-specific purchase page for the price, currency, and renewal terms. Current plan options

The published notebook and source capacities are:

• Standard: 100 notebooks per user; 50 sources per notebook.
• Plus: 200 notebooks per user; 100 sources per notebook.
• Pro: 500 notebooks per user; 300 sources per notebook.
• Ultra, 20 TB plan: 500 notebooks per user; 500 sources per notebook.
• Ultra, 30 TB plan: 500 notebooks per user; 600 sources per notebook.

These capacities are subject to change. They describe storage organization, not an unlimited allowance for generation. Plan capacity table

The current usage guide describes compute-based limits affected by prompt complexity, models, features, and conversation length. It says quotas refresh every five hours until the weekly limit is reached. Plus provides twice standard usage, Pro four times standard, and Ultra five or twenty times Pro depending on the subscription. Settings shows usage; Studio shows expected generation cost. Current usage limits

This makes older tables of fixed daily chat or audio counts unreliable for planning. Track the actual work your account completes. Larger source capacity and more computation solve different bottlenecks; neither substitutes for well-organized evidence.

NotebookLM Privacy and Source Freshness

Google's privacy documentation says Notebook content is not used to directly train foundational AI models unless you provide feedback. Consumer feedback can include prompts, sources, uploads, outputs, and related context; reviewed feedback may be retained for up to three years. Qualifying Workspace and Education uploads, queries, and responses are excluded from human review and AI training, including when feedback is supplied. Content shared with other Google services follows their applicable notices. Privacy and terms

For a team pilot, identify the account type before adding internal material. Decide which documents belong in the notebook, who should have access, and how generated deliverables will be reviewed and stored. Feedback and cross-service use deserve their own review rather than an assumption that all Google AI surfaces share identical terms.

Freshness is another practical issue. Google's current source guide says Drive imports auto-update and offers manual sync. It also says comments and footnotes from Google files are not imported. A synced document can therefore still omit context that matters to your conclusion. Import behavior

Use a short source inventory with titles, dates, and intended scope. Before finalizing a report, confirm that the passages you checked belong to the versions you meant to use.

How to Evaluate NotebookLM Against Alternatives

Run a small comparison using a project you already understand. Pick several documents containing an explicit requirement, a disagreement, a numerical claim, and a known gap. Ask each candidate tool the same questions, then review the evidence yourself.

Measure three things: whether the answer is supported, how quickly you can inspect that support, and how much correction the deliverable needs. Include subscription cost and review time in the decision. A larger model or more polished output is useful only if it improves that workflow.

For research spanning an open-ended web question, SD's ChatGPT Deep Research vs Perplexity comparison provides adjacent context. For an ongoing team knowledge workspace, compare the priorities discussed in the Notion AI review. These are different workflow choices, so evaluate the specific task rather than assuming one universal winner.

If your source material starts in meetings, the Granola AI review explores the earlier capture and note-taking stage. Notebook research and meeting capture can complement each other, but inaccurate source notes will remain inaccurate evidence.

Conclusion

This NotebookLM review finds Gemini Notebook most compelling as a repeatable research workflow built around inspectable evidence. Its source collections, citations, audio, and expanding output tools can make document-heavy work easier to navigate.

Begin with a small project, verify the answers, and measure correction time. Upgrade only when a real limit blocks useful work. The best result is a deliverable you can explain and defend, with the source evidence still close at hand.